Salesforce, Inc. (CRM) Earnings Call Transcript & Summary

Hello, everyone. Good morning and good afternoon to you all, depending on where you're joining us today, and thank you for making the time to be here. Before we move forward, I want to get through some quick housekeeping. So in addition to the content we'll be presenting, you also have additional resources that you can access from one of the panes below. [Operator Instructions] Great. Now with housekeeping out of the way, let's get started. My name is Maddie. I currently work as a lead solution engineer at Salesforce. I've got a diverse background that I now channel into tackling some of the public sector challenges in the APAC region. The areas I specialize in are data, AI and citizen or customer engagement. I have been in the Salesforce ecosystem for nearly 6 years, and I have been a customer and a partner as well prior to joining sales force. So my main mission now is to deliver the best outcome for both citizens and government agencies through our use of our technology. I'll now pass on to Bryan to introduce himself.

Thanks, Maddie. My name is Bryan Wise, and I'm a distinguished solution engineer on our global public sector specialist team here at Salesforce. I started my career in technology, about 20-plus years ago in the United States Navy, getting into technology there and been working in the business intelligence, data warehousing space ever since. I spent some time also at a natural language processing app exchange company that focused on analyzing unstructured data inside of Salesforce org and was the Chief Data Officer of the company called Halosight, and excited to be back here at Salesforce, helping government customers find ways to use AI. And with that, we're going to -- Maddie's going to start our presentation now.

Let's jump into the agenda for today. We will talk about the wave of Gen AI, its landscape and what's changed, followed by a discussion of different kinds of high and more a means for the public sector. We will also then introduce you to Salesforce's Gen AI capabilities and depth, and how Salesforce Einstein trust layer is going to be the most crucial part of any of our customers' AI initiatives. We will also take you through a short demo on how you can realize value through Salesforce Einstein. We will then stop and answer any questions that you may have in the end. Before we jump into the content, I want to remind all our attendees to make purchasing decisions based on features that are currently available within our products. This is especially true for rapidly evolving capabilities with themselves for such as data and AI. Let's now move forward with the content. AI has come a long way in just the past 10 years. And the difference is remarkable. I'm sure each of you are here today because you recognize that. Today, at your fingertips more, if not all of us have many forms of AI from virtual assistance to chatbot, even ChatGPT on your phones. Earlier, the barriers to entry were really high with massive costs and a steep learning curve, which often requires a deep understanding of complex mathematical concepts and the ability to write code from scratch to implement AI models. And that doesn't even cover costs to build these models. But now it's democratized, it's widely accepted and it's also impacting every aspect of the workplace. It went from being science fiction to being a crucial part of our daily lives, breaking down traditional barriers and making us all work smarter, serve our customers or citizens better and unlock the full potential of AI in the public sector. Today, we'll dig deeper into these changes to see how AI is reshaping the way we do things in public sector and the government all in plain simple terms, of course. While understanding AI is the first part, many of us struggle to bridge the gap and translating it into outcomes. And that's precisely why Salesforce's approach to AI is to focus on tactical and realistic use cases that align with your mission. AI in itself is not really a strategy, but how we employ and what we employ a particular AI capability to accelerate or improve will become the strategy. Now our best recommended approach is to explore use cases and outcomes while keeping a keen focus on guardrail for AI and on increasing productivity. Someone very smart recently told me that we stopped calling something AI when it does something useful. One is a recognized space is we don't call it AI. We call it facial recognition. Once it can understand natural language, we call it natural language processing. At Salesforce, we've had various AR capabilities for years, and we call it Einstein, whether it is bot, lead scoring, next content recommendation or best action recommendation. Einstein has delivered many outcomes for our customers over the last decade. Having invested in AI since 2014 and starting to work on large language models in 2018. Salesforce's AI research team now has over 200 patents and has published more than 227 papers. This knowledge allows us to recognize the diversity of users, such as our audience for those who have been out today, you're all diverse, you all have individual and diverse use cases, and so are the capabilities within the AI ecosystem and that a nuanced approach to policy and practice is extremely crucial. Therefore, Trusted AI is the top priority at Salesforce, and Bryan will cover the Einstein trust layer in depth later in our presentation. Now that we've covered a brief introduction to the current state of AI, this is a great time to discuss the various types of AI that are beneficial to organizations. If you ever have been curious and search the definition of AI, you would go back to the original definition from the dark math summaries research project, which is all the way back in 1956, which is when AI was defined as an endeavor to create machines that could mimic human intelligence. Specifically, the ability to perform tasks that typically required human cognitive functions, such as learning, problem-solving and reasoning. So AI comes in 3 different buckets: automation, predictive and generative. The definition of automation is pretty straightforward. It helps automate tasks on the back of various triggers or inputs. For predictive AI, this is where the machine learning is leveraged for making predictions based on what the AI has seen in existing data. An example of this would be the likelihood of someone contacting a disease or being eligible for a benefit based on their application. Now generative AI is what it says it is. Generative AI can produce text, images or videos based on its training. Now generative AI does not predict outcomes, it generates. Now that we are grounded on AI, let's look at what Salesforce has done and is going to do in each of these areas. If you recall earlier in the presentation, I mentioned how AI in isolation is not a strategy. So keeping that in mind, in each area, we are ensuring that the AI is being embedded in the flow of work in order to drive efficiencies and enhance jobs for our customers. For automation, you may be familiar with the tools that Salesforce has been leveraging in various formats over the last 15 years. Using tools such as Flow, the Business Rules Engine, Omnistudio and Einstein bots to automate self-service with natural language processing. With predictive AI, Salesforce in the last 7 years has delivered features that can help predict classification, provide content recommendation or predicting the best time to reach a citizen through some time optimization. We've been a leader in predictive AI, and we continue to expand our offerings. In the last 6 months, we have pivoted our focus to adding the value of generative AI in the flow of work and across sales or ecosystem. With Einstein Copilot, we bring generative AI through large language models such as Open AI, Amazon SageMaker and more. We're doing that to bring it into the flow of work, whether it is write in an e-mail, generate a segment or a cohort and it's relevant communication content, generating a case summary or writing knowledge articles, so on and so forth. I will now hand over to Bryan, who will take us through the generative AI capabilities in depth.

Thanks, Maddie. Let's dive a little deeper into generative AI now and look at how Salesforce envisions it helping in the flow of work across your organization. Whether it's about delivering proactive service or enabling more effective communication we see multiple ways generative AI can enhance all facets of your CRM operations. For instance, if we look at the service domain for a moment, we see generative AI being able to free up some of your agents' time by writing case summaries for you or by providing recommended replies for your agents. On the IT side of the house, generative AI can accelerate your development with tools like Einstein for developers, which can write APEX code based on your natural language prompting or maybe is about generating content that is going to be used as part of your outreach to your customers. There are numerous ways generative AI can help your employees be more productive. About the possibilities with generative AI, they usually get pretty excited about these use cases. However, they are often a bit leery of it as well. For example, they know they need to use their data to make the AI relevant but are concerned about the security and privacy implications of actually using that data. Salesforce recognizes this and has made trust an integral part of our AI capabilities. Let's walk through a quick example to show how the Salesforce Einstein trust layer works. In just a moment, you'll see a demo using Salesforce chat. A citizen will be chatting with an agent asking questions about a benefit application. Salesforce generative AI will provide a recommended response for the case worker to use in their conversation with the citizen. To provide a recommended response, Salesforce constructs a prompt to send to a large language model. The first thing that goes in the prompt is the question the citizen asked. Next, Salesforce adds some additional context by getting information about the person asking the question from the CRM. Finally, Salesforce will scan the knowledge base to see if there are any applicable knowledge articles that might apply to this question. So in summary, we have now grounded the prompt being sent to the large language model with the citizen's original question, information about the person asking the question and information about your organization's policies. And we are almost ready to send that prompt to the large language model and ask it to generate the suggested response. But before we do, we actually scan the entire prompt and mask all PII data in that prompt. We are about to send this information out to a large language model and do not want any personal data going out to this model. In our generative AI solution today, we are sending the request out through our secure large language model gateway to a partner solution. Part of our agreement with this partner is that they will not retain any data on their end. Additionally, we've designed this large language model gateway to support partner models as we are seeing in this example, models that Salesforce host or models that our customers might provide. This trust infrastructure works with all 3 of these scenarios. Okay. Let's get back to our prompt. After sending this off to the LLM, Salesforce gets back a suggested response. Before this response is shown to anyone, it is checked for toxicity. We check for things like racial bias, profanity and hate speech. If the toxicity score is too high, nothing has returned to the agent. If the score is good, the agent will then see the suggested response to the citizen, and this is where we make sure to keep a human in the loop. The agent now has 3 choices. They can use the generated response as is and send it directly to the citizen or they can edit the response and then send it to the system or if they think the response just isn't that good, they can market as bad and write their own response. In each of these cases, the prompt, the response and the agent's actions are all logged in the audit trail in your Salesforce work. As I mentioned, you'll see all this in action in a demo in just a few minutes. Additionally, Maddie will show how this trust layer fits into the larger sales force architecture. So far, we've talked about predictive and generative AI. Here at Salesforce, we're excited for how both of these types of AI can help public sector organizations overcome common challenges like an unpredictable workload or an aging workforce, and help you accomplish your mission by assisting your employees right in the flow of their work. We do this by providing AI embedded into applications that use your data but we do this in a trusted fashion, making sure your data is used appropriately and stays protected. When I give briefings on AI, at the end, I always get the question. Well, that's all very interesting, Bryan, but how do I actually get started? There are 3 key areas to get started with right away. First, make sure you understand your policy on AI. When is it appropriate to use? What is your evaluation criteria for if AI is applicable? Many organizations are still trying to figure this out right now. Salesforce recommends a risk-based approach to evaluating AI. Next, start thinking about the possible use cases for AI to assist you in your organization. Meet with your business users and do your own discovery to see where it might be helpful. Finally, start thinking about your data. Did you have the necessary data to either build a model or properly ground prompts being sent to large language models? Let's talk about use cases a bit more, and then Maddie will spend some time discussing how to connect to your data. As we have mentioned several times, we see AI assisting your employees. So as you're exploring use cases, we recommend looking for those tedious tasks your users have to do each day. Maybe as the call center agent that has to take several minutes after each call to summarize the conversation they just had or maybe it's that brand new agent you hired that is scrambling and searching for the right policy or procedure. These are 2 perfect places where AI could assist. We also recommend looking for ways to enhance your customers' experience. Can you take a paper form or a visit to the office and replace that with a bot or some other self-service experience. Now let's go back to Maddie and talk about the overall AI architecture here at Salesforce, and how you can get access to all the data needed for these use cases we just discussed.

Thanks, Bryan. Let's talk through architecture from the ground up. At the very bottom, we have Hyperforce, which is our public cloud infrastructure that enables us to securely deploy and scale all of our data and AI services and do so while meeting compliance requirements. At Salesforce, everything starts with a solid data foundation. So the next layer app is Data Cloud, which provides our platform with the native hyperscale data accounts for bringing all your customer data together. This is to help you build a single source of truth in the form of unified customer profiles. We also enable you to use data from other data lake houses such as Snowflake, through zero-ETL access in order to create a single source of truth. Data Cloud is also critical to grounding AI models. But Salesforce is unique in that we don't provide you which is one model. We have an open model ecosystem strategy. So in order to bring your own models, we provide secure access to foundation models hosted outside of Salesforce via API, like let's say Open AI or Entropic. We also provide you with the flexibility to bring your own large language models. Just as you bring your own data lake, so that you can reuse your existing generative AI investments within the Salesforce ecosystem as well. We have a trust layer that is embedded into the architecture. And as we've seen in depth, this is to ensure that you can leverage the generative AI capabilities securely and confidently. Additionally, AI at sales force is customizable. And in a typical sales force fashion, we have low code builders such as our new Einstein Copilot Studio, which is a configuration layer for that new Einstein Copilot. Think of Einstein profile as your conversational AI assistant that exists in every sales force app, enabling you to ask questions and not only get answers, but take actions that empower your team. So as you can see, instead of building a separate AI stack, we bake AI into our metadata architecture. So it just works where you and your teams are and on your terms with the highest standards for trust. Now let's talk about the most important enabler for any successful AI application. It's your data. We have seen time and again how AI prediction or generation is of little value without the contextual data that is specific to your organization. Earlier when Bryan took us through the trust layer, we saw that Einstein is able to retrieve that critical contextual data. Einstein is able to do so by leveraging data from data cloud. While our customers have a lot of contextual data stored within their CRM, we have to acknowledge and recognize that especially in a public sector landscape, data of varying types are scattered across several systems. Now take a moment to think about all the different types of citizen or customer data that exists in your organization. Now make a quick mental note of how diverse fragmented they are. And often, they are creating their own data silos. Now these data silos are typically hard to resolve. And it almost becomes either impossible or extremely labor-intensive when the systems are legacy systems or on-prem systems. Now imagine if you could bring all of your data together, and not only just bring them but also normalize and harmonize that data with very little custom configuration and that's Data Cloud. Let's look at the slide from left to right. Let's talk about the first step, connect. With Data Cloud, you can connect to all of your customer data, engagement data within Salesforce ecosystem such as from Service Cloud or from marketing cloud. Extend this data with your data lake house and warehouse data because data cloud support structured, semi-structured and unstructured data. You can leverage almost any data that you have in any of your systems. You can leverage MuleSoft, our integration platform to unlock data stored within any legacy systems and connect that to Data Cloud as well. Now let's move to the next step. Once you bring all of this data into Data Cloud, with just a few steps of configuration you can harmonize and then also map all of that data into a canonical data model that is available out of the box. Now this data model can also be extended or customized to support your business. Once the mapping is done, you are then able to leverage powerful matching rules to identify customers or citizens whose data was previously fragmented, was scattered across various systems. You not only have that view now, but you can also use reconciliation rules to create a unified profile within this data model. And this is how you would create that single source of truth of your citizen or your customer data. This data then in addition to powering our AI model can also be leveraged across other Salesforce apps such as Marketing Cloud for sending a proactive communication while also honoring the consent and privacy of that particular citizen, which is an important call out for our public sector customers. You can also leverage this data within the Service Cloud for any call center or service use cases and also within Tableau for any insights. Now without further ado, I will jump to the amazing demo that Ashwin Kumar, who is one of our very talented solution engineers, has built for us. [Presentation]

Fantastic. I think we've got a few questions come to specifically on the availability of Marketing Cloud in Hyperforce. Just a quick answer to that is [ Bot ] Marketing Cloud and Data Cloud are on the road map to be in region on Hyperforce. We recommend keeping in track of it and keep getting in touch with your AE to know exactly how and when you can get on the Hyperforce. All right. So Bryan, I was thinking maybe some of the relevant questions that we have had during our experience. It will be worthwhile sharing it with the group here. I think one of the first ones, I think a lot of us would be thinking about if I were in customers' shoes is what are the best or successful AI implementation? What does it look like? How would we define successfully our implementation? Would you like to pitch in?

Yes. Sure. So actually, Maddie, I just saw 2 questions come in from customers here while you were giving that question. So maybe I'll take those really quick and then come back to yours. So one, where any of the voice is done by AI? Actually no on that one. However, we have used AI for some of our voices in our presentation. Those were actually all live humans talking in there. So those weren't done by AI. Hopefully, we didn't sound too robotic in nature. And then another question was what is the recommended approach to getting my organization's knowledge base ready for AI? That is a great question. I would recommend -- one of the things I'd recommend is going out to Trailhead. Trailhead is Salesforce's free learning environment. There are some great paths on AI inside of Trailhead. Some of them are about AI in general. Some of the things we talked about today, the trust layer, safety within AI, ethical AI development. Some of them are much more tactical and get into actually how to use Salesforce AI products. And there's also a brand-new AI certification available from Salesforce, if that's something you're interested in as well. So Maddie, going back to your question, you were asking you about successful implementations, right, and just kind of what I've seen on successful implementations. One of the implementations I worked on way was at a partner organization inside of Salesforce. It was with a commercial organization, but I think it's applicable kind of the lessons learned here for a government organization. And what was, to me, so successful is they wanted to use natural language processing to better inform their Salesforce. But rather than building some new dashboard that they had to go to or some new tool, we took the insights that we got from natural language processing and built that into a next best action for their users. And so as their sales reps would go in and pull up an account, they would get the next best action that was based on information from the notes that had been left, from the actual sales that have been done, pull all of that together. But just as you said in the webinar, they didn't really know they were using AI. It was just there on the screen. We had 1,800 people using that every day on a regular basis, and it was just neat to see it part of their daily work environment. And did not have to do any training as we rolled it out to those 1,800 users. I think that also kind of spoke to the success of that implementation there.

Absolutely. That's put on isn't about not even knowing you're using AI, while using AI, and that's the best way you can define it, right? Like AI is here to enhance what we already do. So support business strategies and outcomes rather than having to do something you need just to use AI that usually wouldn't be successful.

Maddie, just one thing to mention really quick on that topic is as we've talked to people about their policy around AI. That's a common question we get, what are other government agencies doing with respect to policy? There's been a lot of announcements this week from like the U.S. White House, the U.K., G7. And one of the big things is how the purpose behind what you're doing, right? You don't go and buy a tool just for the sake of a tool. You buy that tool to build a project, to do a project. So what are you trying to do? What is the outcome you want. And if you're meeting that outcome, that's a good indicator of whether or not your AI was successful.

Absolutely, absolutely. I think it also brings to light I think the next thing anybody we might be thinking about is how do we make this a reality in our organization today. Now I understand AI. I understand how it can help, but there are still steps that we need to take as organizations to get there. So the next identification could be who is the best person to lead this or who will be the best stakeholders to sort of bring this together in an organization? And I think that will be a good call out as well is we can talk through that.

Yes. And this kind of goes back to the question that was asked about how do you get your organizations kind of knowledge-based ready for AI that was asked out there. And when I answered this question like 10 years ago, it's a very different answer than today. So 10 years ago, I would have been recommending getting the data scientists there, maybe a mathematician, who understood the models that were being used by the data scientists and could evaluate the effectiveness. And then somebody from the IT organization that can hopefully implement and maintain those, and then maybe somebody from the business, and it was kind of a tough to get that group together. And I think where we are today and some of the changes in the last year on AI is we can focus on that business user, right? What's the business outcome, how do we want to use AI, how would it impact the business? I think we've got to have security involved, right? We need to look at the privacy concerns, the safety concerns and your policies and looking at that. And then lastly, what we're really trying to hope to do on the Salesforce level is maybe it's your Salesforce administrator that you're bringing to bear here that is utilizing functionality that leverages AI, but we maybe don't need as deep of a data scientist. We definitely don't need that mathematician, which, unfortunately, that was my degree day, putting myself out of a job here. But yes, I think that is maybe the core group of people that I would get together to start looking at use cases, start experimenting with AI, making sure your data is ready to go and so you have a successful implementation there.

That's fascinating, isn't it, how much things have changed. And it's also really interesting and exciting because from what I'm hearing, you say, Bryan, which means a lot of our customers are already ready for this journey because a lot of our public sector customers already have a team, which has policies and security in place. And we already have a lot of our customers who have Salesforce administrators and a lot of other sort of support roles around our technology. So it's really about, I guess, getting up to speed on those guardrails for AI from those teams and they would be pretty much ready to go in exploring how AI can enhance their outcomes today.

Maddie, just got a new question in here asking about the unified customer profile and how AI can help identifying a golden customer record from various data sources included even legacy ones. So that unified profile is part of our data cloud capabilities. And with Data Cloud, you're bringing in data from various sources, including legacy sources, Salesforce sources, bringing in that customer information. And we bring that in at what we call the individual level. So you've got maybe a record for Maddie that's coming out of the Salesforce system, maybe a record for Maddie that's coming out of the claims management system or a benefits management system. I've got a record for Maddie in my subscriber list, where I've sent some outreach e-mails. I've got 4 or 5 records for Maddie. What Data Cloud does is it allows you to easily look for a way to identify where we have matches. And I've got Maddie in multiple systems, and that is 1 Maddie. Now how do we really have AI helping in there? Specifically, we have some models around names, specifically, the first name. So when we're looking at matching, we might say it matches on name and e-mail, name and address, name and phone number, name personal identifier like a driver's license or a social insurance number or something like that. But specifically on the names, we have some AI there that does fuzzy matching on the names. And if you go out, just even to our documentation on Data Cloud and look for fuzzy matching, it will describe the 3 different levels of precision, doing things like identifying culturally different spellings, accent marks and characters, spaces and spellings, abbreviations, initials and takes into account a lot of different things there. And so that's kind of how we use AI within that unified profile process.

And just to add to that, Data Cloud is unique in the way it unifies the customer profile is that it persists your source profile. So we're not actually merging them, but we're creating a layer on top of those profiles. So you don't actually lose context because typically, in a lot of our customers' environments, you do have those independent records for operational reasons and legitimate reasons that you need to use in the sources. So we're not going to go ahead and change anything, but we're going to be able to provide this golden record or unified profile, so to speak, for you to access to either communicate proactively or take action based on insight. So there's various ways to use that data.

Okay. Another question that we got is, if we could share the impact of AI, Gen AI on data analytics. So I think there's a number of interesting ways there that AI can help on the generative -- or on the data analytics side. One is explaining the analytics, right? You build this awesome dashboard that has an incredible visualization in there, and you happen to see what that means or maybe what's driving that change on the chart. Being able to use something that can, in a natural language voice explain the insights and pull that out. Also just the ability to use things like having -- clicking on an outlier in a chart and saying, explain this outlier to me, and having good techniques happen underneath the covers in terms of outlier analysis and looking at the relationships between the features in that data set and say, okay, just automatically looking at that and say, all right. I've got this outlier. I have all these features that impact the model and doing a quick feature analysis there and saying which components impacted that score the most. Being able to describe that and not have to go through that process, okay, I want to do an outlier analysis. I want to look at those sorts of things. The other way we're seeing it, and this has been kind of common in the analytics space is can I use natural language to request the analytics that I want. So natural language query capability and have the analytic built for me. We're seeing that being introduced into our data cloud product in terms of natural language definition of a segment. So identifying a group of a population that you want to reach out to, just write the definition of what that is, and it will build the segment for you, and it builds it in a way that's still editable so you can adjust that as well. And so those are just a couple of examples there. Another question here is about Einstein chatbots. They say they've built chatbots and they've got them deployed on their website. However, when will the AI component be available? So one thing on that, our chatbots, AI is used in a couple of different ways. One is in determining intent just the basic natural language processing. That's been part of our bot for quite some time. So you can define those intents, do some training data, build in automatic recommendation or automatic replies based on that intent. So that's something that is available today. But I'm guessing you're maybe asking more about some of the AI and generative AI capabilities that were shown in the demo there. And so recommended replies is generally available today, and it's generally available for chat specifically. So that would not be though the Einstein chatbot. So specifically, that is a live agent chat. So maybe I started on the Einstein bot. I fail over to the live agent and then we can have the recommended replies. That is generally available. That's part of our conscious decision, though, to have an agent in the loop, a customer in the loop and making sure that we're getting the right information out to the customer before they see that.

We also had another question come through on case summaries. The question is, can Einstein create case summaries especially for cases where the actual case investigation may last several months and the details are recorded across several related objects on the case.

Yes, that's a great question. I think this is a good example of something that's coming in, in pilot right now on the Salesforce side, so not yet generally available is the ability to like generate a summary. I guess we have case summaries available, GA, but that is not the case you're describing there in the question. So this is more like a chat case, and I want to summarize that chat transcript. That is available today. However, I think this is a great example of where that bring your own model capability comes into place. We've seen this a lot with child welfare is where this often comes up, where you have a case that might run for years. And I need -- I'm a new case worker working with this child, I need to get up to speed very quickly. There are several techniques on the large language model where maybe you go in and summarize each case workers notes over the years and then doing a summary of summaries. And we've seen some of that built out in proof of concepts. We've seen some of that built out with partners and the ability to do that. That is something that our public sector product management is specifically looking at though because that investigative case management is often a long-running case with large documents as well as case notes and summarization. So definitely a use case that we're seeing and a place where our out-of-the-box summarization probably won't go directly after that, but we're building an architecture, providing an architecture where we can hopefully work with you to go and solve these types of problems. I think that's all the questions that we have today. Thank you for joining. Thank you for your questions and participation here. and look forward to talking to you more in the future.

Wonderful. Thanks, everyone. Thank you for your time.