C3.ai, Inc. (AI) Earnings Call Transcript & Summary

Please take your seats. This show is about to start. A reminder to please silence your mobile devices.

Good afternoon, ladies and gentlemen. My name is Tom Siebel, and thank you so much for joining us. for our June 2023 Investor Day. The slides are going in advance. Yes, I think they will. Okay. I brought out much of our executive team to spend time with you today. I'm joined by Ed Abbo, who founded the business with me and is President and Chief Technology Officer; Houman Behzadi, who serves as Executive Vice President; my colleague, Nikhil Krishnan runs all data science is Chief Technology Officer and Product. He owns the generative AI, C3 AI -- generative AI product that he will be talking about; Juho Parkkinen, our Chief Financial Officer; Amit Berry, who runs IR; Amy is here, our Chief of Staff; Alex Amato, who runs customer services and is responsible for all the customer deployments going on around the world; Merel Witteveen, who runs our alliances program; Derron Blakeley, our General Counsel; Binu Mathew, who runs our products and engineering organization; Paul Phillips, who assist with IR; and Henry Murray, Who runs our Government Relations in D.C. So you will have the opportunity to interact with and ask questions of many of much of the leadership of C3 AI. I'm also very pleased that we'll be joined today and by my good friend, Honorio Padrón, who is our customer at Pantaleon, which is a large sugarcane producer in South America; Graham Evans, who is a -- who has his high school photo there, year book photo, and Graham is our very strategic partner in the -- as it relates to the Defense and Intelligence Agency with Booz Allen, and Graham will share with you what you're doing. And then later this afternoon, when we get to demos and cocktails, Chris Casey is joining from AWS, what we want to cover today, and it's going to be a kind of if you're going to be drinking out of a fire hose. We're going to give you -- I'll give you a very kind of brief company update, and then we'll get to substance, and Abbo will come up and give you a briefing on what is happening in sales. Honorio Padrón is going to come up and talk a bit about what it is like to be engaged in a trial and what it is likely to go into production and what the decision process is like. Merel Witteveen, who is our Group Vice President of Alliances, is going to talk about our alliance program. Graham Evans will come up on stage and talk about the nature of very significant partnership with Baker, Booz Allen and C3 AI. Then We're going to take a break. And when we take in the break, we'll have demos going on of the C3 AI generative AI product the C3 AI ESG product and the C3 AI Readiness product. And that's there's going to be 4 products there, and so the product managers will be there. You can see hands on what this stuff does and how it works. Then what might be the highlight of the day. is Nikhil Krishnan, who will come and talk about what we're doing with C3 Generative AI. I think you will find that really fascinating. Binu Mathew is going to talk about our product footprint and product road map, and then we'll bring the entire management team up to answer your questions. This is all being webcast live, which makes us all Reg FD compliant so that we'll get -- so we can speak [indiscernible] with you. I'm very pleased that we are joined also by an old friend of many of you and certainly an old friend of mine, Rick Sherlund, who is something of a legend information technology community, most recently as vice Chairman of Bank of America. But Rick really, I made his name as legendary sell-side analysts back of the days of Microsoft and Oracle when those companies were just getting started. So Rick, if you'll join me, that will be great. So Rick is going to join me, and we are going to carry on a kind of a brief conversation about what's happening at C3, and then we'll just get on with this event. So thank you again for joining us. Rick, thanks.

Thanks, Tom. Long since gone are the days where I was writing research. I read all the research, and as you'd expect, everyone is kind of focused on the transition from the subscription model to consumption, and that's pretty much what people are occupied with for the next year or 2 in their modeling. But if we look at a little bigger picture and perhaps a little longer term, you've had a long history of being very visionary with -- back to Siebel days, you invented the whole CRM category. Who wrote the book on digital transformation, and you've been talking about AI for a long time, and you've been very bullish about just how big it's going to be. And now I think it's very speculative to say, well, how big is big because it's going to be like really big. So if maybe we could start just with the perspective from you about what's going on out there? What's your vision of what all this means.

We started this effort -- after we sold Siebel to my friend, Larry Ellison and I sold Siebel system Larry Ellison. I recall that I think they join closed in January of 2006. We started thinking about what was next. And we've got together the number of colleagues in the industry from founders of Oracle, Intel, SAP, McKinsey & Company, [ Exensor ], you name it. And we kind of ideated for the better part of a couple of years, thinking about what was next in information technology. And what we thought next was about Elastacloud's computing, big data, Internet of Things and predictive analytics. And look -- and guys, let's think back to 2006, I mean the AWS was selling like this much in terms of cloud. It was exactly how big the market was. So we believe this was going to be a big market, and so we came up with this idea that we would build a software platform that would allow organizations to take advantage of those -- that step function and technology that consists of those items that I have identified. We set on our e-mail on a Friday, raised $20 million by Sunday and the company was financed, and we began work in January of 2009. Okay. Now we believe that we have been at this kind of very vocally now for 14 years, and particularly the last decade that we believe this market for predictive analytics applications applied to the enterprise was going to be a large addressable market. And we're kind of -- that not everybody share that perspective, and I understand that. But now fast forward, and we wrote books about it and we give keynote speeches about it, and we sold a little bit of stuff into the space in the last 14 years. I think we've sold maybe booked a couple of billion dollars in software in rough numbers. So we have -- we are able to make some things happen. But now here we are, it's June 22, 2023. And I don't think it's an overstatement. I really don't think is an overstatement. But I think on June 22, 2023, there is not one CEO in the world who didn't think about AI today. There's not one government leader in the world who didn't think about and have a meeting about AI today. And there's not One investor who didn't think about AI today. Now there's lots of people who didn't think about China, lots of people Who don't think about quantum computing, lots of people who didn't think about Ukraine or climate change today, but nobody didn't think about AI. So we've never seen anything like this. I mean we've been involved in some large addressable markets, relational database, enterprise application software, CRM, cloud computing, et cetera. But nobody has ever seen anything like this. So however big we thought it was, it's One to 2 orders of magnitude bigger than that. And so it is -- what's going on in the marketplace today is really quite remarkable.

Well Tom, the machine learning, has kind of been your main business of AI in the past. But now with generative AI and large language models, it creates the opportunity for the new user interface to the product, how you interact with the system, which means you can broaden the number of users. But these large language models, I read an article recently, someone said, well, just because you can memorize the Internet doesn't make you smart or intelligent. But because of your unique architecture, you've got the logic layer, which is all the AI applications underneath the machine learning. So when you ask a question, you've got this layer that goes right into your -- I presume it's in the Vector database, but it goes to the AI to answer the questions. So it's not, in your case, just about a natural language interaction. It actually is very substantive because you have the intelligence to back up the questions that are being answered. So if you can kind of talk about what the natural language means to the company in terms of the UI. But also does this reduce the friction for adoption in the market as well as people start thinking that, oh, we really need AI even though they don't necessarily understand what that means? You go much deeper than a natural language, you get right into the substance of the AI applications.

Okay. So if I may, and I think I need a clicker because I want to show a slide I left it up there, I'm going to talk -- no, I don't have a -- go backwards. There you go. Okay. This is a $1.5 billion worth of software engineering, and this is the C3 AI platform, and this is a model-driven architecture. And this is -- what is different about C3? This is what's different about C3, okay? So if you want to -- you may or may not be interested in it when you're done, okay? But this is what's unique about C3, is we spent $1.5 billion. Hello? Oh, you can't hear me? Are we live? Okay. Good. So this represents -- I don't have to count, but some lies between 1,000 and 2,000 person years of work, okay? This software platform, this is the C3 AI platform. And What we have here in one cohesive architecture are all the software services necessary and sufficient to design, develop provision, operate massive scale enterprise AI applications. And we do this in the Department of Defense. We do it in the oil and gas industry. We do it in utilities. So the -- so what are -- on the left side of this or the left side of this, you see these are all the data fusion capabilities where we take ERP systems, CRM systems, SCADA systems, weather, train, social media, open source data about unemployment, stock prices, commodity prices, GDP growth rates. We aggregate those data into a unified federated image, okay? And then we persist -- we have a family here. You see all the data persistent technologies, relational, nonrelational key value store, vector databases, what have you. Down below, we have all the platform services that are necessary for applications of this nature, like encryption and motion, encryption arrest, access control, ETL, okay, what have you. Then machine learning services, supervised learning, unsupervised learning, deep learning, reinforcement learning, natural language processing. Then we have -- on the left side, all the -- this is a 3D image of what you see up there, all the kind of data visualization tools. We've enveloped this in a family of applications development tools, deep code, low code, no code application development tools. And we've used that to build 42 turnkey applications for the defense industry, intelligence industry, banking industry, telecommunications industry health care industry, manufacturing industry, utilities industry, oil and gas industries, et cetera. Now -- As we enter June of 2023, and everybody is interested in how they can use these technologies for -- oh, by the way, when we build an application in this environment, any application in this environment, that application will run without modification. Look at the bottom row of this guy, okay? It will run without modification on the AWS Cloud, the Google cloud, the Azure cloud, behind the firewall and on the Edge. Hey, this is a pretty neat trick. Now so when we -- so now we have these 42 applications. So as we are -- as we enter now the second half of 2023 and this enterprise AR market is exploring and everybody is trying to figure out how to use these technologies to be competitive in their business or else they will go out of business unless if they do not use them. And there is exactly one company in the world that has 42 turnkey AI applications. One company, there was actually one company in the world that's built a platform, one company, okay, that has built a platform that enables these applications. Now let's talk about generative AI. Generative AI did kind of change everything because when this was -- we've been working with Nikhil and his group have been working with generative AI, the data science guys at C3 since maybe 2020 on a number of interesting applications. But we got we started to work on this in earnest in September of '22 associated with the crest that we got from a customer of Graham and mine and the Department of Defense, who said he wanted Google for DoD. And this is the person responsible for all AI platform standards and all the DoD. He said, "Tom, we need to be the Google for DoD." A customer asks a question and gets an answer Okay. And then we -- so then I got in a room with my colleagues Ed and Nikhil and Henrik, I don't think Henrik's here today, and who's our lead data scientists. And we started sketching our solutions. What is Google for DoD mean? And so the idea there was to -- in generative AI, I know it looks like a step function to the rest of the world. For us, it just looks like the next step. As this AI has developed from IoT on supervised learning, the supervised learning, say, someplace in there back here, you have NLP reinforcement learning and now you have generative AI . What happens next, I don't know. But it's certainly when this thing came out in November of last year. From November through to today, it has certainly fueled the market. Again, everybody scared. Everybody wants to know what this means. And -- but it looked to us, basically, it was just another tool that came available, and our platforms allow us to just put it in our machine learning in our AI level and now we have generative AI. Now in terms of what large language model, we put in that, okay, in that model, I mean it could be chat, it could be Craft. It could be a Plenity from Google. It could be -- what is the other real popular one from -- No, barb is doing in LLM. Pom , A Pom, we don't share. We don't care What is in there. We can take advantage of any one of them. So what is the effect of this in terms of what we're doing in LLM. Now I can tell you, and these are great companies, okay? And we're not inventing the LLM technology. There is billions being invested in this by Microsoft, by open AI, by Amazon, by Microsoft, by the academy, MIT, what have you. And as these technologies become available and the next one becomes available, that's more powerful. We'll just -- or it becomes available for the banking industry. We'll just take advantage of it. And that's the way that this architecture works. As these new technologies become available, we'd take advantage of it. If somebody wants to use, people want to know, gee, how do we compete with databricks? We don't compete with databricks. Databricks is just a data virtualization technology. We have data virtualization. It's right here. We have been -- Databricks came out of the University of California, right at Berkeley. It's a project called AMPLab . It was developed by 6 graduate students in a year, 6 years of software in many years, person years of software engineering. Reported to the Chairman of the Department then by the name of Michael Franklin. Yes, the Thomas Siebel Chair computer science, okay? Now these guys took that production that they call Spark and they put it into the open source community. And they came out where they commercially a supported version of Spark, they call it data bricks. Great idea, a great product, a hugely successful company. Out of the box, our product comes, we happen to use the successor to Spark, which is called Ray, which we think has a little bit more utility. But many of our customers sell DoD, Booz Allen, others, they want to use databricks. No harm, no foul. Take out Ray. Put in databricks. Okay. And you're running. So these things that appear to be competitors, they're not competitors. Now let's talk about LLM. This will be the last -- I will answer your question, Rick. Of the all of these companies, great companies. Microsoft, Google, Amazon, Accenture, IBM, Open AI. Not one of them today ships a production software product for enterprise to deploy a generative AI at one level, not one. Okay? Now the way that all of their product architectures work, if you were to sell them, they work like this. You can install any data you want. You're going to integrate any data you want into the LLM as long as it's text, HTML or code. The LLM has direct access to these data. Okay. And so what happens then? Number one, the answers are stochastic. You've used ChatGPT, you use your power, you know there's stochastic. Every time you ask a question, you get a different answer. Okay. Secondly , there's no traceability. You can't see where the answer came from. I mean if you're running you want to know whether you're the Chair of the joint chiefs of staff and you want to know where your satellite gaps are and into Paycom, and wanted to give you the answer. You got to know where the answer came from. So they're not traceable. Stochastic also, if 2 people ask the same question, they get a different answer. Thirdly, none of our enterprise access controls at Bank of America, at Department of Justice, at DoD are in force. So anybody get access to any data. We have a risk that you've read about, and it's very real of LLM caused data exfiltration. See Samsung for details. So the LLM has direct access to the data. It pumps all your IP out of the unit app. In a lot of organizations, that's kind of problematic. Like any commercial implementation, it's problematic. And finally, all of these solutions, all of the slideware to work, okay, when they ship it in when they get it into production, it's prone to hallucination. For example, if the -- if it doesn't know the answer, it makes it up. I can go on to Bard and I can ask who are the largest institution of shareholders of C3, it will tell me the answer. And then the next question I can ask to the question that it can't possibly know the answer to. I would say, who are -- who has the largest short position in C3, it will come up with 11 institutions and then tell me what their short position is. It's absolute poppycock, it's unknowable. You guys know it's unknowable. But it'll just make -- so with equal authority, it'll just kind of wing it. So our solution is a little bit different. So we're putting generative AI on top of a couple of billion dollars’ worth of software engineering. We are the masters of the universe at data fusion. So ERP, CRM, SCADA systems, sensor data, whatever it might be, we can aggregate those data into a unified federate image. When we do so, okay, when we do so, rather than exporting these, bringing these data into the LLM, we're bringing it into a deep learning model. And that deep learning model is storing these data in a vector data store. Now we have built a firewall, okay, between the LLM and the rest of disguise. So the LLM interaction with us, okay? And these are where the data are stored and the data are secure. So in a way, we kind of think about this as the brains and this is the memory, and we have severed the -- we have disconnected the 2. So when the LLM figures out what we want to ask, it passes the answer through the firewall to a retrievable model. The retrievable model gets the answer out of the Vector data store. It gives us only the answer, and then it presents the answer to us. So what is -- what are the advantages? The responses are deterministic. Every time we ask a question, we got the same answer. If 2 people ask the same question, they get the same answer. Secondly, everything is traceable. Once it gives us the answer, we can click on it, and it tells us exactly where the answer came from. It takes us right to ground truth. Thirdly, all of our access controls at Bank of America, CIA, defense counterintelligence security agency, whatever it might be, they're fully enforced. Fourthly, there's no risk of LLM cause data ex filtration because the LLM has no access to the data. So we don't have to worry about IP disappearing in the. And finally, the darn thing doesn't hallucinate. If it doesn't know the answer, it just says, "Hey, I don't have access to these data. So it's fundamentally -- then the most important thing about the C3 AI generative AI product is just shipping into production today. And we're installing it today at the Missile Defense Agency, a large intelligence agency that I can't name. Georgia-Pacific, Fit Hills resources and some other organizations that I'm not prepared to announce yet. But the -- but we will have, I don't know, in 8 weeks, I'll have the first 6 users up in production before anybody ships the product. So it's a little bit different.

So yes, What I do it's unique in that it leverages the platform, which touches data all through the enterprise, no matter whether it's on-prem, hybrid, cloud, if you were to ask Oracle for an answer is going to look at Oracle database. It doesn't have the ability to look at all the data sources that...

It is Small language model Exactly. So we're not connected to the Internet. And for these types of applications, nobody wants to be connected to the Internet. So you can basically get -- you have Access to the information within the Air Force or within Georgia Pacific or with Ever Charles Coke, within Coke enterprises. And so you're limited to the firewall. You're limited what we have to do. I think our time is up, Rick. And I think that with this to keep this project on schedule, I'm going to turn this over to my colleague of now many decades and Chief Technology Officer and President of C3 AI, Ed Abbo. Ed, please.

Okay. What I'm going to do is Just give you an update from the field on how we're seeing the broad interest in AI translate into sales and the sales organization activity over the quarter. And so if I start by just laying the context of what are our field teams doing out in the market. So they're basically engaging with prospects and customers and identifying basically a pilot, which is think of that essentially as deploying our applications into production. So it's a pilot, but it's production use of the application. In a limited fashion at a customer. So that application might be using AI to improve demand forecasts. It might be using AI to improve reliability and throughput of manufacturing. I mean so the first conversation that we have is where should we really start. Many customers have appetite for putting AI everywhere. But you have to start somewhere, and that's the pilot that we're basically engaging in discussion. Now that discussion has accelerated, but it's roughly just under 4 months to get that pilot identified and then basically an agreement concluded. After that, we basically have up to 6 months to put an application into production deployment. And sometimes it takes 3 months if we're talking about generative AI. Those are pretty fast, and we can deploy those in just roughly 12 weeks. And other times, it takes longer because you're actually integrating many systems and putting something into production from there. At the end of that, we've actually demonstrated value to the customer. So they see the value of improving reliability in their plant. And then we convert that into a production usage, and that's where the consumption that Tom talked about kicks in So that's $0.55 per virtual CPU hour. And as they expand the deployment, we basically -- they get more value out of it, and we get more consumption revenue from it. So that's the model that we're basically driving in the market. So with this broad interest in AI, let's see what it's doing to basically the metrics that we look at on a day in, day out and weekly basis. And so first is the pipeline, and the pipeline is basically doubled in qualified opportunities from the prior year. So we went from roughly just under 300 qualified opportunities to over 600 qualified opportunities at the beginning of this fiscal year. Sales cycles are shortening. So 6, 7 years ago, they were over a year in discussion about the cloud, about IoT, about AI, et cetera. And now everybody is basically on board and want to deploy very quickly. So we're down to under 4 months average sales cycles here. Pilots. So significant interest in engaging in pilots and doing them. And so this is quarter-to-date. So roughly halfway through this quarter as compared to halfway through last quarter, you'll see this triple the number of pilots that we've actually closed relative to last quarter. Number of deals have also accelerated. So it's a comparison to, again, midway through the quarter a year ago for first quarter and now. We've basically seen a 60% increase in the number of deals that we've...

Quarter-to-date.

Yes, yes. It's quarter-to-date. So at this point in the quarter, roughly halfway through. A year ago, we had 10 deals in this year -- this quarter, excuse me, we've done 16 to date. Where are we -- that's roughly 37% of the pilots that we closed in the last quarter. Manufacturing is also strong with 15%, 16%. And then high technology, oil and gas are just over 10%. And a smattering of interest from other industries representing 5% each. So that's where we're seeing pilots. Let me talk about specific pilots that we're doing and then also conversions and expansions that we have underway. So last quarter, we did 19 pilots through the quarter. And you can see each pilot is with a -- has a specific focus area. So for example, if you pick Owens-Illinois, they were focused on reliability of their facilities. If you look at Nucor as also a new logo, that was production schedule optimization. So that was kind of where they wanted to start with us. This quarter, so far, we've done 11 pilots, and so we're well -- we're on track to exceed last quarter's number. And you can see we have now also closed a deal with Nucor for generative AI. So last quarter, they started with production schedule optimization. This quarter, they're doing Gen AI with us. Let me talk about conversions and expansions for a moment. So last quarter, we announced renewals at Con Ed; renewals at NIPA; renewals and expansions at the U.S. Air Force with RSO, which I'll talk about in a moment; and then also pilot conversions. I think this is basically drilling into a pilot and how it converts, which is another metric that we're tracking and you all are tracking. So if you take an example of an industrial manufacturer, they wanted to basically do -- take advantage of all of our applications, but they wanted to start with one. So the first one we did was reliability, and this was to improve the reliability of certain equipment in their facilities. And so we picked one facility and demonstrated that through applying AI to this -- to the data from the equipment, from the maintenance systems from the process systems. We could actually improve the run time for that equipment by 20%, pretty significant. What that means is that the asset utilization for that facility goes up by 1% to 2%, which is worth tens of millions of dollars to this organization. So that's a pilot that we did in about 4 months and then converted. In fact, let me share with you the time line for this particular customer. We met with the CIO and their team, June of last year. Three months later, we had signed the pilot. So we went back and forth as to where we should start, but we ended up with the reliability case that I shared with you. Within 4 months, we had completed -- concluded the pilot and basically demonstrated to them the 1% to 2% asset utilization improvement. And then we converted that to a production deal in another couple of months. So this is a very typical sales cycle, starting with the pilot delivery of a production application and then essentially the conversion to production. One more example is a large food processor. And here, we're applying AI to improve the demand forecast. If they can figure out what their customers need, they can basically manufacture the right stuff and actually distribute it to the right locations. And for this food manufacturer, we improved the forecast accuracy by 12 percentage points. So that's massive. It means they basically can reduce the amount of waste -- food waste that they have pretty significantly, and now that's converted and expanding across all of their SKUs or their products, if you will. And I'll conclude with the work that we've been doing with the U.S. Air Force and the Rapid Sustainment Office. Again, this started with one aircraft platform, which was the AWACS. Then they gave us the second one, which was the C-5 Galaxy. And from there, we now are designated as the AI predictive maintenance system of record for not just aircraft but for everything in the U.S. Air Force. So it's a pretty big accomplishment. If you look at the expansion, we cover 16 aircraft platforms in 16 months, and this is pretty remarkable in terms of the speed at which we accelerated from a couple of aircraft platforms across from 16. And the economic value here is measured in billions, so this is not insignificant in terms of the readiness, the throw we associated with the Air Force as well as reduction in maintenance costs that they can get out of this. So with that, in summary, we're seeing increased sales activity from all the interest in AI in the marketplace. Sales cycles are shortening. And basically, the pipeline is growing. And I would say that our prospects are very encouraging. So I'll conclude there, and really excited to welcome Honorio Padrón to join us. Honorio is Chief Technology Officer over at Pantaleon, and I'll let him tell the story of the pilots that we're engaged in. So thank you for joining us, Padrón. Mr. Padrón.

Thank you.

Thank you. There you go.

I'll take one of those just in case I need it. Sparkling, right? Thank you for inviting me, Tom and Ed and the team and for trusting me up here with such an important audience. I'm going to tell you very quickly about Pantaleon. You probably haven't heard of us unless you looked at that reasonably just to give you a context of where we are and what we're trying to do. So Pantaleon is a sugar company. We're a family company, 173 years old in headquarters in Guatemala City and with operations and some other countries, and I'll talk about that. And one of the key things here, which is a message that I will develop a little bit more, is that we want to be in the top 10 sugar companies in the world. And the only way you can do that is not only with great talent but with great partners, and I'll tell you why I think this is a great partner for many reasons. So the family owns several businesses. Today, we are the #1 producer in Central America, and we're about 16 in the world. We want to climb to 10. We do about 1.2 million tons of sugar a year, but we also -- and this is a very interesting business. And by the way, I tell you more about me, but this is the first time that I've been in the agricultural business. It's fascinating. Also very complicating. Every sugarcane they get to this mill is different because of all the environmental conditions and how it grows and fertilization and the rain and all of that. And this industry is also extremely complicated and very sustainable in talking about today's world with all of that. We have no waste. We use our bagasse to generate electricity. We generate alcohol. We generate molasses. Obviously, sugar is our main product. And our objective and what C3 is helping us with and what I'm going to talk about is how do you maximize the best sugar came to the mill. And once it gets to the mill, how do you maximize the extraction of the sugar. Because even though we use everything else, it's not at the same price point. And we're complicated. We are in 5 countries. We have a trading office in the United States in Miami. We have one in Chile. And then we have operations in Guatemala, one mill. One mill in Nicaragua, difficult to do business there. And very difficult to do business in Mexico, where we have 2 mills. Mexico, I'll give you these details, you understand how difficult it is in Mexico. You cannot own your own sugarcane. You got to buy it from small producers, and the government tells you what you're going to pay those producers. Imagine. And we got to pay them in advance, but you never want to get ahead of them. So perhaps, AI can help there, too. But that was just to give you a little bit of a background to show you that we are vertically integrated in some places but not in others. It's a very complicated business, and we need to be -- digital transformation, AI, all of those things needed to be in the picture. Okay. So let's talk about -- a little bit about me now and why I'm here and what are we doing at Pantaleon. So I know this team. This is my second time working with this team. Worked with them at the beginning of Siebel Systems and had a great experience with all of the aspects of working with a partner like this. So when it came to making a decision here, this was also a component. I have had 6, 7 CIO roles, been around. I'm one of those CIOs that was hired to change stuff and move on to the next one. I was actually retired from being a CIO. I was consulting. I did 15 years of consulting, and I was consulting in Latin America when Pantaleon to semi arm to come in and do a total digital transformation. And we went at it back in 2019, with changing everything end-to-end. This was -- I was in Disney World. This was a dream, end-to-end, from buying toilet paper to processing the sugarcane. We selected an SAP basic platform and then got us to the situation where we now have a decent foundation to be able to move into the next level of evolution in the company where we begin to apply AI. And AI has been around for a long time. We all know it, but I will tell you a little bit about my perspective in terms of being a head of IT. And I would tell you that I learned this, I have a lot of scars on my back because back end, I'm going to date myself, back in the '80s and '90s, okay, we were all out there with our ego in our hands, trying to think that we were developers as Head of IT. And you may remember when ERPs were not really -- or you may not remember because mostly we are young, but you can read about it. Back in those days, there was no SAP ERP, Oracle ERP. We were all putting our own stuff together, not very successfully. Well, guess what's happening today? Most of the CIO, CTOs that are there think that they can do that, and they're grabbing tools from everybody and trying to put it together and not doing very well. And companies like Booz Allen that recognize that there's a platform out there are jumping ahead, but there are others that are perhaps making a lot of money because I was at a conference last week with a bunch of CIOs, and there was a telecommunications company there that has been working with one of the major provider tools for 2 years now, and they're not getting there because IT departments should be applied technology. They're not in the development. That's what people like Tom and his team specialized on, and they've been doing this for years. So as a CIO, I'm very, very keen to that as a CTO to buy apply technology. So I went out there and I looked. And Tom is right. There's not another platform out there. That's already integrated with all of the services. And one of your guys made -- one of the businesses that I made to their Compo headquarters made an analogy and said everybody thinks they're a general contractor, and they go to Home Depot. And they buy the wood and they buy the tools and they're building a house. 1.5 years later, their family is still homeless, okay? What happens here is in terms of the model up here illustrates it is -- he's got the house prefab, but you can change the roof, you can add another room, you can do multicolors. That's what IT departments should be on. And I think it's just a matter of time before the [ kneeling curve ], and I think we're already there. And folks like me realize that don't be messing around with all those tools and trying to do it like the old ERP folks that didn't succeed. Go jump on a platform that accelerates your success in terms of what you're trying to do with your company. So finish digital transformation. Basically, we got the last part of it in October, coming up now in Mexico, and we selected C3 to begin to do the kind of work that we need to do now to be more competitive. And in our case, it's not just being competitive, surviving because our margins are so small. We're in a controlled commodity environment. We can't make mistakes, big mistakes. We got to be good at what we do, but we won't be around. And the company has been good in a market for 173 years, but that's not going to continue unless we do better. So the aspects of what Tom was talking about is the integrated open architecture. It can use any tool really. When we brought C3 and some of our data science were saying those of you that know the language open. We're using Python. So what? Keep using it. It fits into the open architecture. The -- one of the big aspects or one of the biggest challenges that we have was the integration of the data. And for every model that we were going to do the old way, you had to do a data extraction because I can't let them touch the live data because they may bring the system down. The way that C3 works without getting into a lot of details, that problem goes away. Goes away because you set pipelines and then you bring the data. And all the calculations are being on their side, not on your side, you'll never bring the system down. So things like that we looked at in detail made us very excited about the selection of C3. Now this one here is one that I don't think there's enough play. There are not enough people out there to take care of all the AI demand. They're not going to be. And with this platform and the way that C3 works, which I have it on the next page, you have the need for smaller teams. You don't need 10 data scientists. And the model that they have is one that it is not like a systems integrator where they want to move in with you. They want to teach you and get out, and you do it with your people. But everything that you're doing is reusable, so like that reference data model, to use it every time. You want to add 3 more data, you add it, we put them in their relationship. But everything you did before with the prior 20 models is reusable. So it really reduces the need for talent. And like I said, that one doesn't get enough play. So what are we doing today? Very exciting. We started with harvesting planning. So what happens? You do harvesting plan at the beginning of the harvesting season, which for us that runs November through May 7 by 24. You start and you run out of sugarcane and the rain lets you, you don't stop. And again, the key here is getting the best sugarcane to the mill. And you set your plan, but stuff happens. And with the optimizer that we're using here because one of the key things is that the software that comes as a part of the kit that you can select, the 40 different options, they're not just predictive-type activities. So we're doing optimization as well on top of being predictive, which most of the models that are being done out at there do not do that, again, as you put it together with your kids. So we want to know 2 weeks out, where do we go harvest because we have a plan that says you're going to go here, here, here. And for instance, in the Guatemala mill, we have 7 crews at the time. Where do you send those crews tomorrow? Yes, criminal burns. Something probably strange to you, but our fields get burned all the time criminally or a range to mulch or you have molasses. So all of those conditions are going into the model. And then you know where to go next. We also have -- and the second one that we're looking at is what we call undetermined losses in the mill. This is stuff that happens. And interestingly enough for every batch of sugarcane that comes into the mill, you adjust your machinery. For every one of them, we take samples, there's a laboratory testing process involved. It's very complicated. And we got to get that measurement to the floor really quick so that they can adjust the machinery. And imagine doing all of that from paper to manual to now all automated and now putting intelligence into it, it's going to be a fantastic closed-loop integrated process that is going to allow us to minimize what we call undetermined losses. And most exciting of all, and we have a little note here that says pending confirmation, well, they got confirmed while I was sitting there. So we have now 2 Gen AI projects starting. And the first one, I'm going throw a number out there. It's going to be $3 million to $5 million a year benefit. So what we have today is a number of service contracts through all of our mills that we really don't know much about. That is one area that has not been integrated. So we're going to put the Gen AI activities on top to be able to analyze those contracts, optimize them. We know that we're coming from, again, from being acquisitions, from being disintegrated. We may have with one vendor that fixes air condition 5 contracts in 1 mill. And all of those things, you don't have enough people to really understand that, analyze it and optimize it. And then the other one is one that's very key to us because in forecasting and contracts. We sell 80% of our production before the season starts. Now we also sell everybody else's sugar. But our own production, 80% of it is on new contract before you cut the sugarcane. How do you price that? And today, there's a spreadsheet. There's about 50 columns by 150 deep, that one person does, and I've been wanting to build that person in a cell with life insurance because if we lose that person, we're in trouble. And we're about to fix that with Gen AI because a lot of documentation that comes in from the industry and what the El Nino, La Nina all of the weather components, all of that, that today, it's just not humanly possible to do that in the most effective way. So four, and we're looking at more possibilities and I know they're endless. So one of the things that -- I don't remember if you had it in your statistics is that you may take 3.7 months to sell your first pilot. But your second, third, fourth and fifth is now I got everybody in the company hungry. They all want one of those. So the issue that we're having right now is prioritization more than anything else. I think I'm running out of time. I want to go through fast here. We're in process of doing a lot of the stuff. I would tell you the experience that I have with this team is being repeated in terms of the trust that you can have, the confidence and the quality of the work. And one of the things that the team has brought when we decided to do business with C3. We didn't just partner with a platform, we partner with an excellent team. And we're amazed that the number of people that have had to apply for somebody to be selected. So they have, what, last year was like 92,000 people for 300 positions, and we see that. And one of the other benefits that we're getting from the staff is that they're not coming in to automate the processes. They're very, very smart and learning new process in suggesting how to fix it with the technology by reengineering it, not just a copy paste type of activity. So that has been an added benefit that we sort of knew that it was coming as we interview the folks, and we saw the team that they were going to put on this particular project but it is materialized. Our head agricultural guys said the other day, and he's a guy that I have a lot of healthy arguments with because they always want to go do their own thing, and they always come up with their own technology, and they want to be stand-alone. And he said the other day, and if anybody doubted, I'll get him on the phone to tell you, that he wouldn't be doing the project, the first pilot, the harvesting optimization with anybody else other than C3. And he was out there looking at all kinds of companies to bring me. And typically, what they do is they go over there, they get the technology and say, I want you to use this. But it doesn't fit. That was a cycle. In this situation, he said, "I will not do this with anybody else other than C3," and that's coming from the agricultural guy. Trying to expand, looking at a strategic plan because I can't do it any other way. So I'm sitting down the teams and say, "Hold on. We have a tool. We have a partner. We have a platform. We have ideas. We need to sit down and figure out an AI strategic plan," which most companies are not doing. Again, they're going out there. Let's go use this tool. Let's go do this over here. I have one executive that had already worked out a training plan with HR, and they were going to tomorrow release ChatGPT to the company. I said what are you going to do with that? How are you going to use it? Do you know that there are security issues? What platform is it going to be on? Are you going to allow them to touch company data with that. And a lot of companies are doing them. They're not doing their strategic plan, and they're not looking at, can I use a platform that's already put together instead of me trying to buy a Lego thing and assemble it and not get there effectively. Thank you very much. You're next.

Honorio, that is -- hello. Testing, testing. Good. All right. Honorio, that is a really tough one to follow. I mean 2 pilots during your presentation, so I'm not sure if I can top that. But let's give it a go. So good afternoon, everyone. My name is Merel Witteveen. I lead our alliances organization at C3. I'm very excited to be here. Prior to joining C3, I was at McKinsey offices, and I came to the Bay Area for Stanford Business School, fell in love and met C3, and the rest is history. A little fun fact in case some of you were googling. I actually used to be a professional athlete, and I went to Olympics in 2008 and won a medal there. It's not what we're going to talk about today, but it's probably just as exciting, so it makes you feel how I feel about my work here at C3. All right. Let's do this. At C3, we have a global partner ecosystem. Now and you can slice and dice that partner ecosystem in many, many different ways. You can draw a big circle and then add some system integrators on the outside. You can create a cube or a box. But I think what is most relevant for you today is actually to sort of categorize it into our go-to-market partners and into a whole set of additional partners. Now I can speak for hours about the second category, but actually, we're going to focus today on the first category. When we go to market with our partners, we always put the customer central. So we really start with the customer with a prospect, and we say what will actually benefit the customer. And we try to assign the best cloud partner to this customer, either go to market together with them or be introduced to them or actually introduce them in some cases. And depending on the customer, we introduce a system integrator as well. Now you may ask, what's in it for the customer? Well, for the customer, they actually have to deal with a team that is fully aligned. There's no daylight between the partner and C3. But what's in it for the partner? Well, the partners, our cloud partners, when they deploy an out-of-the-box C3 application at a customer, they see consumption almost instantly instead of having to build these solutions over time and sort of slowly building up that consumption. So there's a lot of incentive for cloud partners. Now of course, for system integrators, they increased their footprint. And for both of them, it's truly a unique product offering. Let's go next. I'm going to highlight a few different examples of our partners today, and I'm actually going to start with our Google partnership. Our partnership with Google was signed in 2021, and it truly is unique because it has sponsorship from a -- from the top level. So our CEO, Tom Siebel and Thomas Kurian, the CEO of Google Cloud, are personally very involved in this partnership, and they really make sure that it runs successfully. Google has made a significant investment in C3, and we're actually -- they have a dedicated co-sell team that works with our sales team to go to market jointly to our joint customers. The partnership is buzzing, it's humming, it is very strong, it's going fast. And actually, in the last year, we've increased our pipeline, our joint pipeline, 4x. So it's a huge partnership. Now an example of how we would partner with Google could be, for instance, our generative AI solution so that Honorio was just referencing to. What you could do is you deployed a C3 AI, generative AI on Google Cloud, and you can leverage several additional Google services. In this case, that could be your large language model, the POM 2 model from Google. So we leverage their services. So it's very clear. We're not a competitor in this case, but we're truly complementary to each other. Now to show sort of the success of this partnership, the Google team agrees with that as well. And last year, we actually won the Partner of the Year award in AI and ML. The applications are in for this year, and I'm very hopeful that we're looking good for this year as well. And I actually want to highlight, Honorio, in your conversation as well is that our Google team was very helpful in this deal as well. This is a little bit of an exception though because C3 introduced Google to Pantaleon. And that is truly sort of, I think, what shows the strength of the partnership. It's mutually beneficial. Let's say, usually C3 gets a lot of benefit from leveraging the Google customer base. Let's move on to the next one. Here we go. AWS. AWS has been a long partner of C3. So almost since the inception, what Tom was referring to when he founded C3, and we have strengthened or sort of expanded our partnership with them back in December of last year of 2022. And the AWS team came to us and say, "We see you do all these successful things with other partners. We want in. We want to go to market jointly with you. We want to co-sell." And they made a huge investment in us, in for instance, our C3 law enforcement application. They invested in a demo and that now leverages natively AWS recognition, Open Search and several other AWS tools. So again, truly showing that you deploy it on the AWS infrastructure, you leverage these tools, but it's not competitive, and it truly shows the unique product offerings. Just to sort of show. Currently about over 50% of our customer base is deployed on AWS, and we have increased our joint pipeline by 24% in the last quarter alone. So just sort of that recommitment from both parties really increased some additional go-to-market work here. One of my favorite examples of AWS as a working relationship here is our work with Koch Industries, one of the largest private companies in the U.S., where Koch Industries has selected AWS as the strategic cloud partner and C3 as their AI enterprise platform of choice. And the 2 of us actually worked really well together. We started out in Georgia Pacific for reliability for paper machines with one successful use case. And we're deploying that now in many, many other business units. Now the beauty of this is that AWS sees the value in the way C3 deploys these pilots and then goes across business units to scale up, and they're actually investing in quite a few of these pilots just to truly show the commitment there. All C3 AI products are available through purchase on the marketplace by purchasing either the C3 AI Pilot or our generative AI product offering. And what's in it for the customer here is that it truly simplifies the procurement process. So going from these lengthy cycles that Ed spoke about, purchasing through the marketplace simplifies a lot of the procurement. Now what it also does for the customer is that it allows them to draw down their commits. So in case the customer has a large commitment with Google Cloud, with AWS or with Microsoft, they can purchase our software on their marketplace and then draw down on that commit. So it's very beneficial for the customer. Now my last example, certainly not my least example, is a very new partnership at C3. Tom alluded to this already, and we're very excited about this new partnership with Booz Allen. It is -- we signed this partnership back in December, I believe, Graham, if I'm correct, yes, in December. And since then, we have already closed 18 deals together. So this partnership is speeding up very, very quickly. Booz Allen has also made a big investment in training a lot of their people on C3 AI platform. So what that means is that we're going to be co-delivering a lot of projects, mostly currently in the defense space. An example here is the work that we do for the Department of Defense, Chief Digital Artificial Intelligence Office or CDAO. And what we're doing here is we're focusing on a set of use cases, C3 AI contested logistics, C3 AI [ commander's ] dashboard, presidential drawdown, Raven, RSO CBM+, and all of these are deployed on the Department of Defense Advana's platform. Now in these applications, these use cases, we're deploying them in these agencies here on the right-hand side, so at the U.S. Air Force, at SOCOM, at Transcom, DLA, the joint staff, J4 and OSD. And I'm sure Graham, who is to present right after me, and he's going to go into a lot more detail of these. So I'll give him a little bit of a nudge here. Now what's next for the partner ecosystem at C3? Today, over 60% of our pipeline is codeveloped with our partners. Now what does that mean? Well, in many cases, partners come with new prospects or new customers to us and also the other way around. And then we keep track of which partners on which account, and we continue to develop and truly go in and sell together. So the alliance ecosystem at C3 is very strong. I am very excited to see what the future brings or bring the future as you may since I'm very involved in this. And with that, thank you. And I would love to introduce Graham Evans, Vice President at Booz Allen.

Thank you. I got it. Thanks. So Tom, you made a comment about my photo in your intro. It's true, that photo was taken before I met you. And I love -- if we have a break, maybe you can give some tips on how to age gracefully in the tech industry, that would be great. All right. All right. So thank you very much for having me, C3. I'm really excited to talk about our partnership. I'm Graham Evans, Vice President at Booz Allen. I lead a lot of our data platforms and enterprise analytics work we have with our federal customers. But first, I want to talk a little bit about what Booz Allen does. We were formed over 100 years ago. We are -- the base of our capability was formulated out of the commercial consulting industry, but now we are primarily focused on the federal industry, so defense, civil and intelligence industries. We have over 30,000 employees. And we mainly focus on consulting engagements related to analytics, digital, cyber and engineering. I'm really excited, and I'm super proud that almost 1/3 of our organization is veterans or folks that are currently active in the military, which gives us great connection back into the mission of those customers that we serve. And it's truly inspiring to be working alongside the war fighters and veterans and civil servants that we are serving every day as we deliver our capabilities to some of the nation's most challenging problems. As Tom mentioned, the market for enterprise AI applications is substantially larger and growing at a much greater pace than anybody, I think, ever expected. And for our federal customers, this creates both opportunities and some threats. So from an opportunity perspective, this might take the form of creating some type of economic benefit or financial efficiency within a government agency like mitigating fraud waste and abuse or, as we've learned about earlier, using predictive maintenance to identify cost savings for an organization. Now there's also threats. Obviously, our adversaries are leveraging AI as an instrument of warfare, and we have to be prepared to be able to defend against that, and so we're really excited to be able to be part of that solution to create AI capabilities to improve the decision-making within the defense organization from the battlefield to the boardroom. Of course, Booz Allen has also an outspoken leader in the areas of the ethical use of AI in the federal space, where it's become a hot button, especially with the ubiquitous use of AI through autonomy capabilities as well. So we are positioned for sustained, diversified revenue growth given our central role in critical AI pathfinder and enablement programs, which C3 is a partner of ours on a number of them. And we're also one of the largest providers of AI solutions in their federal space. We have over 150 projects within the federal government, where we're actively producing AI capabilities to solve, like I said, some of the most challenging problems that our customers have. So this appear is sort of our comprehensive approach towards delivering AI capabilities to some of our customers from strategy through operations. So this obviously involves delivering some type of IT capability, and that's not without its challenges, as most of you probably already know. I'd like to kind of illustrate that with a little scenario with one of our projects, and I can't really talk a whole lot about it. But from what I can share, we have one team, which is basically a digital SWAT team. And they get deployed into challenging government organizations with a specific challenge to say, "Hey, go. I've got this issue. And I need to go collect data, and you just solve this problem using AI/ML or whatever. Makes sense because I can't figure it out." And so looking back over the last couple of years, the team was engaged during the COVID-19 pandemic to integrate all of the different data sources that were coming in from the states, from the different services to give commanders a better sense of what's actually going on, how do we support our war fighters, how do we make better decisions, how do we continue to fight. And in that day, we built a data platform. We spent many hours and days and months and weeks trying to create an integrated AI/ML pipeline to solve some of the challenges around predicting when the pandemic would cause a particular outage in a supply chain or something like that. Fast forward, we took that same lesson learned with that digital SWAT team, and we applied it to the Afghan evacuation when the Taliban was coming in. So we had a team deployed that was helping look at how we could leverage data and AI to essentially bring our allies and partners from Afghanistan to locations around the world that had the appropriate medical facilities and housing and things like that. And then finally, in the most recent update, we had that team deployed looking at -- when Russia invaded Ukraine, how could we ensure that the supply chain for our allies and partners was not interrupted, and we were giving the right types of capabilities to our allies and partners. The sort of the challenges that we faced during the entire time was the same. We had these bespoke solutions that we created every single time we had an engagement. And again, we spent all of our time building out the infrastructure. We spent all of our time orchestrating data operations. And what we identified is that in order for us to grow, in order to continue to repeat that for each of our customers, we needed a common foundation. We needed a common tool set that allows us to focus more of our time building out the more interesting insights, that were going to allow our war fighters and our partners and our veterans to get the capability that they need as opposed to spending over time on the infrastructure. And that's really where C3 came in. And I think Honorio has talked about this. With an organization as large as Booz Allen, we can't have custom software solutions because it's going to create too much time in training our employees to think about the speed on how to use different capabilities, and we're not going to be able to deliver capability at the speed of relevance that our customers demand. And then finally, in the space that we deal with, there's a number of providers that want to do business with federal government, vendors and whatnot. And we're looking for differentiated partnership. We're looking for organizations that can help us deliver our capabilities faster but also help us deliver something that's unique to our customers. And really, that's where C3 has come in. C3 from day 1 has been helping us with this challenge across our federal government, and there's really 2 main reasons why we selected C3 as our partner and a number of those customer areas that Merel was talking about. One is technology. It's world-class. The other is the actual partnership itself. From a technology perspective, C3 provides that turnkey AI/ML solution that I was talking about that's such a challenge for our customers. We can rapidly deliver complex decision support capabilities, look at the next best action, we can predict what the consequences of those actions are and the results. And it fits right within our customers' workflow, and it's not something that's like a static dashboard that they're used to seeing. It's something new and unique that differentiates us in that space. Also the open architecture that Tom was talking about, where you can plug in different tools. Our customers already have investments that they've made in things like data bricks or other data visualization tools that we can easily plug in as part of that AI/ML architecture. But most importantly, the relationship that we've had over the last 6 months has really been a true different experience for me from a partnership perspective. From the first conversation that we had with the C3 leadership, it was very clear that they were an organization that was focused on the economic outcome of their customers of the solutions they were providing as opposed to just selling more licenses, whether that's important. It's really about aligning to that ethos that your customers' mission is the most important, and we really saw that with C3. It also -- the partnership also gives us an insight into what's going on in the commercial industry and see trends before they can actually be part of the federal government. The technology life cycle in the federal space, as some of you may know, is a little bit behind sometimes in what we see in the commercial industry. So what's been really interesting is to see solutions that have been implemented in the commercial space around supply chain or readiness, and we can easily apply them to our customers in the defense and intel space. And as Merel mentioned, we've had a very successful 6 months over this first initial part of our engagement. As Merel mentioned, we have 18 closed deals. We have about 19 in our pipeline that we're looking at very closely that are super exciting. And one of the things that we've been really focused on is trying to train up an army of developers. So it's not just a one application at a time kind of deal. We have folks that are ready that are making the C3 application, that are making the C3 platform a part of our regular delivery tool kit as we go and deploy new capabilities to our customers. So the first, I'd like to just articulate one of these successful programs that we have using C3. I can't really talk about who the customer is. But let's just say, in general, when we talk about delivering these capabilities to our customers in the federal space, senior executives in DoD. We're talking the Deputy Secretary of Defense. We're talking [ combatant ] command commanders. They expect smart applications today. A static dashboard they might be used to is okay, but it's not really the expectation anymore. Leaders want to be able to look at scenarios. They want to be able to predict outcomes. They want to create alternative plans on the fly using the information they have available to them. And that's really what our customer asked us to be able to do. But where we've seen some early success in some of our deliveries together have been in really 2 main areas. One is what we're calling commanders' insights. And so essentially, what that means is you've got those senior executives like I was talking about, they want visibility into their vast enterprise. I mean the Secretary of Defense and these command commanders have like unmanageably complex businesses that they must run from a people perspective, acquisition perspective, readiness perspective, very much like a commercial business but with the aspect of an impending war, if you will, that they have to be ready for. So they need information at the speed of relevance, and they need it to be dynamic. They don't want an army of PowerPoint rangers roaming around the Pentagon to be able to give them a week old data. They want that at their fingertips, at their desktop to be able to make decisions, billion-dollar decisions, life or death decisions on the fly. And so we've had some serious success with building out capabilities rapidly with the C3 platform in that area. Another area that's emerging, and it's extremely important, given some of the external conflicts that we're seeing in the world, is around -- let's talk about logistics. And there's a phrase in military circles that amateurs talk about tactics and strategy and professionals talk about logistics. And the circles that we operate within Pentagon and some of our joint customers, they're all about ensuring that their logistics, their supply chain pipelines, the inventory optimization, their predictive maintenance. That sort of thing is all encompassing within this contested logistics swim lane, and we're really excited about some of the discussions we've been having recently with -- in those areas. And again, I mentioned this before, but it's really been interesting to see the analogous use of some of the existing applications that C3 has put a lot of investment in around the commercial application of logistics or supply chain optimization that we can then very easily extend into the federal space. One very interesting use case is a case study we did for one combat and command organization. One of these customers asked us to look at how we could consolidate some of the myriad data feeds that would go into briefing the Commander about the health of their organization and their warfighting capabilities. So we looked at C3 to help us build a smart application to replace some of the existing static dashboards that were currently being used, again, whether it was PowerPoint or some type of data visualization capability. It was days old data, weeks old data that wasn't giving the command of what they wanted. And so the application that we built helped create additional risk scenarios. They wouldn't have been able to get through a PowerPoint presentation, for example. It used AI/ML capabilities to look at the vast amounts of data that we had access to and create alerting and other types of information that would be available to that commander about the readiness of their force, and the result of that was a very well-received pilot by the command. This activity, as we talked about before, was done in a 12-week execution, a pilot execution. As Tom mentioned, it was on spec, on budget and exactly what the customer was looking for. And the result of that for us, for our partnership was a continued extended contract to build this for the entire command in a longer-term engagement. All right. So looking ahead, what's in the future? A couple of things that I'm pretty excited about and leveraging the C3 platform for. Number one is leveraging the progress that we've made so far and some of the logistics organizations that we've been talking with. Merel mentioned the joint staff, J4, for example, in DLA. Getting inside the decision cycle when we're talking about contested logistics, being able to link planning and the actual delivery of capability is something that's a huge challenge across the enormity of the DoD enterprise. And that's something that our teams have been working on very closely together, and we're really excited about what we can do in those areas, and there's lots and lots of opportunity ahead for that. The other thing that's really interesting too is, again, I go back to this concept of reuse. If we can build a commanders inside dashboard for one combatant commander or one service, we can build it for the entire organization. Each organization needs a different view of their version of that business. The same thing goes with contested logistics. If I'm a joint combat and command and I care about this portion of the world around logistics, the Navy might be interested in some portion of it, the Army, et cetera, we -- there's probably going to be a presentation today about predictive maintenance. So if you build a predictive model around airframe predictive maintenance, there's probably not too far of a venture to go and look at it for surface ships or ground vehicles. And so we're really excited about how we can potentially reuse and create a flywheel with growth on the C3 AI platform, in addition to looking at civil and intel use cases, which are also becoming more interesting to us as we start to grow the awareness of the C3 platform across the Booz Allen enterprise. And then finally, as Tom mentioned, we've got a common customer in the DoD space who likes to talk about their vision for the Google of DoD, for the search for DoD. So the example here would be the Chairman of the Joint Chiefs is sitting down at his desk, and he wants to know how many tanks are service-ready to go fight in this next conflict or how many E5s do I have in INDOPACOM that know Mandarin. And today, that information, that like seeking that information out would take weeks and weeks of data calls across the entire department. Over the application of something like C3's generative AI capabilities, all the data that we've collected in these data platforms that we produce for our DoD customers can be leveraged to quickly answer those types of questions in seconds. So extremely exciting. The future is extremely bright for that particular capability. So in summary, we at Booz Allen are super excited about integrating C3 into our delivery tool kit for some of our federal customers. I look forward to working with the team going forward. So thanks very much. [Presentation]

Okay. Let me introduce my colleague, Nikhil Krishnan. Nikhil has worked with, I think, 11 years. He runs all data science. He came to us from McKinsey & Company, where he worked for a number of years. Before that, he was at the faculty in Colombia. He is a kind of distinguished expert in all things kind of related to data science. He was kind of personally has been driving this C3 Generative AI products. Nikhil is going to talk to you about that. I think you'll find this quite interesting. Nikhil?

Thank you, Tom, for the introduction, and I'm going to launch right into it. Let's see. So generative AI, we've been -- As Tom mentioned -- we've been working on this in the AI/ML world for a few years now. And then really last year, got into it in great earnest in terms of a broader generative AI product capability, and I'll talk more about this in the next few slides. But as we approached it last year, we put together the best of C3.ai. So over a decade of our experience in building very large-scale AI applications have gone into this generative AI product. And that includes our capabilities in managing big data, cloud compute, all of the purpose-built enterprise AI models, supervised learning, deep learning, unsupervised learning, NLP, reinforcement learning, all coming together with the generative AI and enterprise search capabilities. What does this mean? Let me double click on that. There's really 2 things when we talk about our generative AI product suite. The first is generative AI when it comes to an attachment of this product with C3 AI applications. By this, we mean a next-generation human computer interaction model. Imagine a situation where a human can ask any question in a natural language interface of our C3 AI application, and they would be able to get the top results, get the sources, get the summaries, be able to chat with the system. That's what we mean by generative AI attached to our existing apps. We have 42 of them but attached to, for example, reliability, supply network risk, et cetera, to transform the human computer interaction model. This is great because it's great for our customers because it drives additional value. It drives democratization of the applications. It gets a larger community of people to be able to use our applications and unlock value from them. And it's also great for us because it drives additional consumption for C3.ai. But the second opportunity that's even larger with generative AI and the second stand-alone capability is what we call it, is really an enterprise search capability. And this is the ability to map any data set in an organization, structured data and structured data, tabular data, sensor data to these generative AI product and be able to answer questions, be able to reason on it, be able to get the right information to the right person at the right time. And this is really the bigger part of what our generative AI story is, and this is what I'm going to be talking about and focusing on in this presentation. So Tom covered this, I want to just double-click on this. We have really been razor-focused on the application of generative AI to the enterprise in our product capability. What does this mean? When you look at the standard implementation of what others are doing when it comes to applying generative AI to the enterprise, they're often using an LLM either directly or using an LLM fine-tuned on top of the enterprise's data sets that could be text HTML code, and you have the LLM, these things directly feeding into the LLM and then you have a chat like interface for humans to ask questions and interact with that language model. The problems here are numerous, and Tom covered them in his talk, I'll double-click on them quickly. First, you have stochasticity, you have random responses. If I ask the same question twice or 2 people ask the same question, they might very well get different answers. And that's a function of the nature of using an LLM to reason and answer questions directly on top of the data. Secondly, traceability. So what is the source of a specific piece of -- what is the evidence that supports a specific statement or assertion by the LLM and how do I have transparency into that. That is absolutely not supported in this architecture. There's no enterprise access controls built into this. That is if the CEO asks a question and somebody in the front line or on the factory floor, ask the question, it's very hard to parse out what a person is actually allowed to see and only show them the information that they're actually able or allowed to see. There's also risk of information leakage. These LLMs are susceptible to prompt engineering attacks, and this is actually quite problematic and Honorio referred to this, Tom referred to this. Samsung case, for example, this is seriously problematic for the enterprise. And then lastly, you have this problem of hallucination. The LLM might just very well make up something in a certain situation. So in contrast to this, which is what a lot of companies are experimenting with and trying out and providers are providing in the market today. The way we approach generative AI was really building on a decade plus of our experience in applying these technologies to the largest and most complex enterprises in the world. So in our architecture, we wanted to from the ground up. So first point number 1 is you'll see on the green area up top on the funnel on the right-hand side, we have tables, we have applications, sensor data, log files. We wanted to consider data sets that are much broader beyond just the unstructured text type data. We also have the separation between the knowledge model in blue and the LLM itself. What does that mean? That means we're actually using separate deep learning knowledge models to embed the enterprise's information and storing that efficiently in a vector store. And then when a human asks a question through chat or through search to the LLM, the LLM in turn -- the language model in turn has to ask that question of the knowledge model. And the knowledge model -- between the knowledge model and the LLM, there is a barrier or a firewall where the access controls are applied. And so the LLM only receives the information that the user is actually authorized to see. And so we've separated the memory module in blue from the reasoning capabilities of the LLM. So the LLM is not answering your question directly. That LLM relying on its memory, the memory resides in that blue box. And then one more thing I will highlight here is you see have the chat interface. You have a search interface, but you also have an orchestration interface. And this is a key part of our product as we want to be able to orchestrate other applications. We want to be able to orchestrate purpose-built AI/ML models. We want to even be able to chain LLMs together and orchestrate other LLMs. By the way, in many cases, in our -- in that LLM bucket, you might have much smaller models and actually many more of those fine-tuned smaller models all working together in a much more efficient way than some of the large LLM implementations today. The benefits of this are numerous: We have deterministic responses. You asked the same question, you get the same response; full traceability, transparency, I'll show you this in some of the product demos; full enterprise access controls that we're able to leverage all of our experience in doing this; no LLM caused leakage of proprietary information because LLM doesn't know any proprietary information; and then lastly, no hallucination. We have the temperature of these models turn weighed out. So that's, in summary, what we've done, that's super unique. And I think this differentiates us very uniquely in the enterprise space, and we have this as a production GA product available to deploy today. This builds on a decade plus of our experience with the C3 AI platform. So this slide is one that Tom walked you through earlier today. This was all of the software engineering we've done over the last decade. All of our applications run on this platform, including the generative AI product. And the generative AI product, I want to highlight some of the capabilities in the platform that our generative AI product uses that are -- would take other software companies quite a lot of time to put together. So for example, run time management, all of the hardware profile, hardware management, GPU and CPU run time management and orchestration. All it takes -- our product takes -- generate AI product takes advantage of everything in the platform there. Vector store, we've actually extended the platform stores to include a vector store as part of this architecture. A distributed file system, so the vector store can spill over to a distributed file system to disk, if needed. We have deep learning retrieval models and large language models. And on the right-hand side, even a visualization component library to actually render results in the right chart format. All of that is part of our generative AI offering. So a summary of what we've done as a stand-alone generative AI product is it really serves as a knowledge co-pilot. And what I mean by that is think about a capability where you have the C3 Generative AI product, which serves us a deep domain assistant with expertise that can orchestrate APIs, tools and help users take action. I'll give you a few examples of what this means. So first is in manufacturing, and I'll come back to this example later in this talk. But imagine a situation, and this is a real example of a case we're doing, where you have a company that -- where you have machine operators and technicians that are fairly new and a retiring -- aging retiring workforce. And these operators are operating a machinery that's hundreds of millions of dollars in economic value. But they're struggling, they struggle to do that because the operation procedures are very complex. There's a lot of information, thousands of pages of standard operating procedures, maintenance manuals, operational manuals, recipe cards that they have to consult, tens of thousands of sensors that they have to look up. And if something goes wrong, they really don't know what to do. They end up calling an experienced tech saying, "Hey, can you help me out." And this is not sustainable. With the C3 AI product, this same machine operator tech, this novice operator can now talk directly to our C3 Generative AI product. And our product has already read the standard operating procedures, the maintenance manuals, the operational manuals is aware of all of the sensor data, for example, in OSI PI is aware of all of the recipe cards and may even be aware of purpose-built AI/ML models that have already been configured. This is really the future state where the operator can ask any question and get an answer back. Here's another example. This is an ESG domain, sustainability. Companies have data contained in dashboards in purpose-built applications like Watershed or Sphera or C3 AI. They have spreadsheets, they have slide presentations, sustainability reports, news seeds, all sorts of their investor data. But at the same time, they're not -- and this data is all fragmented throughout the organization. But at the same time, the challenge that the organization has is that -- most organizations have said they can't answer even the basic questions. How am I doing against my CO2 reduction plans. What are my best performing ESG issues or topics. So what are my worst performing ESG issues or topics or what are my worst performing plans or which projects are -- how am I doing against my budgets? And here's an example where an operator could -- with the C3 AI -- Generative AI product, you can ask any question of the system. So for example, how am I doing against my CO2 reduction goals that could be a question that I could ask. And I just wanted to walk you through an example set of visualizations of what this could look like. There's my question again, I've just asked it in a very simple Google like interface. Our system has gone back, the LLM interprets the question, the question is then sent to the retrieval models. The right embeddings are retrieved based on access controls. And if there's a right match found, our system will even visualize this in -- a product will even visualize this in the right visualization. In this case, it's the wedge chart that shows you how I'm tracking to my goals. It also summarizes it in text. So here it says, what does my plan look like? What do my goals look like? And I can actually go directly to a specific application page or a document or I can further chat with the system. And then I also have supporting data. What are the relevant search results that -- from the sources across the enterprise from C3 AI as well as from other sources or in this case, Tableau or Microsoft in a rank list of order of relevance to the end user. So that's the second example. We have filed this patent quite broadly of the capability that we've developed here, which includes the orchestration of all of the components that I talked about before. So summary, we have deterministic responses, full traceability, full enterprise access controls, no LLM cause leakage of proprietary information and most importantly, also no hallucination. So what are a couple of examples. We've been doing work in multiple domains. One of them is in banking or financial services. This is an example of a customer of ours that has very complex documents -- financial services documents. These are -- this is an example of a bond document that's 80, 100 pages long. And it has introductions, has appendices, has tax treatment, has Board of Directors, and it takes an operator 5, 6 hours to go through a single document. They have hundreds of thousands of these documents and thousands more that arrive each month. So in this kind of a use case, our system is super powerful. For example, I can ask in any question, see if it plays. What is this bond, for example, I can ask the question. and I can get an answer -- a summary answer. In this case -- of this 84-page document is summarized directly for me as an end user. And I can further chat with it. What is the [indiscernible] of this bond. I can expand that chat window so it takes the full page, it's going to answer that. What is the maturity date of this bond. It's actually gone through all the tables and all the details of the document to give me that how much is being borrowed, $5.4 million in this case. What is the interest rate, how is it paid, et cetera. So you get the idea. Some of these are very detailed type questions that would take an operator quite a lot of time or a user quite a lot of time to find and in this case, it can go through not only the details of the document, but also all the tables, diagrams, et cetera, and extract the right information for me. Another example is documentation. So here, we've applied C3 Generative AI on our own documentation as part of our version 8 product. And we're making this available to all of our customers. Our documentation is fairly complex. There's formal documentation that's written by our documentation specialists. There's type level documentation for each C3 object or entity that's part of our release that's often written by developers, there's tutorials that we write and there's many, many of these that are written with each release. There's in-depth guides, there's a full rich developer community. Think about this is our equivalent to stack overflow, it's called community.C3.ai. And then there's interactive videos, video courseware, 101 courses, 201 courses, et cetera, that are often updated as part of each individual release that encapsulate all of the features that we have in that release. And this is fairly complex because the platform is very rich with a lot of capabilities. And so this is a game changer for us internally as well as our partners and our customers. So I can ask it a question. For example, what HPO techniques can I use, which is a fairly difficult question, and I'm able to get an answer directly by HPO I mean hyperparameter optimization. This is a question that a data scientist might seek to answer. And I can further chat with that. And I can even get code snippets, in this case, you're seeing an example of code snippets for grid search that I've asked it for. So it's a really powerful capability. Product road map. So switching gears to where we are and where we're going with this. We're on a quarterly release cadence with the generative AI product. Our next release is in July, and I want to highlight a few things. So first, we're going to be releasing 3 new fine-tuned LLMs for specific tasks. One is for question and answering. So an LLM that basically is used to orchestrate the question-answering tasks as part of the chat or search capability. Second is for summarization to summarize very long documents and data sets. And third is for comparison, to compare things to baseline. We're also excited to be releasing a true generative data visualization capabilities. So think about this as seeing a data set into an LLM that's fine tuned to see a data set, interpret a data set and then generate the right chart format on the fly. Improved administration flows and then multi-LLM chaining and orchestration will be part of our July release. Highlights in October. So we're going to be releasing fine-tuned LLMs for manufacturing and oil and gas. And then a second big area of focus for us, is we have -- we know how to go from about order of 6 customers for generative AI to order of 60. We know how to do that very well. But we really want to go from 6 to 60 to 6,000 customers with this product. And we think that this product can actually be self-serviced where people can launch it from a cloud marketplace from one of our cloud partner providers and basically set up their users, set up their data sets, set up their configurations and be off to the races themselves. So we're super excited about self-service deployment. That will be a big theme of our October release is making it so that people can actually launch and set this up themselves and accelerate their time to value and accelerate for us the time to consumption. We're also going to be releasing in the October release, a fine-tuned LLM that is for code generation for the C3 AI type system and DSL. And this will also be game changing for all of our partners. Booz Allen, for example, as well as for all of our customers that develop on top of C3 AI because it's going to make it much easier for them to actually write their own applications or extend and modify our applications. And then the last thing I'll highlight in October is developer tools for third parties. This will make it easier for people to use our generative AI capabilities. In January, we're planning 3 more LLM -- fine-tuned LLMs to be released, 1 for defense, 1 for banking and 1 for CPG retail. And then we're also thinking about very specific, planning specific capabilities for C3 AI applications. So think e-mail memo generation as part of C3.ai CRM, think about report generation as part of C3.ai ESG. The summarization of AI evidence packages for the smaller models we have in all of our applications, being able to summarize that for a human readable interface. That will all be in January release as well as enhanced multilingual support for organizations that have multiple languages. And in April, we're thinking through fine-tuned LLMs for data generation, and fine-tuned LLM so DevSecOps. Now in terms of application support in April, we're also thinking about additional capabilities for our supply chain suite, for coaching and recommendations as part of C3.ai CRM as well as our law enforcement application. And then lastly, multimedia support; images, videos, audio, et cetera. So I wanted to cover a few examples or a little bit more detail on our pilots. We had talked last quarter about 3 specific pilots, generative AI pilots, one with a large U.S. manufacturer, second with the U.S. Federal Government Agency. This is on starting with technical papers and then going on to other technical classified data sets. And then third, with a large U.S. refiner. And this is looking at price and volume data. This is looking at market data related to crude and commodity prices. Double-clicking on the first one. What is a pilot generative AI pilot look like? Typically, for us, this looks like less than 12 weeks in duration. There might be a design and analysis phase where we're looking at what data sets to really bring in, how to bring that in with the customer, working with them through that. There might be structured data, unstructured data, sensor data that may write on a map and bring in to our object models. All of that is part of that design and analysis phase. And then we're configuring the application and then bootstrapping the training and improving the performance of the end-to-end system. That's what the rest of the time looks like. But these typically go quite fast and they're typically 3 months in duration versus our standard pilots, which are up to 6 months. In this case, in the manufacturing case, we're bringing in standard operating procedures, which are very complex documents, maintenance manuals, recipe cards and then data from OSI PI sensors. There's tens of thousands of sensors per machine here that are mapped into our system and then the generative AI user interface that I showed you up front. Let's double-click into that. So here, I really want to showcase the support for operators, for novice operators to make them equivalent to experienced operators. I can ask, as an operator, any question. This example is showing me asking a question that's fairly complex, which is a procedure for Annubar cleaning. Now this is something that I would have to otherwise find in a specific procedure or maintenance manual. And it's very difficult for the operator to know how they've done it -- how to do it. The AI summary immediately shows me the steps that I need to go through. And not only does it show me the steps, but actually links to the sources. Please join us in our demo station after, and we can also give you a live demo of a similar system. That's a synthesized AI response across all relevant data with the sources and there's no hallucination in any of this. The second thing that you'll see here is a ranked list of relevant search results from other documents. So you can see some of them are documents specific to vacuum systems. Some of them are bearing manuals, et cetera that are in ranked order of relevance. And then I can also further chat with this system in this example. It's basically allowing me to ask follow-on questions and it's context aware. So each question that I'm -- that the system is asking -- answering based on what I'm asking it, it's actually aware of the original question that I asked. So for example, I can ask it what is the frequency of cleaning or what are the tools needed. And each time it gives me a response and then also links me to the source documents. So interactive chat with context and follow-up. So this is hopefully giving you a sense of the power of the system. There's AI summary, all of the sources, a ranked list of relevant search results and then an interactive chat in context. And the benefit of this really is this current process that is incredibly inefficient and incredibly difficult for these novice operators to operate these very expensive machines is going to completely change where these machine operators or text can interact directly with our system with the C3 Generative AI product and access the right information at the right time. With that, I think my last slide is how -- a summary of how we are unique. And the standard LLM package approach that we see from many others in the market, you have all of the data directly going into an LLM, the LLM might be fine-tuned on those data sets and have problem with random responses, no traceability, no access controls, risk of leakage, and prone to hallucination. And on the right-hand side, with our architecture, you have all data, not just unstructured data. Everything embedded and stored in a vector store and then an LLM that's capable of chat, search as well as orchestration, deterministic responses, full traceability, full enterprise access controls, no LLM cost -- leakage of proprietary information and no hallucination. So with that, I'd like to conclude this section, and I'll hand over to my colleague, Binu Mathew. Binu is responsible for product and engineering at C3.ai and formerly from GE and Baker Hughes.

Thank you, Nikhil. My name is Binu Mathew. I run products and engineering at C3.ai. And I came to C3.ai from Baker Hughes, where I ran digital products. And then prior to that, I spent 16 years at PeopleSoft and Oracle. A major reason for the Baker Hughes joint venture and for me to come to C3.ai 4 years ago, is that C3.ai is unique in its focus on enterprise business value. We're the only company that does it in this way. And this is in sharp contrast to what you see out there even today. A focus on tools, a focus on technology and effectively science experiments that do not scale because customers are really looking to collect data, to apply tools, work on individual models. And while you might actually get some very interesting results, there are very good tools out there. You don't get enterprise AI applications. You don't get applications at scale. You don't get applications that are tuned to business value. Now our focus is very different. We start from the value chain. So if you're looking at a large enterprise and you look at what are the business processes involved? How do you go through from customers to supply chain to manufacturing? How did the business processes work? How does AI apply to these use cases. Once you have that, then you can think about how an AI application should be built, what models you should focus on. You need that context and you need the expertise and you need the technology to do it in that way or you do not have scale. And again, we're the only ones who do this. To do that properly, you need enterprise AI applications. You need applications that collect data at scale that can be deployed easily, that have user interfaces and can actually be adopted. And we have a very comprehensive suite of these. We've been doing this for well over a decade. So we started in 2009. Our initial focus was in the energy management space and then utilities, financial protection. We expanded then into predictive maintenance. We're really the masters of predictive maintenance now. And we did that in many different cases. We expanded into financial services. And over the last few years, we've now built these up to 42 out of the box turnkey applications. You can deploy these applications, you will get value. You will get these. We can do these in pilots. We will have a production application running for you in less than 6 months. Now we've started aggregating all of these applications into suites. We now have enough applications where we're really taking a sweet focus to the whole thing. So if you take supply chain as an example. We have applications like inventory optimization, supply network risk, production schedule optimization, sourcing, demand forecasting. And all of these applications work together. We have deployed these at some of the largest customers in the world. And part of where we're truly unique besides our overall focus on enterprise business value, we are also the masters of data fusion in our ability to bring data from many different sources. To be clear, we're not in there replacing your systems of record. There are good applications to do that. We complement you. But we can aggregate your data from all of these sources. If your underlying sources have performed enough, we don't have to copy the data over. We can actually access it live from those systems. And we integrate that into one unified data image. And what that means is that you can get in representation of your entire enterprise from one place. So think about this from a supply chain perspective. What that means is that I can look at a part that I've sourced from a supplier. I can move that part through the sourcing process. I can move it into warehouses. I can track how it ships through logistics. I can get into my manufacturing plant, I'm making -- I'm transforming that part into something else into a product, and I'm tracking how that product then gets out to our customers. We can replay that all the way from the beginning to the end. And when you can do that, then you can also learn and you can predict. And this is where the AI portion comes into play. Our applications are going to -- you heard Nikhil talk about generative AI. We already have very powerful data fusion and predictive AI capabilities. But generative AI is really going to make a difference in terms of how customers interact with these applications. We have very, very good looking UIs. We spent a great deal of time on our UX process. But generative AI, if you can actually start asking questions. For instance, what is my on-time delivery performance, which suppliers are late in deliveries. That brings a whole level of interactivity and the whole level of adoption that we can drive with this. You can expand, including the generative AI process into much more elaborate what if scenario planning. And you can continue expanding the ability to scale and deploy these applications. And this road map, by the way, is illustrative of all of our other applications. Our reliability suite built up with all of our experience from predictive maintenance in various industries and various companies. And this is an area where it's an example, I think, of where all of our other applications will go. At this point, our reliability application has been deployed at very large facilities at scale. And that means you're talking about thousands of pieces of equipment. You're talking about billions of rows of data a day and trillions of rows of data that are being analyzed in aggregate. And that means that we can get enormous amount of information by learning patterns of how equipment really behave. Now to contrast this with what traditional applications do. Traditional applications, and again, there are some good ones out there, but they are based on rules, they're based on how machines were designed, they're based on physics, which again, is very good from a design perspective. But if you truly want to understand whether a machine is going to fail or not, you need to learn how that machine, that specific machine is behaving. And once you have that, then you can get a great deal of insight from that. And what that means is that the amount of noise that you get. If you're dealing with a large manufacturing facility, you're inundated with thousands of alerts from dozens of different systems. We can pull that all down into one area. We can give you actual insights in terms of what's likely to happen. And that means that we can increase the availability of your plant by a couple of basis points, and that ends up becoming a great deal of value. Again, deployed in some of the largest customers in the world. And you'll see a pattern here. We can integrate data from multiple different data sources, some traditional, some nontraditional and pull it all together. We do the same thing with sustainability. And some very interesting things that are happening in sustainability. Sustainability is an area, by the way, where C3.ai really got a start. We started working on energy management in 2009 -- energy and emissions management. And with the current focus on climate technology, ESG, materiality, stakeholder management, our capabilities in generative AI, we can actually provide comprehensive analysis of data sources that are relevant. This is not us coming out with random invention factors. We can actually analyze the data, we can look at what's relevant from external -- in terms of what's relevant to external stakeholders, and we can give you updated materiality assessments. And our energy management solution is integrated with our other tools. So for example, since many of our customers have implemented our reliability solution, the same data elements, the same sensor data works immediately for sustainability. If you have Scope 3 emissions that are coming in from a supplier, that integrates directly into our energy management solution. Now we're going to take this much further. We're working on AI decarbonization planning. Going after supply chain emissions, as I mentioned. We can give you far greater insight into that than you can today. And we're going to expand that with generative AI capabilities. ESG reporting is a major burden for customers today. We can help you generate those reports. Our CRM solution. Again, you'll see the same sort of pattern here. A couple of things that I want to point out with CRM. Most CRM data or most of the CRM solutions today rely on what a salesperson is entering into a CRM system. And what that means is the salesperson is going to enter just enough information to get their manager off their back. You actually want to see what's going on. What's going on with e-mail interactions, what's happening in the broader market? If you can pull all of those data sources together, unified image, you can learn from the behavior then you have the ability to optimize, predict what your revenue is going to be with a much greater degree of precision. You can intervene earlier. And so our ability to help drive true sales impact through sales growth is much higher. Defense and Intelligence. You heard Graham talk about what we're doing with the federal space in Booz Allen. Again, this is an area where we've been established for quite a long time. We have applications like Readiness. This is -- you heard Tom state that this is the system of record for the United States Air Force. We're active in the intelligence community and in terms of tools -- the smart tools that we're providing to very senior personnel, and again, deployed at a wide number of agencies in the U.S. government. State and Local. We have an application that is proving to be quite successful in terms of law enforcement, again, being able to aggregate data from multiple systems, make it easily available to drive a criminal investigation for us to be able to do data-driven property appraisals at scale, at speed. Now all of these applications, the reason we can do so many of these applications. Well, we have domain expertise. We have a great data science team. You heard with Nikhil's team. But the biggest reason we can do this is because we have a platform where 95% of the code is the same, application over application over application. So when we're coming up with a new application, it's really a declarative process. We need to know what the data elements are. We need a certain amount of domain expertise to model that. We can build models. But that's less than 5% of the code. The platform provides the rest. The platform allows us to deploy on AWS, on GCP, on Azure on-premise. Same code. The platform allows us to scale to trillions of rows of data. The platform uses the same technology that we do for data fusion to integrate application platform services. And because we can integrate platform services in that way, it is essentially future-proofed. And so we're constantly moving the technological frontier forward with our platform. That platform has been given the top ratings by multiple analysts, the Forrester Wave, Constellation. And so our products -- our focus right now is on making our products as easy to adopt and deploy as quickly as possible. So we have a very simple pricing model now. Our pilots, you pay $0.5 million. In less than 6 months, we will have that pilot deployed for you. We will have that pilot not just deployed, we will have it running at in production grade. And that means that at the end of the pilot, you will have had something that has proven its value. You will have it live. And if you like it, you can just keep running. And again, doing that in the space of 6 months for an application of this class is not something you'll find anywhere else. Once you get into the production phase, I know that most of you are familiar with the consumption-based model, you pay for it by VCPRs. Now we continue to expand our talent base because to build and drive all of these applications, we have a very strong R&D team. One of the things that's unique about C3 is our culture. We have everybody in the office. And because of that, we have a very collaborative atmosphere. We're able to drive, I think, a great deal of productivity by being together. And we continue to drive that through our platform technology and our application technology. In addition, we're building up new development centers. We have a development center that's coming up in Guadalajara, Mexico. It is very close to us from a time zone perspective. We have a great relationship with the government there. And so you'll see us increase our application capacity by a significant level as that team comes online. So to conclude, we have a unique focus on enterprise business value. We have 42 out-of-the-box applications. We have a platform that is $1 billion-plus investment that allows us to scale, accelerate, stay up to date in terms of technology, and we have a unique talent base. So with that, thank you all.

We have -- there's some people with microphones. And we'll be very pleased to field any questions that you may have. Alex, do you have a question? No? Come on. I know you got one.

In a bit, coming in online.

Got online? Okay. I know Alex has one, so let's get -- let's start with Alex. [indiscernible] It's a hardware problem. It always comes down to that.

Thanks, Tom, like the old days. So I have maybe a couple of questions from general to more specific. There's a point of view out there that -- and many comments about AI and...

Can you hold the mic maybe a little close to your mouth please. We're [indiscernible]

Many comments about AI, by investor community as well as companies we follow, some comments focus on productivity, meaning head count reductions and other comments are on productivity in terms of velocity of development. What do you think takes precedence now over the next year, head count reductions or productive in terms of development in how quickly can applications be put out? And second question has to do with -- you talk about pipeline doubling and sales cycles decreasing by 20% or 30%, that tends to produce decent financial growth. What should we be looking at as investors to see this kind of productivity improvement in your own model as your pipeline doubles and sales cycles decrease?

[indiscernible] backwards the pipeline has increased by 100% year-over-year since the end of the fourth quarter in terms of qualified opportunities in the next -- in the following 12 months. Our sales cycles have decreased from, I think about 13 months to 4 months, okay, in the last few years. I think the shortest sales cycle we had was the one that Ed had mentioned that just closed yesterday. And that was 4 days, so that's getting in the ballpark. So we're -- so we clearly see a lot of interest dramatically increase in AI. And with -- aided by the consumption base, a pricing model, which clearly in the short term would put downward pressure on revenue growth but the medium and long term puts -- is a revenue growth accelerant. We're pretty optimistic, Alex. As it relates to productivity, I have not been involved in any discussion where people were interested in productivity as a means to reduce headcount. They just want increase in -- increased productivity to get even more product out of their factory, to be able to deliver products on time, to the right place at the right time so that people don't starve in North Africa. The United States Air Force is getting -- looking to get 25% more effectiveness out of their 5,000 aircraft, so they're not laying off pilots. They're just getting 25% more airplanes in the air on any given day. Next, other question?

Tom, just to follow up on Alex's point that, Generative AI burns a lot of computer cycles. And last I heard, you switched to a consumption model. So I would presume that should be very positive switch in pricing mechanism for you.

Ed, do you want to comment?

Yes, I think that as I kind of alluded to, the first step is just to get a pilot in place and get -- to take the example that I gave, which is reliability in one facility and then basically scale up across all facilities in an organization. Consumption goes up. And then as you heard Honorio talk about, then there's another application that might gets deployed, it might be production schedule optimization and you roll that across all the facilities and there's more consumption. So the answer is yes but the consumption is commensurate with business value that the customer is getting. So if the customer is getting value, they'll use more of it and the consumption will go up and we get paid.

Paul, you have some questions online or on [indiscernible]?

Tom, first question, here is, any comment on the federal business. We heard from Booz Allen, how has deep thinking in the market evolved?

Ed?

There is significant interest in our products in the federal market. And I would say that not unlike other large organizations, there were tens or hundreds of individual projects that were going on. But models are being developed but there was an inability to kind of scale them up across these very large organizations, whether it's in defense or whether it's in intelligence. And so there's recognition now and you heard it from Graham and Booz Allen that what you need is an AI platform to not only scale these models in 1 domain but also reuse the capabilities in multiple domains. And so whether it's contested logistics, if you do it in 1 service, you can take that same capability and apply it across multiple services and scale up very quickly. And so I think that's really the essence of the demand in federal and defense, intelligence and civilian over time.

You can imagine that if -- we were recently selected as the AI program -- as the program of record for predictive maintenance for the United States Air Force. And as a comment, I believe, we're the only AI system to be -- this is the first AI system to be selected -- to be designated as a program of record in any portion of DoD, so this is a real milestone. But these people are looking at trying to save billions of dollars a year. And those contracts are not really in place yet, so there's some opportunity there going forward for growth.

You have one more question online. Shall I keep going?

Go.

Another one here is, do you expect to land more deals in Generative AI use cases because it is easier to get up and running?

Well, Ed, why don't you comment on the pilots for last quarter? But I think actually a minority of them were Generative AI.

Yes. That's right. So I think we're seeing broad interest in all the applications, I'd say that's fair to say and Generative AI as a companion interface, if you will, for those applications. So I'll go longer here, which is, I talked about reliability in a facility and basically scaling that up across all facilities. Now that application, you have to train an end user on and there's a specific maintenance operator that basically is using it. You can complement that with C3 Generative AI in the reliability realm and you make the information now available to anybody in the plant, the Plant Manager, the Head of Manufacturing, the Head of Operations, basically everybody becomes a user of that information. And they can ask questions and get us -- get an answer without having to be trained on an application, without having to log into an application. So we see Generative as very complementary to the applications and just a broader net for AI.

Sort of accelerate for all of our business.

Correct. That's right.

Another one here. You have clear leadership in the oil and gas market, how's C3.ai helping other verticals like healthcare?

Who wants to field that? Who's up here? Is Houman up here?

So I think, yes, we do have very strong leadership in oil and gas. But I would argue that if you look at some of the leading providers across manufacturer -- manufacturing, financial services, defense, intelligence, others, we're also getting leadership across those customer bases as well. In healthcare, we're getting -- we are starting in life sciences, where we're doing a lot of work, applying the applications that we have, whether it's in reliability or supply chain into that industry and I think that's where we'll be able to scale from there as well. And our goal will be similar to oil and gas, right, get the leaders within that industry, show economic benefit, move to the next industry and scale it. And I think that's what we're doing.

For whatever reason, well, I know the reason but it's clear that the utilities were the first -- the utility industry was the first industry to adopt enterprise AI. We can talk about that some other time if you want. Okay. The second industry that really was just kind of blown open to us was oil and gas and that was primarily due to our partnership with Baker Hughes. And so that's been a hugely successful partnership that we are closing a lot of business around the world. I think in the long run, you're going to see the -- the enterprise AI is clearly going to be adopted by all industry segments, federal government, defense, intel, state, local governments, chemicals, pharmaceuticals, consumer packaged goods, travel, transportation, pharma, you name it, okay? And the -- so what the industry looks like in 5 years, the industry opportunity looks about like the distribution of GDP across those market segments. There is no segment that is not significantly altered and facilitated by the application of AI.

Thank you for your presentations. Just wondering how you would evaluate your competition, not a lot mentioned about that. It doesn't seem to be a whole lot. I would think that a lot of the SAPs and Oracles and even the Accentures of the world would have that on the front of their dashboard. And I'm not sure why they're not already delivering products -- performing products and services like you. And what do they need to do to catch up?

Well, pretty good question, [ Graham ]. SAP has a competitive product to us in any segment in which we operate. We are [ unaware ] of it, okay? If Oracle -- and we are pretty familiar with what they do. If Oracle has a competitive product to us in any segment in which we operate, we're unaware of it. We understand that they have a plan for a plan for a Generative AI product with some third-party partner. So if and when that happens, they might be a competitor in Generative AI. And what was the third company, Accenture? Accenture is a professional services firm. The way they make money, honestly, is to make these projects as complex as possible, okay? And as long and as hard, with as many hours as possible. Now we have this technology [indiscernible] but this is going to be the antithesis of Accenture. And that's why these guys are kicking it and -- that guy in the defense and intelligence community. I didn't take any shots at Booz Allen. I could do that fast. And these guys are the past. Now, but to our knowledge, I mean, tell me when Ed and I, really invented the CRM market, sometime in the last century. I mean what took SAP 10 years and Oracle had 1,000 people on it for 10 years, 1,000 people, okay, on building a CRM product and 10 years later, you had less than 1% market share. Same thing for SAP. As CRM guys [indiscernible] simpler than this, I mean it's trivial -- this is [ nontrivial ]. Now a decade later, you could have less than 1% market share is baffling to me. But sometime these big companies, you kind of -- you have these innovators [indiscernible] them a problem, where they kind of missed the next big thing.

Tom, what about Palantir in the defense area?

Why doesn't somebody who will handle that professionally? Ed?

Where do I start? I think Palantir, again, does a fraction of what we do. And Tom showed the tech stack. They do the data fusion piece of the equation and they have much more of a services model. And so they typically -- the pilot projects that we deploy are 3 full-time equivalents and we get an application up and running in production with users in a period of 3 to 6 months. That's -- there is no other solution that's out there that gets that to [indiscernible]

Again, in a case like [ BP ], Palantir will send 100 people in for 10 years, okay? And that's how they get paid, 100 people for 10 years. And one of the other difference in our business models is that when we're done with the deployment, the customer doesn't find that the software company owns their data, okay? That is kind of -- it causes some friction in the market. And with our customers, the customer owns the data. With our customer, the customer owns all the derivative works. And so Palantir's business model is a little bit different. I think they are -- look to me to be -- they have pretty good data fusion capability. They clearly have a lot of professional services capability and they kind of are able to monetize the fact that they own the customers' data that the customer finds out a few years later to its regret. I did that pretty well, right? I didn't come out of my shoes.

Yes, you did. That was very professional.

Two questions. I wanted to kind of reposition the competitive question and think about it differently in the sense of, in a world...

A little closer to your mouth, please. Thanks.

I wanted to reposition the competitive question and in a world where we are starved of GPUs and if Facebook can't get GPUs, I would imagine enterprises are struggling as well and will for some time. And so how do you guys think about your competitive positioning versus NVIDIA themselves in their DGX offering, right? I imagine many enterprises would contemplate going directly to them and rent out their compute power in order to execute against their LLM design .

Was it -- we're really having trouble hearing. Was it, a competitive position versus NVIDIA? Oh, NVIDIA, is a partner. Okay. Jensen is a partner. And -- you can go ahead.

Yes. I mean it's a different level of the stack. So we're focused on the applications and then we take advantage of NVIDIA technology. I think that was the essence of your question.

Well, no, DGX, NVIDIA actually has their own [indiscernible] LLM libraries, like you literally rent them, like in libraries. So I am just trying to understand how you fit within that.

We're agnostic of the actual libraries. So if a customer is using NVIDIA's LLMs, we'll take advantage of those. If the customer is using HPE's recent announcement with their LLMs, we'll take advantage of those, Bard [indiscernible] or any LLMs that come out basically are pluggable into our model architecture. And so these are all partners rather than competitors to us.

The second question I had was, as we moved from more of a training orientation to this phase of AI to more inference, how do you guys foresee your position as we make that handoff in the future?

Well, we've been doing inference for a decade now. All of the work that we've done in deploying AI applications or training and inference and so that -- we've been doing it for a full decade. I don't know if there's more to that question.

Yes. Is it more additive or not?

Yes. We all -- we know whether training or doing inference, it's just CPU cycles to us, okay? And so they're doing -- whether they're training the model or running the model, it's a CPU cycle and it's just -- and we get paid for CPU cycle.

Is the computational power for training significantly large than the inferences are?

That's a fact.

If I can just...

Yes, please.

So we've always had a blend of training and inference in our consumption model as well as in our base business case. So we charge for, as we just said CPU hour for -- around $0.55 a GPU hour, depending on the workloads. And all of our applications require both training and inference on an ongoing basis, for both the small models. And we expect this to also continue with the -- as we scale the use of the large or the smaller language models in the future. And so I don't see really any change in the fundamental architecture of our compute profile, whether on CPU or a GPU, just -- we just see a significant growth .

Thank you, C3 management team for doing this presentations. It was really helpful. So my question is, again, going back to competitive nature. So Merel talked about alliances, right? It's grown -- hyper growth, right? Was wondering, like from a hyperscale perspective, right, eventually, like the large one like Microsoft, right, they have a partnership with OpenAI, right? Can you comment on like this symbiotic relationship where like you guys are friends at some points when you go to market. But at the same time, they could build their own AI models as well, right? How do you balance that?

We're going to use their AI models in our architecture and they do, so their's -- an AI model is not competitive with what we do. An AI model is complementary with what we do.

How about OpenAI?

I'm sorry?

How about OpenAI with Microsoft?

OpenAI, I mean -- it's complementary. In other words, OpenAI, whatever large language we can use, Microsoft's LLM we can use. Nikhil, can you take it?

So we can use the PaLM LLMs. We have our own fine-tuned models. We can use OpenAI's LLMs. Keep in mind that the LLM is not the application. The LLM is 1 part of an application. Even for our Generative AI product, the LLM is just a small part of that application. There's so much additional capability, there's so much other technical work in that application, the vector store, the knowledge embeddings, the ACLs, the encryption, the interaction with database systems, sensor systems. All of that is part of our product, even in Generative AI. So we're actually LLM agnostic. We actually think there's going to be a proliferation of LLMs. We actually try to track it very closely. There's a Cambrian explosion of LLMs that will occur in the next year.

Can I just follow up 1 quickly. Like how do you make sure, I guess, like when you go to market with Microsoft, right, instead of them, the client using OpenAI stuff, that you could able to steer to C3.ai?

Okay. So at the same point, right? If a client wants to use OpenAI through let's say same Microsoft, we can plug that into our Generative AI product. They cannot -- it does not make sense, given an enterprise context, to use the LLM directly on their data. If they do that, they incur great risk, hallucination, stochasticity or random responses, risk of LLM exfiltration, proprietary information exfiltration , all of the things that I mentioned in my -- in the talk. But we can actually take advantage of an LLM like OpenAI's GPT-3.5 or GPT-4, as part of our stack, if a customer wants to do that.

And we are completely agnostic. I mean, whatever they want to do, it's fine. But look at this from the hyperscalers perspective, okay. We're an application running on that, okay and that cloud infrastructure. And whatever you're running, an application, doing for example, all predictive maintenance for the United States Air Force, let me help you out, that's a big application using -- or missile defense agency, okay, or Shell, which would be larger than the Department of the Air Force. Okay. So we're using massive amounts of GPU hours, CPU hours, okay and storage. And those hyperscalers want that workload on their platform and that's why they want to partner with us to go close that deal.

Because you have the industry knowledge?

Because we're burning. We're raising the temperature of the planet. Okay. Literally, okay, with the amount of capacity that we're utilizing for this application.

We have applications that are rapidly deployable and that consumes cloud very quickly.

And they want it. Microsoft wants in on Azure. Google wants it on GCP and Azure wants it -- and AWS wants [indiscernible], HP wants it on their cloud now, that's why they want to partner with us. They want that workload on their cloud. Graham?

Sorry, Graham Tanaka, Tanaka Capital. Sorry, we're going to give you a little action here. Just wondering if you could share with us your longer-term target profitability model. We've talked a lot about the AI model but your financial model -- and it appears breakeven -- you're hoping for breakeven by the end of next year, I believe. Is that around $80 million of revenue per quarter, it looks like. Maybe you can share that, plus really your target percent gross margin, operating margin, et cetera, at what revenue levels?

All right. Thanks, Graham. So we shared in our last earnings call, is the mic on, just confirming? Okay. Perfect. So in our last earnings call, we shared the plan to profitability, so Q4 of this year, so FY '24, we're now in Q1, we plan to be operating profit -- profitable cash flow, profitable on a consistent basis there on out and with the long-term goal of generating 20% operating margin in the long term, medium to long term.

I can take a question online here. There's a question, from an FTE perspective, is there any limitation on how fast you can scale pilots? In other words, do you have enough FTEs to support continued pilot expansion from here?

Who wants to comment? Ed?

I can take that.

Sure. Alex, that's your field.

Thank you. So as we moved into this pilot followed by consumption model, we've been hiring as needed, growing the number of FTEs we need to manage the number of concurrent pilots at any given point in time, plus the ongoing support services that we provide to customers based on what comes next after that pilot, right? We talked -- Ed talked about scaling that first application out across an organization, scaling to adjacent use cases, using additional C3 applications, right? And so we're hiring as fast as and only as fast as we need to, to support that number of concurrent pilots, which continues to grow over time.

But the business -- the forecast that we have given for this year, we have sufficient headcount to meet the pilots and to meet the revenue growth that Juho has talked about. And to be non-GAAP profitable and to be cash positive in the fourth quarter of this fiscal year and beyond.

Time for more?

Simultaneously, with, of course, hiring our own FTEs, we are also enabling our partners to be able to staff pilots and projects that we're doing with customers, so we have a formal training program for our partners in place. Many of them are going through it already. And I think Graham alluded to it, I believe Booz Allen currently has 50 people trained, that is not directly for subcontracting to C3 but there are ways where we can very quickly staff up with partners like Booz Allen on many, many of our projects.

Thank you. What was your question in here?

Would you share some colors on international expansion? Because I assume that AI is spreading out like wildfire.

International expansion? Yes. We are building -- from day 1, we've been building a multinational organization. We have offices in Rome, Paris, London. Go ahead, where else we got?

Singapore, Bangalore, Munich. Yes.

Amsterdam.

So you can expect that in a steady -- Sydney, you can expect that in a steady state, we'll be -- we can expect to see overall 20% of our business in Asia, 30% to 40% of our business in EMEA, okay, 40% of our business in North America and likely 10% to 20% in the defense and intelligence community. And I know that -- I think that adds up to about 110% but [indiscernible] and adds up to 100% to 110%, maybe it'll be 110% instead of 100%, maybe, we'd like.

There's a question. Could you discuss more about the self-service Generative AI product coming in October? How does self-service look like? Nikhil?

So the concept there is, we have -- from the -- from our partner cloud marketplaces, so think Google Cloud, Amazon, Azure, being able to -- a customer or a prospect to be able to find the C3 Generative AI product and then actually launch it and configure it themselves, basically map to their data sets, set up their users, set up their permissions, set up their structures and [indiscernible], this is -- be able to unlock value seamlessly by themselves.

Okay. What do you consider a healthy level of investing in sales and marketing? How do you balance that with growth?

Juho, you want to field that?

Yes. So again, on the long run, when we think about the long-term expense portfolio, we are planning or would expect to invest about 10% of revenue to marketing, another 18% of revenue into sales. And obviously, during this time, when we're going after the market aggressively, we are investing in the sales teams a little bit higher in the shorter term and marketing, of course.

What are your most popular applications today? How do you see that evolving?

The -- our largest footprint today is in Reliability. I think we're -- in Reliability, I think it's almost noncompetitive out there. I think we absolutely have the -- I think we win a very high percentage of the competitive situations in there. Our second largest application today is in supply chain, supply chain optimization, supply network risk. The third largest segment, this doesn't have to be an application, our largest -- well, it's hard to tell, actually, what's our largest market segment is today. I think in the fourth quarter, our largest segment was defense, intelligence, if I'm not mistaken. But it's -- before this is over, I mean, AI is going to be involved in every business process. I mean this is not a small phenomenon. And so this will be involved in customer relations, customer service, all of demand forecasting, demand chain, supply chain, asset optimization, cash management, customer churn. I mean there's no aspect of business and government operations that is not touched by AI, as this kind of phenomenon explodes in the next decade. I think that if you look at the positioning of C3 in this space, I think if I'm not mistaken, we're roughly 1/3 of $1 billion business today. So at some point in time, we're larger than that. The -- this market is really big, okay? So the worst case is, I think, so if you look at the worst case for C3, okay? The worst case is that we're a relatively large and rapidly growing cash positive profitable business in a huge and rapidly expanding market, okay? How bad is that? That's not too bad. How many customers, how many companies will be successful in that market? I think quite a few. What's the best case? The best case is that in one or more or many of those segments, we turn out to be the market leader, in which case -- that best case would be pretty good. So I think the worst case is -- I think the difference here, now realizing whether we hit the worst-case scenario or the best-case scenario is not going to be to the competitive dynamics of the market, it's going to really the execution of this team. So it's basically us, okay and our colleagues around the world that we will gain -- the extent to which we seize the opportunity.

Tom, a follow-up that kind of came in as a derivative of that. As it relates to reliability, is there any material difference between verticals in that respect? And this is -- I think it's kind of inferred from the question, is Reliability applicable to all verticals? Or is it more focused in some than other? I think that's...

It's a great question. And it's the same application. So see whether we're doing -- it's actually the same whether we're doing it for the human heart, okay, or whether we're doing it for an F-35 joint strike fighter or whether we're doing it for a paper machine at Georgia Pacific or whether we're doing it for an offshore rig at Shell. And we use -- we do all of those -- no, we don't do the human heart today. So every one of those use cases, okay, we have in production, deployment today and it is absolutely the same application. Although it varies -- are from application to application, are the data sources, okay, the machine learning model will change, maybe a different machine learning model or certainly it'll be trained on a different set of data and the user interface expression will change somewhat from industry to industry. But it is the same application across all segments. Binu, go ahead.

Yes, I can just add to that. So as Tom said, data is the only change -- things that change are the data sources. We have what we call asset templates for different classes of equipment, which drive machine learning pipelines and this is how we can train models at scale, so those are really the only things that change. You might have a UI change or so for a specific type of equipment.

So we have a user interface for oil and gas. We have a user interface for manufacturing. We have a user interface for chemical and we have a user interface for aerospace that's used by the Air Force and the rest of it is the same application.

Thank you for the presentations. Just as Generative AI takes off, how do you think about gross margins because on the inferencing side or API fees to call on LLMs, are you going to charge a premium to maintain gross margins? Or how does that outlook change over time?

Well, the -- it's a very good question. And thank you for asking it because what's important to understand is, this is being run in the customer's cloud, so it has no impact. The fact that it is -- they might be using greater CPU or GPU capacity, it doesn't affect our margins at all, okay? I think it's a shorter sales cycle and it's a shorter implementation cycle, so it actually should be increased gross margin, okay. And as that becomes a increasing part of our product mix, it should increase gross margin. So thank you for asking that question because it might have led to an understanding -- misunderstanding out there.

Maybe 1 or 2 more, Tom, in no order. How do you think about the co-optation with Google and their ability to do enterprise AI apps with Vertex AI?

Merel?

The way we see it is that Vertex AI truly is complementary with the C3.ai applications. What we do in many of our products, with any of the Google services, including Vertex AIs, we mostly leverage them as data sources. So anything you've done in Vertex AI, you can leverage within the C3 application. Now of course, this sometimes leads to a little bit of confusion. But I think, as you know, as we've worked on partnership, we've really cleared up a lot of that confusion and there's a lot of enthusiasm within the Google sales force to sell these C3 applications because they see that it generates Vertex AI consumption.

Another one here. Can you explain how you compensate and resolve hallucination problem with AI models?

Nikhil?

Yes. So the summary is, it's related to the architecture. We're not asking the LLM to answer questions directly. We have LLMs basically interacting with retrieval models. Retrieval models are providing the relevant embeddings to the LLM and then the LLMs summarizes that in our Generative AI stack and we have the temperature of the LLM turned way down. Combination of these things and are fine tuning -- basically encourages the LLM to -- if it doesn't have enough information to just say, I do not know. So we really don't want the LLM to make up anything, we want the LLM to reason on available information, summarize and answer based on available information. If not enough information is provided to the LLM, the LLM will reason and answer that I don't know. This is how we avoid the hallucination problem.

Tom, can you share with us some thoughts on what you think about the rest of the software industry and how they'll leverage Generative AI? They don't have the AI models underneath to fully inform responses to questions. But I would presume everyone will begin to use this to improve customer support or perhaps change their own user interface. So thoughts on the industry broadly?

I think that it varies from all this kind of yap out there, a lot of which is absolutely bunk and I won't comment on any names. Okay, to the other extreme is highly credible, which would be Microsoft. I mean these guys at Microsoft, this guy, Satya, is clearly a genius. Okay and you can see how they will leverage these LLMs in virtually all of their products, okay, in VS Code to help people write code, okay, in Microsoft Word to help us all write documents, in e-mail to help us respond and craft documents in their -- whatever their search engine is, okay, to -- what is their search engine? Bing, okay, I mean, those guys are going to leverage it in every product they have. And so you can see there's one where they're going to be able to use these LLMs incredibly powerfully right now, okay, to making their products even more competitive than they already are, which is pretty darn competitive.

And they've built into that -- to an ecosystem of [indiscernible]

What those guys will do quickly is really impressive. A lot of the rest of the, you know [indiscernible] out there, it's just all a bunch of yap. I mean, there's nobody who hasn't announced something dash GPT in the last month, right? And -- but Microsoft has -- they're clearly going to do it. Okay. I think, ladies and gentlemen, thank you so much for the courtesy of spending time with us today. It is -- we appreciate to share our thoughts with -- the opportunity to share our thoughts with you and give you an update and our perspective on the business. I think that it is today, June 22, 2023. I think by the time -- I think there's some probability, by the time we get to June 22, say, 2028, you'll be looking at one of the world's great software companies. So we thank you for your interest and we look forward to keeping you posted. Thank you, [indiscernible] I think that we have -- I think we have 4 product demos going on. And I think that they're serving red wine and cocktails or Coke or 7 Up or whatever you want out there. So please join us for a refreshment. We'll be happy to show you Generative AI or whatever you want to look at. Thank you.