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

[Presentation] Hello, everyone, and welcome to today's session entitled Easy AI - Generative Sales Emails. Thank you all so much for joining us today. My name is Simon Carolin from the Solution Engineering team here at Salesforce. Before we begin, I'd like to share a few quick notes about our webinar platform. Today's webinar will be available on demand after we wrap up, and you can access it through the URL you're currently on. Please note that the slides will advance automatically throughout the presentation. You can enlarge the slide or media button so by clicking the square button located at the top right-hand corner of the widget. If you require technical assistance, simply click the Help, which is located on the right side of the console within the Engagement tool. We've also provided some additional resources accessible through the Resources window to the left of the slides, where you can find related content. Lastly, we encourage you to submit your questions at any time throughout the presentation using the Ask Question widget at the right side of the console. We'll do our best to answer as many questions as we can at the end of the presentation. With that, let's get started. First and foremost, we just have to bring it to your attention that all purchasing decisions should be based on generally available software. Everything that will be shown today is currently generally available. Before we start, I'd like to begin with acknowledging the traditional owners of the land on which we meet today, the Gadigal people of the Eora Nation, and recognize their continuing connection to land, waters and culture. Would also like to pay my respects to elders past, present and emerging. Introducing our panel today, myself, Simon Carolin, already introduced myself, and also Tiffany Maruta. And our agenda for today, first and foremost, we'll have a look at how Generative Sales Emails work. Then we'll have a look at how to set it up and get started. We'll look at the future demonstration. And then towards the end of the webinar, we'll look at some resources, and we'll have our Q&A to round us out. Companies are battling inflation, supply chain bottlenecks, regulatory uncertainty and political disruption. Amongst all this uncertainty, sales teams are under pressure to keep producing and stay productive. A recent Gartner survey said that 90% of sellers are actually burned out. They have a higher intent to leave and lower quota attainment. Sellers are juggling many hats as they adapt to changes in consumer demand, evolving buyer preferences and new ways of selling. Approximately 70% of the seller's time is spent doing nonselling tasks like admin or getting across the ever-increasing complexities of products in the modern workplace. One way to reduce seller burnout and drive productivity is to look for ways to improve business processes. Many sales teams have turned to data to help them to adapt their sales strategies. The Gartner Future of Sales 2025 report reveals that 60% of B2B sales organizations will transition from experience and intuition-based selling to pure data-driven selling by 2025. Are you on a path to be one of those organizations? Data-driven sales teams seek growth and accelerate these deal cycles, and 83% of sales teams with AI saw revenue growth. Let's now take a look at one way to leverage data as well as AI to improve seller effectiveness and reduce the time spent on nonselling tasks. Introducing Sales Emails, one of the many features of Einstein for Sales. How many emails are sent very day? Absolutely tons. Imagine if we could select a type of email I want to send, introduce myself, for instance, nudge for follow-up, schedule a meeting. An Agentforce Assistant will write that email for me based on CRM context, contact, accounts, previous business, et cetera. How much time would that save me? When was the last time you click compose and sat with that empty email window, willing the email content to write itself? Sales Emails help you get over what I call blank email window inertia by presenting great email content based on your initial selection or instruction. Now there are 2 ways to generate Sales Emails within Salesforce. Emails can be generated by Einstein in the email composer in Salesforce or even in Outlook or Gmail integration panels directly. This gives sellers the choice to generate emails in their composer of choice. Sellers can also utilize the Agentforce Assistant inside of Salesforce and instruct it to generate an email based on their stipulated criteria. You can then ask the assistant to further enhance the generated email. For example, you may want the email to be more detailed or to have a call to action at the end of the email. Let's now take a look at the concepts around how Sales Emails work. When we want generative AI to write an email for us, we write out a set of instructions as to what's the large language model or LLM, i.e., the brain, can generate. We call this a prompt. Now a prompt can be simple. As an example, I could ask the Agentforce Assistant to write me a Sales Email asking for a meeting. Now the LLM will generate an email, but the content will not be great, and it will be pretty vague. This is an example of no or limited grounding of the prompt. Grounding refers to the additional context and information that is sent to the large language model for it to generate better-quality emails. With Salesforce, you can ground your prompts in your rich CRM data. You've collected all this rich data in Salesforce over the years, now you may as well use it to create the best emails you've ever seen. Salesforce provides a number of out-of-the-box templates such as Introduction, as an example, to get you started with your generative email journey. Now the real secret sauce lies within the unique tool within the Salesforce platform called Prompt Builder. With our low-code Prompt Builder, sales teams can create their own prompts customized to their precise business and processes. You can build, test and fine-tune trusted AI prompts within the Einstein Trust Layer that masks any sensitive information, and it can also limit bias and toxicity. You can ground your AI prompts with Dynamics CRM data using merge fields and Salesforce Flow for Dynamic data input involving complex layer, and we will be covering that both in our demo shortly. But first, I made mention of the Einstein Trust Layer in a previous slide, and I would like to chat briefly about this hugely important aspect about Salesforce to our strategy. Our AI platform is delivered with the collection of safety guardrails that we call the Einstein Trust Layer. This is how we enable all of you to get those powerful benefits of AI without increasing risk to your business. Trust is our #1 value, so we want to take you in-depth of how this works. Everything starts with that prompt, as we mentioned previously, the instruction you send to the large language model. These prompts can be grounded and enriched with your unified data through the power of some of our tools, such as Data Cloud. The output of the prompt depends on permission to access the relevant data, and we make that possible through secure data retrieval. We can also dynamically ground prompts through Flows, Apex code, MuleSoft API codes, pulling you data that is relevant to customer and specific scenarios, including performing semantic searches against your knowledge base. And we protect these prompts by masking their personal identifiable information and cleaning instruction defenses that steer the model towards relevant and safe outputs. The prompt is sent to your choice of large language model. And if that large language model sits behind an API, we enforce zero data retention so that your data is never stored outside of Salesforce. On the way back to Salesforce, we employ toxicity detection to ensure the prompt is usable and won't put your brand at risk. And all of this, the prompt, the response, the scoring is logged in a secure order trail, and it will power reinforcement learning based on your CRM outcomes for models that we host or whose weights you own. But enough of these slides, let's take a look at this all in action. Now our first demo, we can have a look at how to set up Sales Email within Salesforce and also a quick little demo of how I think you can use it immediately out of the box.

Hi, there. My name is Tiffany Maruta, and I'm a Solution Engineer here at Salesforce. Today, we're going to see a demonstration of how to set up Sales Emails, and then we'll also see it in action. So first up, let's walk through how we can set up Sales Emails for your team. But before we get started, be sure to check that you and your team have the Einstein for Sales license, which is an add-on or included, depending on your sales card addition. First, let's navigate to set up to turn the Einstein generative AI preference. Use a Quick Find box to search for Einstein setup and click select. Switch the toggle to on, and you've enabled Gen AI in the org. Next, search for Einstein for Sales in the Quick Find box, then select it. Turn on Sales Emails and enable add email prompt instructions. This step can take a few minutes before it's ready to use. Okay. Let's now assign the Einstein Sales Emails permission set to the team. I can select 1 or a few users at a time. But for now, I'm just going to add it for Cindy. Next, we want to let our sales reps take advantage of this feature right from the Inbox. When you use Salesforce email integrations for Outlook or Gmail, the Sales Email component is automatically added to the default layer. However, if you've customized these, you'll need to add this. Simply edit and drag the Einstein Sales Emails component onto the page. So now that we set up Sales Emails, let's get started and see it in action. Sales Emails empowers your busy sales reps with a tool that enables them to quickly draft up personalized and informed email content for their prospects and customers. It works by leveraging generative AI through prompts and grounding this in your sales and CRM data to generate tailored emails that resonates, all of which takes place within the secure boundaries of the Einstein Trust Layer. All right. So today, Cindy has a number of leads, which have been assigned to her and are waiting for how to reach out. Instead of having to write an internal email to run from scratch, thinking about what to say, how to make it catchy and potentially facing the dreaded writers' block, she can now leverage Sales Emails to help her with this process. Out of the box, there are standard templates to help her get started. Here, we can see 3 templates for leads, and there's more available for contacts and personal accounts. Let's select introduce yourself and select the product that Ron is interested in. Great. That's a neat starting point. And from here, Cindy can make any edits to the email before she hit send. Another way that Salesforce can use this is by drafting up their own instructions directly here. Now some businesses may only want their team members to use premade instructions, and that's fine, too. We can disable this in the setup.

All right. So let's have a poll. Do you currently utilize an AI-generated tool for email creation? All right. We've got 4 choices over there, and we'll give you 30 seconds or so to make a selection. [Voting]

All right. So let's take a look at the results of our poll. So first option was, No, but we have been considering, 16.7%. No, we manually create email content, it looks like the vast majority of people, almost 60% there. Yes, we use it regularly, so just under 14%. And yes, but we do not use it often, sort of just above 10%. So interesting results there. In our second video, we're going to cover off our prompt template builder, and we're going to see 2 examples which basically use our merge fields in Salesforce, and then our second example, which cover off slightly more sort of detailed functionality around how we use Salesforce Flow in the process as well.

Now those out-of-the-box prompts can be very useful if you're looking to generate a simple email to send. But generative AI technology really shines through when you properly ground the prompt that is sent to the LLM in order to give you an even richer and personalized generated content. So let's see how we can start to customize the generative email content. And let's go back to Ron's lead record, where he's interested in renewable energy solutions from Sirius. Cindy is going to send out a generative email using Sales Emails, except this time, she's going to select a template from the custom section. Now custom templates are prompt templates that have been created by our Salesforce admin or even by you, if you have the correct permissions assigned. Just like that, Cindy's created with a rich Sirius-specific email highlighting the why Sirius pitch to Ron. With 1 or 2 minor edits, she's ready to send off this email to Ron. Let's now take a look behind the curtain to understand how this came about. Prompt Builder is one of the tools included as part of Salesforce's generative AI tool set. It can achieve a number of generative tasks. But today, we are just looking at one use case for, namely, Sales Email templates. Okay. So here, we have opened a self-introduction template, and now let's break down to see what makes up one of these. There are 3 main areas: the prompt template workspace, preview area and configuration. Now the workspace is where you will type out the instructions for the prompt and select resources, which will be used to bringing dynamic data from Salesforce specific to the lead or contact. Now the first thing to remember is that writing prompts is a skill set in itself. It is definitely something that needs to be understood and practiced in order to get the most out of Gen AI technology. For anyone who's new to prompt writing, we recommend that you check out this trial, get started with prompts and Prompt Builder on Trailhead today. Back to our template, we start out by explaining who Sirius Solar is to the LLM to ground his understanding of who Sirius is as a company. This is crucial as the LLM has no prior knowledge of Sirius as a company, other than what is on the public Internet. We then move to stating the goal. In this case, it's to write a compelling sales outreach email for a prospect to purchase Sirius solutions. Next, we proceed with instructions that the LLM must follow. Things like using a framework to develop a full paragraph outreach email by stating an intra paragraph why change, why now, and finally, a why ask paragraph. Further instructions are then declared. This is very important as it keeps the LLM on track and reduces hallucinations. And finally, we declare the tone of the email to bring up the brand voice of your company. You will probably have noticed the blue merge fields that's scattered throughout the prompt template. As mentioned earlier on, these merge fields bring in dynamic data straight from Salesforce that is relevant to the lead or contact that the template is running for. If I click into the Resource box, I can select sender or recipient and then pull in fields from Salesforce for either. Now the second component of Prompt Builder is a preview area. As the name suggests, it's used to provide the output of the prompt template as well as what is returned from the LLM. Let's now run this template for Ron. Go ahead and click preview. Now in the resolution box, we have the grounded template before it gets set to the LLM. We can see how Salesforce has filled in the merge fields with data relating to Cindy as sales rep and Ron, our lead. And the response section is what actually comes back from the LLM and, in our case, what we will show in the email editor. It's important to remember that with generative AI, the response can be subtly different each time you execute it, even with identical prompt. And that's because this is how LLMs work. And the third area is the configuration area. This is where we select the AI model type, either standard or bring your own model. If you select standard, you have access to a large number of options from LLM providers like OpenAI and Anthropic. You can also pick different types of models within a vendor, depending on your use case. Now we are going to build on what we've learned, thus far, and add some more detailed information to one of our prompt templates, courtesy of Salesforce's process automation tool flows. So Lauren, an existing customer, is currently engaged in a sales process with our sales rep, Cindy, to purchase additional solutions. The deal slowed down, and Cindy is looking to reignite the opportunity. But instead of sending Lauren a basic follow-up, she wants to personalize the message and tailor it based on the interactions to date. Cindy selects the custom email and inputs the product that Lauren's interested in. Now what's different this time is that the LLM has generated an email that shows empathy towards Lauren. So how did we do this? As we open up the prompt template, you may have spotted some green text in the prompt on this occasion. In this example, we are using Salesforce Flows to retrieve data from Salesforce and utilizing a tool called Record Snapshot, which adds the fields from the page layout in the prompt resolution. Now the Flow extracts a wealth of information about Lauren, such as her sales opportunities, open cases and communications from the last 3 months, including emails, log calls, meetings and tasks. Now this information, along with the Record Snapshot, is appended to the prompt template and then sent to the LLM, which is then able to review the current state of Lauren as a customer. If Lauren has outstanding open cases, the generative email content makes the reference to this as a sole point and attempts to smooth things over. It's like having someone doing the automatic research for you, and we think it's pretty cool. Today, we've explored how AI can supercharge sales productivity by personalizing email outreach. Extending sales emails with Prompt Builder, we can craft targeted, brand-aligned messages that meet unique business needs and resonates with your audience. Ready to see the difference? Get started with Sales Emails and empower your teams to connect better today.

Thanks for that great demo, Tiff. All right. So how do you get your hands on this technology? So what we would recommend that you do is get in touch with your Salesforce account executive. And what you're looking for is Einstein for Sales. And sales emails is essentially just one component of Einstein for Sales. As you can see, there are quite a number of other fantastic options available with it. But it is a well-rounded AI product that doesn't just give you predictive AI, but full access to that generative AI suite as well. All right. And with that, you'll find those resources on the ON24 console, as I mentioned at the beginning of the webinar, and feel free to leverage those to find out more information. All right. So we've reached that time, Q&A. So please fire away with your questions, and we will do our best to cover them off. Please don't be shy.

All right. We have a question from the audience. Can generative emails be used for email replies or just new emails? All right. So with regards to that, what we've shown you today is essentially intended for new emails. We do, however, have, as part of our Salesforce suite of products, we do have, as part of our service card offering, Einstein for Service, we do have the ability to automate replies in chat context for service cases and the like as well. But in this particular case, it's a sales outreach kind of product and basically allows for new correspondents like that. All right. We have a question here as well. How are the prompts licensed? All right. That's a fantastic question. And essentially, when you purchase your Einstein for Sales license, you get a certain number of what we call Einstein requests with that license. Now every time you make a call to a large language model, as some people would know, those are third-party products, and they cost money to use effectively. So what we do is we give you these Einstein requests, which cover off that aspect of it. And you can then use those Einstein requests how do you see fit. Now a typical Sales Email would use one of these Einstein requests. And then, essentially, these can be tracked as well from a customer perspective to see how you are tracking with regards to your allocation. Now the amount of Einstein requests also pooled across the organization. As an example, if you had 20 users of Salesforce, you'd have your allocation per person added together across the entire organization to pool that allotment, essentially. We have a question here from Carla. She says what permission is required for Prompt Building? All right. Tiff did cover that off in her initial video, with a set up there. The one specifically for Sales Emails, I believe it's literally called Sales Emails. It might be Einstein for Sales Emails, off the top of my head, but it will be something along those lines, and that can be assigned to the users that require it. It doesn't have to be used across the entire organization. Quite a lot of people just pilot this initially and give it to 1 or 2 people to play around with, and you can do that obviously with the permission sets in Salesforce. A question from Ian. Can an email be sent out to multiple recipients or just individuals? At this stage, it is via the composer within the activity time line, Ian, and obviously, within Outlook or Gmail as well using the other integrations we have there. But it is, at this stage, intended for 1:1 correspondence. What we do have as well is our Marketing Cloud products, those also more geared to one-is-to-many kind of communications, and we have all sorts of generative AI smarts around copy creation, everything around that sort of things as well. Sara writes, what are some best practices for using AI to personalize Sales Emails effectively while maintaining authenticity and avoiding a robotic tone? It's a great question, Sarah. And that sort of goes back a little bit to what Tiff was mentioning with regards to prompt engineering. And I think a lot of us would have a bit of a crash course with regards to how to get the large language models to work in the best way possible for us, and it is a new skill set essentially as well. What I would highly recommend, though, that Trailhead trail that Tiff referred to goes into quite a lot of detail with regards to the best practices about how to construct certain types of communications and to get the best out of those large language models. So I would definitely give that a read. It's not too detailed, but it's definitely helped me with regards to utilizing this technology. Alberto has a question. Will this functionality be used in a mobile setting? At this stage, Alberto, that's absolutely correct. So within Salesforce Mobile, you have the ability to use the Agentforce Assistant, formerly known as Einstein Copilot. So within that, you saw it was in the demo. Tiff actually mentioned the option with regards to actually just then and there typing a prompt directly into the area within the Agentforce Assistant. And you can then just get an email done sort of without having to create a template beforehand. And you can do exactly the same thing within Salesforce Mobile as well. All right. A question from Matt. How can Einstein be used to support responses to technical product questions? That's a fantastic one, Matt. And this is another area where the Salesforce platform really shines. And one of the things that's making this all the more possible is our Data Cloud product. Now for a lot of businesses out there, you generally have other systems beside Salesforce as well, and you can have valuable company information in those other systems. Now with regards to generative AI, AI in general, it's always best to point the AI to a collected data set, essentially, so that it's looking at one sort of single source of truthful data. And our Data Cloud product, essentially, enables that. Now with regards to those prompt templates that Tiff showed you, you actually have the ability to pull data directly from Data Cloud straight into those prompts as well. And one of the really exciting new features of Data Cloud is what we call vector databases. And vector databases allow us to bring unstructured data into Data Cloud. So think of things like chat conversations. It could be -- we could even upload the video from this webinar into Data Cloud. And that transcription, the text, obviously, of this webinar, you could search on that information as grounding for prompt outputs and Agentforce outputs and things like that as well. One of the things we always advocate as well is storing important technical information, particularly if you need to send this on to customers within Salesforce knowledge as well because you can refer to that in prompt templates as well. As well as Agentforce output as well with regards to getting up the organization to have autonomous agents that takes some of the load off of your service agents. And the Agentforce platform is essentially not just a service as well. It can be used across the organization for a variety of purposes. A question here. Can internal processes now be fed to complement the rules and select the appropriate response to an email? Well, I think this is also where you can use Salesforce Flow as well. So Salesforce Flow really is your declarative version of your business processes, right? So by declarative, I mean, we're not to write code, or clicks and drag and drop tools and things like that. And you can make use of Salesforce Flow, either in a basic or very, very advanced ways in order to achieve that sort of outcome. You can serve things like flows with inflows, right, if you really want to get fancy, but it's all possible within the Salesforce platform as well. A question from Ashley. How do I know if my company has Einstein for Sales? Good question. The first portal I'd go to is your Salesforce administrator, and they'd be able to go into the console and look to see if you have purchased those permission set licenses, which enable this functionality. Otherwise, feel free to reach out to your Salesforce account executives, and they'll be able to look at your account in our Salesforce, which will be able to tell us immediately if you have purchased that. And thanks, everyone, for these questions. They're really good questions. Keep them coming. Don't be shy. All right. It looks like questions are up. Like I said, thank you very much for your time today. I know time is precious at this time of the year, and we really appreciate you joining us for this webinar. If there's any other questions that come to mind, please feel free to reach out to your Salesforce account executive. And we'll be really happy to help you on your journey towards leveraging this fantastic technology to potentially make your lives a lot easier and more efficient. Thanks once again, and enjoy the rest of your day.