How to Build Agentic Revenue Solutions & Sales Agents

Welcome to Dreamforce and thank you for taking time. Here we're going to showcase how Salesforce builds agentic revenue solutions. I'm Sanjeev Arora. I lead the revenue technology at Salesforce and I have with me Suvra, who is the product director for CPQ and Diwansh, the senior engineer in the same team. We are very excited to showcase how Salesforce is building revenue solutions. We are customer Zero. What does that mean? We are the first adopter, first tester of our systems. We work very closely with our product team to ensure what we are delivering has been tested and is available to customer after customer Zero's testing. Today we're going to talk about the agentic journey for revenue systems. We're going to talk about process flow. We have a small demo for you where we can showcase you how we have increased the productivity of our sellers. And last but not the least, we'll share the challenges we encountered while building this agentic journey. Salesforce has grown over the years with various merger and acquisitions and we have added multiple selling models. This really complicates our landscape and we took a stab into simplifying our lead to cash processes and build the Agentix solution over it. So we are moving away from UI and click based interactions to a more conversational experience. This is the transformation we are building here at Salesforce using Agent Force. This really supplements our productivity for our sellers as well as our operational staff and agents take care of the rest. So this is the transformation. We're going to take a deeper dive with Suvra who can talk through the agent process, how we followed as well as a demo on coating and deal agent.

So I'm going to walk you through the example of how we made our seller's life easy using agentforge by building the deal agent and the quoting agent. It starts with Ideate. Basically it starts with our users. What are their problems so we understand their use cases. We design our agent, we configure it using Agent Builder and then we run rigorous testing using our testing center. Once we are satisfied with the test results, what we're gonna do is we are going to deploy it using our Salesforce CLI or DevOps Center. Once it is deployed, the story does not end there. It is a cycle of continuous improvement by testing, tracking and adopting. And this particular user feedback loop helps us figure out what is the next feature is going to be. So now that I have given you the introduction, let's jump into the real example of the deal agent. Now as I told you before, we start with the user journey, the user problem. What does sellers do? They sell, right? Simple. But if it is a new logo, it is always we have our entire catalog product catalog to sell. However, most of our customers are returning customers for whom we need to do our upsell, cross sell cash and we need to renew. And in order to sell, the first thing that comes to mind is what is there to sell and how to know that. For that we need to find out the customer footprint. Now sellers are selling through multiple different channel, sometimes partner led, sometimes self service based and sometimes direct. And it is being implemented like all the products are getting provisioned across multiple different orgs across multiple different accounts. In order to do that sellers what they do, they actually go ahead and log a case to our dealdeck support and it takes more than three to four days. It used to take more than three to four days to bring that footprint report to create it and share it with a customer. But think about a seller trying to close their quota at the year end or a quarter end. What do they do? They do not have time for that. So now we understand the problem of the sellers. They need fast accurate customer information across the multiple different orgs that they have purchased through. We decided to build an agent which is going to give a conversational experience where they can ask follow up questions because they do not really want to log another case if they have any questions based on the footprint. So what we do is we went ahead and built the agent using our agent Force builder or Agent Builder and we connected all our existing deterministic logic like Apex Flows and APIs. And that's where the configuration comes in. And once the configuration is done, we are testing it for hallucinations, for topic conflicts and for anomalies using our testing center. Now we are satisfied with our test results. What do we do is we go ahead and deploy it in production in seconds. A complete customer footprint across multiple different accounts. If there are subsidiaries or of one parent account, you will be getting in seconds. Wherever and whenever your seller does not need to be in Salesforce, he can be in any Salesforce page account page, opportunity page, contact page, or he's probably having breakfast with a customer talking about a potential deal and the footprint report is ready at their fingerprint through Slack. Yes, we have made the agent available in Slack too. Now that they are ready, they know what exactly to sell to the customer. It is time for quoting and guess what? What you are using for quoting again. Agent force to rescue. But how do we do that? For that my friend Devansh over to you.

Thank you, Shubhra. So let's talk about how the agentforce works behind the scenes and what is the process flow like for coding agent. So let's say I want to add a 10% discount on one of the code lines in the code and I'm talking to the agent. I send that message to the agent and the first thing it needs to do is identify the right topic. Once it has the right topic, it needs to read through the instructions. There's so many actions that are available in AgentForce. So the instructions actually tie that prompt back to the action and it's actually the APIs that are connected with these actions that are making the magic happen on the Salesforce platform. And once the job is done, it will return the response back to the user saying that, hey, your 10% discount has been applied. So that is the process flow of the coding agent. Now let's jump into the demo and see how a customer or seller interacts with the agent to find out about their customer footprint and how can they action on the insights that are generated by a coding agent. Over to you, Shubhra.

Okay, so now you're going to see our Salesforce environment. This is the agent window. And this is where a seller can in natural language ask what is the customer footprint for one particular account. Now he's asking for the footprint for the account called Acme. And what it is doing is AgentForce is looking through the entire environment to find the particular account called Acme. But there can be multiple account with the same name. So it is giving option to the seller to select which account they are looking for the footprint for. Now that seller has selected the right account. The Agent force is now looking for the account footprint for the account Acme. It is giving the details, it is churning data. It is pulling from all the different sources of truth and then creating a report. And let's see how long it takes. It is still retrieving data.

And.

Right there, how many days it took probably a few seconds. And this is the G sheet link which has the entire customer footprint report. And this is the power of Agent force pulling data through the entire org just in a few seconds and letting the agent know what to sell. Now it's time for quoting.

Right? Thank you, Shuvra. So now we have an opportunity with these two products that Shuvra just talked about from the customer footprint. And now I want to create a quote with start date today and term 12 months. I also want to carry these opportunity lines over to the quote and also add a 10% discount to the sales cloud product. So let's see what the coding agent does. Looks like it's fetching the code line item details and updating them. And here we go. We have the code created. Let's navigate to this code. And we can see that both the products were actually there on opportunity and now they're there on the code and are converted as code line items. It even updated the discount as 10%. Now for my next use case, I want to create three codes with service cloud product and I want to incrementally add the discount by 5%. So the first quote will have a discount of 10%, the second quote will have a discount of 15% and the third quote will have a discount of 20%. Now this is actually one of the requirements that was given to us by account executives because they initially create multiple draft quotes for our customers. So let's see what the coding agent does. We packed all of our requirements in one single prompt and sent it to the agent. And there we go. We have three new quotes created for the customer right here. If we go to the first quote, we can see that the product is added with a 10% discount. Second quote will have the product added with a 15% discount and the third quote will have the product added with A 20% discount. And with these features, we were able to cut down 75% of our sellers time. That's a lot of time they can use to develop relationships now with that. This is the power of agentforce with codetocache and we will end our demo here. But before we end the presentation, we do want to talk about certain challenges that we faced while building this agent. The first one is guardrails, how to ensure that agent never creates incorrect data. There's many use cases when our users are entering incorrect information. Whether it's by mistake or it's a typo. We don't want to add products with negative quantities or negative discounts. So what we did was we added validations in our custom flows, apex and actions to make sure that these errors are caught earlier on in the process and user always has an optimal user experience. The second one is testing how do we accurately measure the quality of phenation. This is the first time we are building large applications at an agentic enterprise level. And we want to make sure that the user experience is great and the quality is absolutely perfect. And luckily for us, as Shuvra mentioned, we have the testing center available out of the box where we can test bulk utterances and it can exactly Tell us which topic and which action this utterance will execute. And the last one is security. How do we ensure the right access for our users? Now in order to solve that problem we created and assigned custom permissions for our users so that they have the right access to the actions and write access to the entire agents. And now over to Sanjeev to talk about our agentic journey and what's more for revenue.

So how amazing is this, isn't it? You can see the example where we tell the agent what to do created three quotes with 10%, 20% 30% discount. So let me take you through the quoting agent journey. We started with a very simple MVP use case where we asked to create the quote and the quote lines. This was done amazingly in three weeks. This was our basic MVP use case of quote agent. We followed by updating this agent with updating the quote lines and adding the product search feature. And now this month we have launched the most wanted feature which our sellers were looking for which is the pricing guidance and intelligent approvals. So this has been very effective in increasing the productivity of our sellers and making the whole operational experience even better. And this is not the end of the journey. We have picked some special use cases for our revenue systems as part of our code to cash transformation where we are helping our deal desktop to build efficiencies with quote and pricing processes with these agents as well as our collection teams is being helped with payments and reconciliation. Last but not the least, the self service agent is really helping our operational staff with case deflection and other improvements that we are trying to bring in. Thank you and have a great rest of the day and great reinforce.