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

Good morning, everyone. My name is Sundeep Kamath, and I'm going to be your host for the morning. It's a privilege to have you all join us. Thank you so much, and let us take you through the next 45 minutes odd on how to scale service with generative AI and Einstein GPT. So before we begin, I just wanted to give you a quick reminder. Salesforce is a publicly traded company. This is our safe harbor statement. Please make your purchasing decisions based on products and services that are currently available. We will be discussing a lot of future-looking visions today in our conversation, but we will make sure that we are calling out specifically what is available today and what will be available as we move forward in the future. That said, a little bit of housekeeping for everyone on the call. All attendees are muted, so please drop your questions and comments in the Ask Questions box, and we'll round back to them during our Q&A session towards the end. You will also see a related content box where you can download some great resources on automation and AI. So thank you, first of all, for joining us today. We are really excited to be discussing about GPT, more specifically around our solution on GPT. There's a lot of buzz in the market around generative AI, right? We are excited about it. I'm sure all of you on the call are very excited about it. Our customers are very excited about it. But they're telling us, they don't know where to start, which is why we're having the session today. I'm going to be joined by a great lineup of my fellow speakers. Joining me is Rajiv Garg, who is our Lead Solution Engineer at Salesforce for the Service Cloud portfolio. And a little later, we will also welcome our trailblazer customer, Mr. Sanjay Zaveri, who's the Head of CRM and Cloud Technologies for HDFC Limited. So let me just walk you through what we have in store for you today, right? We're going to start off with sharing our vision on what is generative AI. We will then talk to you about Einstein GPT and what that is, run you through how it works, and then finally, demo our solution, right? Always the best part, see how it works. We'll then have a fireside conversation with our customer, Mr. Zaveri, and then close out with next steps and resources for all of you to take away from this call. So the future of generative AI, right? Let me talk about it in context of service. I think every one of us service professionals understand that setting new excellence standards is the norm. Exceptional customer service, service excellence, doing more with less. All of these are day-to-day realities for a service organization, right? Everybody is trying to drive their competitive edge through their service experiences. But they have to balance it out with economic headwinds. They have to balance it out with cost. They have to balance it out with resource pressures. Generative AI has the potential to bridge that gap. And if we try and understand what generative AI is all about, right, I think it is the buzzword of the day. I'm sure a lot of you in the audience would be in a much better position to explain what generative AI is than I am, but let me take a stab at it. Bear with me, all right? We are all exposed to what traditional AI is, where there are neural networks, multiple parameters fed in, a lot of training, which will allow for predictions and image recognitions and text recognitions so that there can be some semblance of artificial intelligence which can be used for our day-to-day businesses. In generative AI, there is something called large language models. What are large language models? They are a kind of artificial intelligence that uses deep learning with massively large data sets to understand, summarize, generate and predict new content. LLMs, or large language models that we speak of, can be very accurate in their relevant field of application, which means we don't need to have those huge training data sets from our side anymore. And LLMs also have the ability to be very flexible with the output that they can provide. It can be text, it can be audio, images, video, code, much more. And more than anything else, I think there is a deeper understanding of the relationship between the data, which means large language models are more cognitive. They are able to generate more human-like responses than traditional AI. This is what generative AI is in a very short attempt by me. Please feel free to add in more points or your definitions of what generative AI means to you and let us -- we are happy to take that. It's been disrupting the industry. It has been a buzzword. It's been on the headlines for the last few months. I think without question, it's been one of the most revolutionary technologies that we have seen, at least in the 2020s. There is a buzzword that generative AI is becoming, right? But the thing about buzzwords are they come and go. But here, there seems to be general consensus that we are on the edge of something very fundamentally shifting change. Generative AI has the ability to unleash far-reaching opportunities. And with each of those opportunities come risks, and this is beyond what we can even imagine sitting here today. So when we look at how AI is disrupting our world, I think the first and foremost is how ChatGP (sic) [ ChatGPT ] has been impacting people's day-to-day lives in their personal front. If I need to write out a letter, I go to ChatGPT and I get that letter out. And I get a beautifully written letter, which I don't even need to worry about, no need of reviewing it, no need of spending half an hour writing that letter. It's done immediately for me, brings down the turnaround time from, let's say, 45 minutes to maybe 5 and in a much better way than I could ever have done personally myself. Now imagine if this is the norm that consumers are getting used to in their personal lives. Will they have any less expectations from our business? And this, I think, is becoming the challenge for all our customers who serve their customers, more speed, more personalization, more convenience. And they expect it to be tailor-made for them. They expect it to be on their terms, they expect it to be on the channel of their choice. How do we provide for that while, at the same time, ensuring that the following slide that I'm going to talk about is taken care of. What is the problem that we have? The problem as a business, when you look at AI, especially generative AI, there's a huge trust gap. The reason I say this is customers as it is do not trust companies with data. And without data being there, there is not going to be any AI because data has to be the foundational element for any artificial intelligence to be effective. We are seeing that there are inherent risks within the model of GPT where there is a possibility for data to be having privacy and security risks. We have possibilities of things like, I'm sure you heard of it, hallucinations coming in, toxicity, bias creeping in, which is why we wanted to focus this webinar today on the whole story of artificial intelligence, which has to include data and trust and give you all a playbook on how we can navigate this together. So what is Einstein GPT all about? Einstein GPT is something that is not new, but has been in the works for quite some time. And Salesforce has been doing this for almost a decade now. We started off with traditional AI in 2014. Over the last 6-odd years, we've been having trust as the foundation of our artificial intelligence through a think tank that we keep internally. And we make sure that everything that we are doing addresses to our core values of trust. We've done over 1 trillion predictions a week with our AI models. And we only want to take that higher and higher as we move forward into the world of generative AI. I spoke about trust, right? Why are we saying this is so important to us? It is a core value for us, nonetheless. And more importantly, we are in the business of serving businesses. And these businesses have their customers. The most paramount thing on your mind is going to be security and the ability for your customers to trust you, which means you should be able to trust us. And for that specific reason, like I said earlier, we have been -- for the last 6 years, had an ethical AI practice, which has been part of our Office of Ethical and Human Use. We're building a framework which includes a human in the loop, so customers can ensure that everything they have as an output from AI is first verified by a human and then is allowed to be sent out, shared, et cetera. So it continues to become more trained, it continues to become smarter, more accurate, more relevant and so on. From a relevance standpoint, I think artificial intelligence, like we spoke earlier, is only good -- as good as the data you use. And the great part is we are the custodians of your customer information as Salesforce, which means we have your CRM data on Salesforce, which is automatically available for our AI engine to use and to drive you relevant insights. Third, we have security. We spoke about why security is so important. We have a proprietary framework for Einstein, where you can use a variety of models without sacrificing security. GPT maintains secure data access, protects personally identifiable information and is purpose-built to work with the 360-degree view of the customer on our CRM. Finally, we have the ecosystem, right? While we have an amazing research team in-house, we are building a large language model framework of our own. We are also partnering formally with all the other players in this space, whether it is OpenAI or any of the others, to make sure we are bringing the freedom of choice to you as our customers. So what's exactly coming up in Einstein GPT? Outside of all the wonderful things you've already come to expect from our Einstein products, and here, I'm going to speak specifically about service, let me talk through about what is -- what are the use cases that we are seeing for Einstein GPT in service. Like I said, I manage the Service Cloud business for Salesforce, which means I'm a little biased towards Salesforce, but -- towards service. But look at it this way: service is how you as a brand differentiate yourself with somebody else, especially in a commoditized business, especially where the only selection of your product vis-a-vis your competitors could be a fighter of price. That is where service acts as that differentiator. Service allows you to have that premium over your peers in the marketplace. Now if we need to provide this differentiated service, generative AI can actually level up this whole experience. And there is where our first product, which is Service Replies, comes in. It has the ability for auto-generating responses, which are grounded in your CRM, with a flavor of Internet-based information around, let's say, location information or anything else. Let me take an example here of someone traveling into the Himalayas, and looking at a sweater or a jacket as something they want to buy on the website. And generative AI can come back and say, okay, listen, I know this jacket because the catalog is sitting in CRM, and I know that this jacket is for a particular temperature range. And right now, because the customer has been speaking with me, I know that the customer is on his way to, let's say, Srinagar, which is a much colder temperature running there right now, because of which maybe this jacket is not going to be enough. So let me recommend a much better jacket for the customer, a much warmer jacket for the customer from my side. Now as a human, did I have that information handy with me? Possibly not. But what generative AI can do is give me that information as a service reply because the generative AI has the information about the products in my catalog. It has the information about the weather in Srinagar right now, and it has access to the conversation I am having with the customer. So taking all of these in context, it can give me a nudge saying, you know what, you ask him to recommend this jacket instead. Now that is differential customer experience. And that is what is going to make a customer stick by you, right? So I think that is the first product that we have generally launched on Einstein GPT. The other piece is, of course, Work Summaries. All of us know as service professionals, how much time it takes to summarize, to disposition after every interaction, after every conversation, after every ticket. What if Einstein was able to do it for you? And what if after summarization, Einstein had the ability to create knowledge articles on the fly because this was something unique that was -- that came up in your organization. This was something which can have repeatability tomorrow, which can help other agents be more efficient. Can we create knowledge articles on the fly? And of course, can we generate answers to agents' and customers' questions that are directly coming in from your knowledge base but in context. Instead of traditional AI where we would have given you -- surfaced you a response from 1 knowledge article, can I combine responses from 3 or 4 knowledge articles and give you the answer you are asking? That's the remaining features that we are currently on our way to launch. Service Replies and Work Summaries are currently available. We launched them just recently last week on the 19th of July. We are going to be launching Knowledge Articles and Search Answers very soon on the product as well. Moving ahead, what's coming soon for field service? The ability for a technician to understand what is the work which is coming towards him. Can the work be summarized in a quick, short textual or voice format that the technician can understand because he doesn't need someone to explain to him what this is. He doesn't need to go through the whole job card to understand what exactly he's supposed to be doing, but get a quick summary. Is it possible for him to use the same knowledge search on the field? Is it possible for him to summarize the post work the same way an agent can do it in the contact center? Now all of these are functionalities which require an additional level of complexity because they are happening on the field versus a contact center, which is a much more controlled environment, right? Now that is the next set of innovations that Salesforce is planning to bring, especially on the field service part. Please tune in, you will hear about this very soon. Moving on. It's not enough to have an AI strategy alone, right? It is very important for you to have an AI strategy along with data that you can activate at your convenience. But it's easier said than done, right? At various points over the last few years, you might have had requirements for which you brought in point solutions. Now these point solutions work in isolation or they may be integrated with each other to pass on point information. All of this basically means if you need to have a holistic engine where you can understand and read data for artificial intelligence to be more efficient and more helpful to you as an organization, it becomes tough. These systems are siloed. There could be trust issues. There could be multiple other challenges which can hinder your AI strategy, which is why we want to, as Salesforce, provide you the ability to do an end-to-end AI strategy with Salesforce. And how we are able to do that is something I want to hand over to my colleague, Rajiv, to explain and show you on how this all comes to life. So thank you, everyone. And Rajiv, over to you.

Thank you, Sundeep. Good morning, everyone. I'm so excited to be here. And what I'm going to talk today is how Service GPT work. So let's take a look at under the hood and see how it works. All right. So what we have done at Salesforce is build an Einstein GPT trust layer, which is bridging the gap between your data and the LLM that allows you to use generative AI while still maintaining the control over data security and privacy. What we have done is using a very clever technique called Dynamic Grounding that allow customer context from your CRM data to enhance the prompt, which is given to the AI model. With this trust layer, what we can do is remove any kind of personally identifiable information, which is PII, from your prompts using different masking tools so that sensitive information is not processed by the AI models. And the generated results from these models are also checked for harmful and bias content and are also audited for compliance. More importantly, your data is never stored outside of Salesforce. It is very, very important as far as data security and privacy is concerned. Once an external model processes your prompt, both the prompt as well as the generated response or the output is immediately forgotten. They are not stored or kept for any reason such as monitoring or training because we believe that your data belongs to you and that's the end of it. So how does it work? So what we have done, if you look at the center, the first step is building the standard gateway for calling large language models. Now this single gateway gives us a standard set of integrations to interact with these models. Now either you can use any of the partner models like OpenAI, Cohere, Anthropic, or you can also use the model that we have developed at Salesforce, which is CodeGen. Now this GPT trust layer that I just talked about in the previous slide is the single central point of governance and the middleware services for tasks like prompt engineering. The next key component here is the Customer 360 data. With Data Cloud and Customer 360, we have access to all the relevant data to build prompts that are tailored to any use case. And our ability to ground these prompts in Customer 360 is critical to protecting trust and mitigating risk of model hallucinations or toxicity and, at the same time, delivering relevant and personalized output for the users. And once those responses are created, this stack allows you to do things like generate service replies on any channel, write work summaries based on conversation data, draft knowledge articles on or surface relevant and accurate answers. So let's take a look at -- through an example of how a service reply will work. When a customer is chatting with an agent, Einstein GPT will take that message and use it as a prompt template to create a service reply for that customer. But if the only thing that we're going to use it as a customer message, the response won't be any better than ChatGPT. So you can go to ChatGPT and ask that query and you will get a very generate response. There will be no personalization. And this is why we have this second step called Dynamic Grounding. And in this process, what we do is ground the prompt with the CRM data and add a bit of flavor in respect to company's language, the procedures, while masking the personal information. So then when we do send the prompt to the LLM in the step 3 with all the relevant information that we have taken from Customer 360, we receive a very accurate service reply that the agents can use either directly or in case they want to modify it and add their expertise, they can do that as well. And it is quite important to keep this agent in the loop, right? So we have this concept of human in the loop so that they can improve the model for future use, right? So I want to take next 5 minutes to demonstrate this through a demo video and show you how it comes to life. [Presentation]

All right. So I hope what you saw through the demo gives you an understanding of what Sundeep and I have been talking about and where we are headed to, right? But you all would want to know how to get started with GPT and this can be really a daunting and overwhelming task, right? But Salesforce has a long track record of democratization of software through clicks, not code. And our goal now is to democratize AI as well for everyone. We want you to set up -- set you up for success so you can create GPT-powered experiences for you and your customers. So the first step that you need to focus on is build and clean your knowledge base, which is a source of truth, and this is very important because we know that generative AI can have hallucinations which essentially means that GPT can make things up. On the long term, we can also bring additional data from other systems like Data Cloud. The second step is to set up your digital channels because your customers are going to interact with your business using messaging, chat, e-mail, self-service, so you need these channels set up ahead of time. After that, you can launch your AI journey with Service Cloud to set the foundation for GPT. This essentially means creating AI models and testing use cases with features like Reply Recommendation, which is built on a similar concept like GPT. And then you are ready to create GPT-powered experiences like Service Replies, Work Summaries and Knowledge Articles. With that, I would like to hand it back to Sundeep for a fireside chat with Sanjay.

Thank you so much, Rajiv. I think it was a brilliant demo, and hopefully, the audience got to understand what Salesforce has to provide in a little bit more detail than when we began. So thanks so much, and I will speak to you soon again. I'm sure all of you are eagerly waiting to hear from Mr. Zaveri on how they're using AI for service within their organization. So without further ado, let me go to the fireside chat. Hi, Sanjay, welcome and thank you for making the time to join us and our audience today.

Thank you, Sundeep. Thanks for the invite. Really appreciate that.

So Sanjay, I just wanted to kick this off. I know we have a lot of people on the call. They want to understand what we have around artificial intelligence, what do we have to offer our generative AI. And I'm sure you also have a particular point of view on this. So I just wanted to get things started off by asking what do you envision generative AI shaping the future of financial services? It's specifically in the context of CRM, specifically in the context of cloud technology. How can it help enhance customer experience? How can it help enhance operational efficiencies?

Thanks, Sundeep. It's a good question. So to begin with, any financial services, they would primarily what do we do? We normally handle the customer documents, the data, the proofs and images. And that's where it's like the future where the generative AI is going to be of help is in more of an operational activities. It depends on the organization to organization, but at least their focus will be how to improve their call centers. That's some -- I saw that you also had some examples, where any customers, if that are the background processes, because it is still at a very nascent state, the generative AI that is happening. So because all the banks or the financial institutions are normally regulated. So most of the organizations is what I feel they would be taking some internal steps that where they use such technologies and have that trust built and then expose it to their customer. So that's where it slowly is going to drive. But the good part, which I liked in the demo that you know what Rajiv was providing is that there is this trust layer. Because that's something which every client and every customer, irrespective of the industry, if it is in -- based out of certain countries, they would have certain regulations as you might have read about the Italian government not allowing Google to -- this is OpenAI. So those kind of regulations will be there. But at least future-wise, it has got a lot of operational efficiencies that any company would be able to look at where they have -- based on industries that you have. So if I can elaborate a bit more, so AI technology is not only ChatGPT. It has got images, it has got voice, it has got the text part of it. So it depends on the industry. But from a CRM perspective, we see that there's a lot of -- the text-related enhancement or the models that will work is going to help us in serving our customer better through the call centers, through the service requests, that might be going through -- the queries and the complaints. Because the way the world is moving towards, there would be time, which we are already in, but nobody would like to visit any financial institutions if they can do it sitting at home. So that sounds what the future would be. And it is going to help. I feel also to manage certain the operational team that if the customer is unhappy, what would be the right approach. And that's where all these AI-based technology is going to be handy for him to answer those questions as and when it gets matured. Right now, it is still a lot of trial and error that is going across because that's how it will be taken up by any financial institutions, which is governed by the RBI.

Absolutely. And specifically, you spoke about one such problem that you definitely anticipate in the world of financial services, which is around trust and security, right? I just wanted to extend that into a larger question because we have audiences across industries here today. Are there any specific challenges or concerns outside of trust, security that you anticipate in implementing generative AI? And how could organizations plan to address these?

So it's like -- see, any organization implementing any AI technology, so there will always be the prime concern that any company would ask -- would be like how do I manage my legacy applications. That would be the prime focus of how they would be doing it or what kind of challenges, how they're going to interact with that? The second point that I foresee that would be the challenge on the data previously and security that any organization would face. Plus there are certain regulatory compliance or the resources because they still are very -- I would say at a very early stages. The technology has matured, but not all the companies and the industries have matured enough that they have the team to manage to define this. And to address this, I would say that it's a kind of a collaboration because you need to work with your partners, the cloud technology partners who help you to implement that because as -- Salesforce is one of them who provides such kind of a support when you want to start a journey on any kind of an AI technology. Plus there are certain other forums where you can collaborate and you get yourself funded because at least I see that in a lot of U.S. and European clients, or the companies and the customers, they do get certain fundings from the organizations who will be interested in driving certain POCs. So that's how I would see that. And -- so you need to build how your data strategy is going to be because you need to ensure, at least what I read and what I understand is that still whether knowingly or unknowingly, there are a lot of developers who might have given some sensitive information outside in the public domain. So there are certain governance of the policies that the data production manager put, at least ensure that it is being reviewed, people are being trained and what kind of data, and I see that your trust lab does allow that you can always delete those data or you can always block and review that. So that is something is what any customer would be finding it as a solution because whenever they would like to go, they would like to control how the data has been flowing from any sources because normally, this large language pool is all data scrapping all across the globe. So there's nothing like it's a private or public. It's available. So that would be the challenges and a few of those is like internal strategy that any company has built and probably even the Indian government would bring their digital bill for the personal data protection. So that should be in collaboration, helping a lot of customers to ensure that whatever they are trying and testing at least that is secured.

I think it's a brilliant way to segue into my next question because the next question I had was around data privacy regulations. So Sanjay, you spoke about RBI earlier, and right now, you just spoke about how the Indian government is coming out with our data security bill, right? What do you see as the interplay between Gen AI and data privacy regulations? From a regulatory government framework standpoint, what steps do you think our customers should be taking to safeguard data and ensure compliance with regulatory frameworks? And I think in financial services, of course, you are 2 steps ahead of any other industry on what you should be doing to tackle this. But if you could share something for our audience, please?

Yes. So like -- see, based on my experience is that what we have been trying to do. So any generative AI technology is like it's going to go and search across the private domain as well as the public domain. So it's going to do a lot of data scrapping. And there is going to be a lot of -- that's for everyone's information. So then there's a lot of data or the information that is being provided to you, which can be private or which can be not without any consent. So that is something is what every customer is going to be having the information that what they are doing can be with content or without a consent. So then it's like -- and plus, there are a lot of providers who help you to when they build their products. And I'm sure even Salesforce would have considered that. But there would be a lot of providers who would have considered multiple jurisdictions and the laws. If you are in Europe, you have GDPR. If you're in the U.S., you have certain HIPAA compliances. So then they normally do get into those kind of regulatory effects and then they build the strategy around that, that any -- so when you want to start any generative AI, so you'd identify what is your ROI, what kind of strategy you want to implement that which are the areas which you want to expose it because you always would like to not play around with the data privacies and the way most of the governments are coming up, they are going to put a lot of power to the customers as well that we're taking the consent of the customers and those kind of challenges. So then this is going to be -- the way to safeguard is to define, first of all, what you would like to do or what you would like to achieve, define those use cases. Prepare or put a guard over the data or the policies that how people are using or the business is trying to use it or the developer they're trying to use it, put certain regulations around that, put a lot of trainings to the people that what can be in the public domain, what cannot be in the public domain when they are trying to scrap anything or finding any information using any of the AI technologies. That should at least initially help them. And by the time I'm sure all the partners with whom you collaborate, they would be coming up with a lot of solutions that help you to protect this kind of the data challenges that any organization or any country would face in the coming future.