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

Good morning, good afternoon, everyone, and thank you for joining. Good afternoon, everyone, and thank you for joining this webinar today. My name is Kevin Domadia, and I'm a service cloud solution engineer at Salesforce. I have a decade-long experience in the service industry. And today, I'm going to talk about the future of service and how AI will transform your customer experience. But before we start the webinar, I just wanted to take you through our safe harbor slide. That this presentation contains forward-looking statements about, among other things, trend analysis and future events, future financial performances, anticipated growth, industry prospects, ESG goals and the other anticipated benefits of acquired companies. Salesforce is a publicly traded organization, and we would request you to take your decisions based on the information which is publicly available to you. First of all, I would like to thank all of you for being our customers and our partners. We hope to make today a valuable use of their time. Your success is a priority for us. We are committed to being your partner in driving customer success together in these days and years ahead. I would want to begin today's session with a story which depicts the future of service. Raj, Simran, and their daughter, Neha, used a generative AI system to plan their vacation. The AI created a personal itinerary, travel guide and assisted them with bookings. Impressed by AI's convenience, they booked their trip through it. AI provided tickets, vouchers and even arranged transportation. During their vacation in Paris and Tokyo, AI guided them and enhanced their experience. On the day of their departure, Simran and Raj packed their bags and got ready to leave. AI assisted with boarding passes, itinerary and personalized travel trips. In Paris, they enjoyed a warm welcome from AI and the hotel, complete with a view of the Eiffel Tower and a thoughtful welcome basket, which contained their favorite wine, chocolates and a teddy bear for Neha. Neha loved that teddy bear. Their final destination was Tokyo, where they spent 3 days exploring the city and its attractions. They used AI to guide them and book tickets, reservations and tours. AI also gave them tips on how to avoid crowds, save money and enjoy the local culture. Returning home, Simran and Raj realized that Neha had left her teddy bear in her Tokyo hotel. She was very upset and she started to cry. She asked her parents to get her teddy bear back. Raj and Simran felt very sorry for their daughter and they tried to calm her down. Neha insisted on getting the same teddy bear back instead of a new one. Using their smart devices, they contacted the hotel via a bot who apologized, promised to help them find their teddy bear and also hinted at a surprise. Grateful, Raj and Simran awaited the e-mail. A few minutes later, they received an e-mail from the hotel. They opened it and saw that it had contained a series of photos, which were generated by AI. The photo showed Neha's teddy bear having fun and adventures in Tokyo. The teddy bear was wearing a Kimono, a samurai sword and a Ninja mask. The teddy bear was also posing with the hotel staff, the bullet train, the Tokyo Tower and also the cherry blossoms. The teddy bear was smiling and waving in every photo. Parallelly, AI checked the inventory of the lost and found department and found the teddy bear, which was left behind. It checked the room number Neha was staying at, and confirmed that it was Neha's missing doll. It also generated an e-mail, which read: Dear Simran, Raj and Neha, apologies for the teddy bear mishap. We found him in our lost and found section, and we'll send him back soon. Before his return, we gave him a Tokyo tour. Enjoy the photos of his adventure and know he misses you, too. He loves you, and hope these photos bring a smile and Tokyo memories. Thanks for choosing our hotel. Best wishes. Simran and Raj were touched by this particular e-mail and also the photos. They showed them to Neha and she was overjoyed. She stopped crying and started laughing, and she hugged her parents and thanked them for the teddy bear. Isn't this the kind of experience that you would want to give your customers? To get to the future ahead of our customers and to greet them when they reach there, we need to find a way to securely and effectively use the technology which is available today. Now service organizations like yourself are focused on making sure the right technician or the right agent gets assigned at the right time, right place and they are armed with the right tools and systems to nail the job on the first go itself. That's how you deliver an outstanding service and create memorable customer experiences. But let's face it. It's easier said than done. Even for service leaders like yourself, you are focused on so many other things like cutting costs, maximizing ROI, streamlining your vendor relationships, adapting to your workforce challenges and also building strong customer connection. It is a daunting list of responsibility that keeps on growing and all -- and you need all the support that you can get. And if you look at today's economic environment, service teams are being asked to do more with less as the customer expectations rise to an all-time high. Today's customer wants personalized service delivery faster than ever before. This means that the service teams must focus both on quality of service and also the speed of delivery. Unfortunately, a lot of times, they have to choose between the -- 1 of the 2 because they are spending too many hours on repetitive tasks, and they have to worry -- work across different systems and applications. And there's a lot of swivel chair, which is required to resolve a particular issue or respond to a query. Technology that will help them bridge this gap between scaling efficient service and that high quality which the customers are expecting today. So how can companies deliver great customer service at scale while driving cost savings? Now today, if you look at it, the talk of the town is ChatGPT. And it is the fastest-growing service ever. It crossed almost 100 million users just within 2 months of its launch. And needless to say, this technology has become a part of your customers' everyday life and their expectations are rising fast. So which is directly also impacting us as customer service agents and field service agents. Your customers now expect a faster response. Along with the speed of the response, they're also looking at tailored experiences and personalized communications. And then along with that, they also want to reach out to your customer service at a convenient time and at a channel, which is -- which they are comfortable interacting with. So the good news is that we have GPT now, and it is available for everyone. And you can use it to finally deliver the service you've always wanted to provide at a scale. When you harness the power of AI and GPT technology, it makes your processes easier and faster for your service teams. GPT-powered briefing and wrap-ups will help your frontline prepare for their next job. GPT-powered search and knowledge will help your teams and your customers find relevant information faster. Also, intelligent automation can help you optimize your current processes and also optimize your scheduling and dispatching activities. This is a very powerful technology, and that's a game changer for service. And that's why 84% of the leaders are confident today that generative AI will help their organizations better serve their customers. But of course, this is all easier said than done. Over 75% of executives struggle to deploy AI effectively. And this is because there is too much of data which is lying in siloed and legacy systems. And these systems, they don't talk to each other. This makes the change management an uphill battle, and suddenly your vision for creating a streamlined and personalized experience for every customer seems out of reach. Service teams now they need a solution, which will help you to harness and deploy your AI effectively. And when we talk about GPT, while it's good and at fluency, but it is very -- it's not as good when it comes to accuracy. It can make things up. And I'm sure that you all must have heard about the term hallucination. Now hallucination is where GPT goes ahead and creates scenarios which never existed in the past. It can also create new security attack surfaces, and it can enable leaking of confidential information as well. Now there are some pronged attacks, which are also called as poisoning attacks, which can happen on your GPT deployments where the users would deliberately provide generative AI system with racist or explicit content which would then cause the chatbot to incorporate inappropriate material into its training data. And this will, of course, lead to offensive output for your customers. And I guess most of us are working in organizations where today, GPT is blocked for internal usage. And one of the reasons is because there is a trust gap within the AI area as of today. So if you look at leaders, right, while they are, of course -- they're looking forward to implementing GPT, but at the same time, they have a lot of questions. The first question is, of course, how is it going to impact their agents and workforce? How would they be able to deliver better experiences with GPT? So finding use case is, of course, another challenge, which the leaders are facing. And along with that, right, what was going to happen in the next couple of years and how would their teams have to upskill or reskill in order to ensure that they're able to provide new service with the help of GPT. But along with this, there is also one big challenge and a big question, which is going in all the leaders mind, and this is what if it produces false or misleading information. Now there are so many news that we keep on hearing around GPT where it has provided wrong information. For example, there was this lawyer in the United States who cited 6 fake cases which were made up by GPT because he was using GPT for actually doing research and trying to prepare his defense in a legal brief, and it was filed in the Federal Court. Now the judge called this unprecedented and scheduled a special hearing to determine the possible sanctions which they can put on this. There was a group of researchers who created 104 examples of prompts, which could make GPT output falsehood, such as claiming that sun is blue or the humans have wings. So these are some of the challenges, which because -- which GPT has as of today. Because of which the leaders are being careful in terms of implementing it. But what if we could take everything which makes this AI unique and rethink our work around that. At Salesforce, we started our AI journey back in 2014 where we first formed an AI research team. And then we started making a lot of acquisitions around it. Then back in 2016, we got all our AI solutions under the Salesforce Einstein umbrella, and we kept on doing more research where we got more features like bots, vision and language. Then in 2018, we were one of the first few companies globally to start an office of ethical and humane use to ensure that the data which is collected by AI and the responses which are generated by it do not cause any trouble in terms of security or in terms of any other unethical behaviors. And then we kept on adding more and more solutions like Recommended Builder, Einstein Search. And then in 2023, we started piloting with GPT capabilities. Now as we move forward, you would keep on seeing new and new GPT solutions built out-of-box on our platform available to you in the future releases. And in this entire journey, we have been -- we have released more than 227 research papers only on AI, and we have around 300 patents today. Now let's talk about the new Einstein platform, which you might have heard about during our Dreamforce event. The Einstein platform is delivered on our Hyperforce public cloud infrastructure, which enables us to securely deploy and scale all of our data and AI services and do so while meeting all the compliance requirements. At Salesforce, everything starts with a solid data foundation. Data cloud provides our platform with a native hyperscale data lake for ingesting, harmonizing and activating your customer data. This will help you to build a single source of truth in the form of unified customer profiles. Our Data Cloud is very critical for grounding our AI models. And what is grounding? It's nothing but contextualizing the outputs and contextualizing the prompts which are sent to these LLM models. But Salesforce is unique in this way that we not only provide out-of-the-box integrations with some of the leading AI providers, but we also give you the option to bring your own LLM models, which will then help you to create your own private mode, and then you would be able to fine tune it based on your requirements. And because trust is our #1 value, we've created the Einstein Trust Layer, which provides you with security guardrails that would enable you to ground those AI models using data in CRM and Data Cloud. But along with that, you would also get controls like zero data retention that would protect your data security and privacy. And when we talk about AI at Salesforce, it is all customizable and it's all build on the low code, no code principles. So you could use our drag-and-drop engines to then go ahead and build AI solutions on top of it. Einstein Copilot Studio would then help you to build copilots for your customers and for your employees, which will not only have the capability of responding to your customers and employees, but also it will have the capability of performing task and taking actions automatically. And this is how we would then be able -- this is how we would then get all our portfolio of apps for sales, service, marketing experience, to name a few, which, of course, are more smarter and more efficient. Now let's double-click on the Einstein Trust Layer. So have you ever wondered that where does your data go when you submit a prompt and especially when you are submitting a prompt which contains sensitive or personal information. Of course, all our customers do. And with Einstein Trust Layer, you will not have to worry. Now let's take an example of service GPT reply recommendations. To build a reply recommendation, the first thing that our system would do is use prompt templates. Now these prompt templates are supported with a number of pre-installed generative AI python libraries like langchain and tiktoken, but you don't have to worry about this part because all of the prompts that you build going forward will be built on our prompt studio, which is our low code, no code UI for constructing, enriching and activating prompts across the flow floor. Now once that prompt is ready, this prompt will typically contain a lot of merge fields. And the merge fields are usually placeholders for things like customer name as an example, their location or any other information which will be relevant to that particular prompt. So using retrieval-augmented generation, what we will do going after that is get this particular data securely from the CRM system, the Data Cloud system and also the Customer 360 profile and added to that particular prompt. There is also an option for vector search where there are text-based outputs required. For example, let's say, like searching knowledge articles and responding to the customer with a summary of that knowledge article or responding to the agent with it. But of course, when we talk about KB articles, it will not be needed for each and every use case. And then we start applying our trust measures. So the first part is that we start using data masking technology. So this particular feature is controlled by the internal teams today and currently is disabled by default, but it speaks of our vision of having readily available prompt-masking in a generative request or response flow. After the prompt is masked and we mask all the customer-sensitive PII data, then we would apply a set of defense heuristics, such as the instruction defense and the post prompting instructions that would further guide you against prompt injection attacks. So what does this mean? So in the previous slides, I spoke about cases where the generative AI systems were actually, with the help of prompt engineering, they were thought false information like the sun is blue or like humans have wings, right? And then we also talked about hallucinations where the AI goes ahead and create fake stories and fake incidents, like the issue that we lawyer faced. Now with prompt defense, you will be able to avoid all of this. And then once that is done, via the LLM Gateway, which will act as a governance layer, it will then go ahead and talk to the third-party AI models. Now when it interacts with this AI model, it will go ahead and also apply certain other security features like zero retention. And all of this data is -- and the communication which is happening with these models is all -- it's all encrypted. So we have a -- when it comes to OpenAI, they have enterprise-grade APIs, which then allow us to ensure that there is zero data retention. But along with that, let's say, if the response has some toxic output, it would then go ahead and inform Salesforce about it without saving that particular data. And if there's any toxic issue, toxicity which has been identified in that particular response, Salesforce will then handle it. So once the model has produced the output and then via intent classification and toxicity and bias detection, we would then identify if there is any data which needs to be withheld before sending it to the customer. And in a scenario where everything is fine, it will then go ahead and demask any of the data which was masked before. And then what will happen is that the user, whether it's an agent or the customer, they would then get an option to either accept or reject that particular solution, which is nothing but a feedback, which will be stored within the system, so that we could then train ahead -- train our AI in the future for better responses. So now we spoke about this particular trust layer, right? And of course, while the previous slide spoke about the architecture and how that prompt will be generated in the back end, let's look at the real results. Now here on the left-hand side, you see that there is one reply which has been generated by Einstein Reply Recommendation for David with -- and for reply to Ashley. Now when he clicks more -- on the More Information button, what happens is that they would then be able to see details about that particular response. So the first one is toxicity, the risk and the accuracy. Now based on this, of course, while the system will automatically take decisions, but it will also help David to then double-click and ensure that everything is right before responding to the customers. And also, you would then be able to test these interactions and see how Einstein would plan in your sandbox environment as well. And finally, you would then be able to govern the access on that through the platform through permission, right? So in this particular case, the AI has recommended to update the contact, but the control is still with David and the decision-making capability is with him to either go ahead and update the contact or probably skip that part. So now let's talk about the Einstein for Service and how the Service Cloud Einstein works. The first step is one standard gateway for calling all the large language models, which is what is provided by Salesforce out of the box. That single gateway will give us a standard set of APIs to interface and to interact with these models and whether they are foundation models from a partner or an internal model which we have developed at Salesforce, all of them would now pass through the same gateway and same set of APIs. So this will then help you in your governance and it will also act as a middleware for all your tasks like prompt engineering. All this together will create a seamless, consistent set of generative AI experiences across use cases and clouds. The next key component to highlight is the Customer 360 Data. With Data Cloud and Customer 360, we have access to relevant data and to build a prompt that are tailored to any use case. So earlier, I mentioned the use of merge fields in the prompts. This is where, again, the Customer 360 Data and Data Cloud will then help to replace those merged fields with contextual data of your customers, which will then go ahead and help in personalization. And all of this is built on our Hyperforce. So the next key component to highlight is the -- we spoke about the Customer 360 Data, and all of this together will then allow us to do things like generate service replies on any channel. It could be your mobile app or it could be a laptop or a desktop. It would also help in writing work summaries based on the conversation data, which also would include e-mails as well. It will also help in drafting knowledge articles and surface relevant and accurate answers. And finally, it will also help in drafting mobile briefings for your field service agents. So they know exactly which task they are going to look at in the entire day and what needs to be done and they could plan their days accordingly. So now let us quickly look at a demo of e-mail summaries, and we will look at how GPT will now be able to summarize all the e-mail interactions which have happened between the customer and the Salesforce agents. [Presentation]

So in the previous demo, we saw how your e-mails would then automatically get summarized with the help of GPT. Now let's move to the next part, right, where we look at knowledge article, service replies and how search answers will help you to resolve the issues faster and how it will help your customers respond to queries even without your agents being involved in that entire conversation. [Presentation]

So now let's look at the Einstein 1 Platform and our Einstein Copilot capabilities. So we spoke about the architecture in detail in the previous slide. Now we are going to focus on two important components of the Einstein Copilot. The Einstein Copilot and the Einstein Copilot Studio. The Einstein Copilot is nothing but a conversational AI assistant, which is available in every app. So whether it's sales, whether it's marketing, whether it's service, or any other app which your agents or your employees are using. And you could also build copilots for your customers, which will then go ahead and automatically respond to the queries. Now to build this particular copilot, we need the Einstein Copilot Studio, which is built on the low code, no code principles. And with drag-and-drop engine, you will then be able to build these copilots. And when we talk about the copilots, the most important part is the skill. And that is the way we augment a copilot or the AI apps and more things that it can do. So as a developer, you would then build a series of building blocks that perform granular task, like retrieving data about the customer, update a record, perform some computational logic. Now these actions together will then make one skill for that particular copilot. So now let's have a look at the demo of copilot and it will first, of course, not only show you how it works, but it will also show you a small glimpse of the entire solution. [Presentation]

Right. And finally, now let's double-click on the Copilot Studio. So we looked at the Einstein Copilot, we looked at how the actions are created. Now we look at the skill builders specifically and how these prompts and skills are created on our -- on the Salesforce platform. [Presentation]

So with this, I would like to end my presentation and open the floor for any questions. All right. I see -- I've seen a few questions come up, and I guess we've already responded to most of them. So if you have any further questions, feel free to reach out to us, feel free to contact our account teams, and we would be happy to answer and then we would, of course, be very happy to get into tailored conversations and how we could help and how generative AI will help your organization transform your services. With this, I would like to thank each and every one of you for joining this webinar, and hope you found this useful. So thanks, everyone, and have a nice day.