How to Build and Deploy Salesforce Agentforce

Key Takeaways

Welcome, ladies. Thank you. Thank you for being here. So, let's start with you, Britney. What was the main problem you were solving and how did Agent Force provide the solution? Sure. So, I was running our business operations team for Amazon Web Services. We were supporting about 900 different builders on the AWS Salesforce platform. I was a bottleneck. Folks had to come to me, get information. It was taking a long time to get it back to them. So, Agent Force is helping everyone find answers to our runbooks, our SOPs, and also query all of our records that are in Salesforce. So, I am no longer the bottleneck. Nice. You get to relax a little bit. >> Yeah. Yeah. I got some time back in my day. Yeah. What about you, Sarah? What problem was Agent Force uniquely designed to solve? Yeah. So, at Engine, we're a travel management platform. So, we get a lot of hotel modifications and cancellations every day. So, our human agents were actually getting about 300 cancellations every single day and it was very repetitive manual work that was bogging down their workflow. And so, we were able to take that use case and and give it to the AI agent. And so, we we built a lot of confidence and trust within that use case and it it helped us scale a lot quicker than, you know, than we were expecting. And so, Eva now handles Eva's our Engine virtual agent. We call her a she and um she is now handling 50% of our uh support cases fully automated end-to-end. And so, not only have we scaled Eva to do other capabilities, but we've scaled to multiple agents across our business. Amazing. Now, as you built these agents and and stood them up and got them in operation, what were some of the biggest learnings and takeaways along the way? So, my biggest takeaway was however much time you think you need to test, you need more. I spent so much time testing myself and with testing center because every so often I would find a little quirk. We have an internal tool. It's called Kingpin. And I found anytime I would mention that to the agent, it would flag it as inappropriate content. So, I had to spend some time training our employee agent to know, "No, this is okay when we when someone asks about this, this is where I want you to go and look for information." So, make sure you spend a lot of time testing every possible scenario that you can well as you're building. Such good insight. I love that. What about you, Sarah? Looking back, is there anything you would have done differently? What did you learn? Yeah, I think the biggest thing that I've learned is just when to be deterministic versus when to let the agent do its thing and actually have agency over its actions. Yeah. Um there are certain things that you want to happen every single time and especially with Agent Script coming out, it makes that a lot more possible, right? There's going to be that determinism when you want it. Um and it it really when you find that balance, it really takes it from oh, this is something that's pretty cool to oh, this is something that we can trust as a business. Yeah. Yeah. Amazing. Now, I want to remind everyone watching if you've got questions for Britney and Sarah about how and what they built with Agent Force, go ahead and enter them in the chat because we will come to those next. Now, I want to ask, was this a team effort what you built? Who else was involved, Britney? So, I am a one-woman show in my org. I am a solo admin supporting this organization of 900 builders, but I was very fortunate. I was partnered up with a solution engineer from Salesforce on the Agent Force team and he was there to help guide me along the way and help me ideate while I built. So, that was incredibly helpful. Wow. What about you, Sarah? Very similar situation. >> Yeah, very similar to Britney. Yeah, it's um one-woman builder inside Engine and so, it's it's been a lot of fun and I've had a lot of help just like Britney has with the forward-deployed engineers, the solution engineers. It's really great to just have someone to, you know, give feedback and keep you up to date with the features that come out with Salesforce because it's changing every single day. And so, just having that partnership really helped excel and helped us scale a lot faster. So, once you got your agent implemented and and you it's up and running, how do you gauge success? Yeah. So, like I said, I was the bottleneck. I think we've all been there. There is the keeper of the knowledge. You send them a Slack and it takes them a long time to get back to you. But now, instead of someone waiting for hours, they get an answer in less than 20 seconds. That's incredible. >> is. So, not only it Game-changing. >> It is. So, I got time back, but so did my builders. So, now my builders have more time to go and do what they're supposed to be doing and that means building, not going and finding information on how to build and how to navigate these processes. Right. But Sarah, what about for you? How do you gauge success? Yeah. We use um CSAT as our number one metric that we're looking at. So, our customer satisfaction. And so, we actually started out with a 3.7 out of five for Eva and we actually bumped that up to a 4.3 out of five in six months. So, we're very excited about that. I mean, when you when you take a step back, you're able to see, "Hey, we're increasing efficiency, but we're also enhancing the customer experience." And so, the combination of that is kind of mind-blowing. Um we couple that with some sentiment analysis. We're able to QA actually 100% of our conversations. So, when you look at the holistic view of CSAT plus that sentiment analysis, you're really getting that whole picture that allows you to tailor your experience with what your customers are are using your agent for. I mean, that must feel really good that you guys solved these the bottleneck and you're you're seeing such great results and kudos to you. Congrats. I'm so excited to hear some of our questions. Are you guys ready to answer some questions from our Salesforce Plus audience? Absolutely. All right. Well, go ahead, everybody. Drop those questions into the chat. We're going to get to those right now. So, um first question, how long did it take you to prepare for Agent Force? Org prep and skilling up? So, I started working on the Agent Blazer program probably in August, but not with the idea that I was actually going to build and deploy an agent. I was doing it because I wanted to be ahead of the curve. I came out of Dreamforce ready to go and by the end of November, I was partnered up with Salesforce and in about 40 hours of build time, we were able to get this agent out the door. Wow. That is so impressive. That's so fast. What about for you, Sarah? Yeah. So, Eva was launched in about 12 days. Uh we had a a consultant firm firm that was able to actually spin up that first use case before I had started Engine and so, it they we really had that foundation that was built within those first 12 days. Um but the real work came in when I needed to actually recreate this agent to do a lot more actions and because the architecture changes so much and at a rapid pace, we actually had to scale down the amount of topics that we had and change some of those foundation level um infrastructure items so that we could keep up with, you know, the latency concerns and some of these other technical issues that you know, weren't apparent to us when we first started the build. Awesome. All right. Well, our next question is, what advice would you give an admin or builder starting their first real Agent Force implementation? I would say start simple. Don't try to boil the ocean. Don't pick a problem that you haven't solved yet and then layer in agents on top of it. So, I think an employee agent is a great place to start because it's doing the things that your users are already doing and they're going to love you for it because they don't have to do all of those tasks manually anymore. Go for the low-hanging fruit. >> Yes, go for the low-hanging [laughter] fruit. What about you? What advice would you give, Sarah? >> Yeah, Britney said it beautifully. I think just piggybacking off of that, starting small, picking that use case that you can then iterate on later. Um but some other advice would just be to jump in and dive in and get your feet wet. Take action. Cuz right? Action is how you're going to get that feedback from everyone and feedback is how you're going to get better. So, ultimately, it seems like scary to to do that and take that first step, but it's the best thing that you can do. Awesome. All right. Our next question, how long did it take you to build, test, and launch? So, like I said, combined, it was about 40 hours of time to get everything together, test it, and then launch it. >> Okay. Yeah, and for for us, for Eva, again, that 12-day was our benchmark, but again, iterations are constantly happening. It doesn't stop once that agent hits production. It's Right. next comes the QA and the governance, the iteration of prompts, the additional actions that you're going to add to your agent. And so, it's it's that's the fun part. It's never ending. You get to always grow your agent. Leads to our next question. How did you refine the experience for customers after the initial build? Like, what does that process look like? Yeah. So, for us, that main thing, the first thing we did, we went from about 35 topics down to eight topics. And this really helped us because our agent was hallucinating a ton because it had we were kind of treating it like a chatbot, right? We were developing like niche paths for every outcome that could happen. We weren't giving the agent the agency that we wanted it to have. We were trying to determine it in the wrong ways. And so, we really trimmed down that to to keep the latency down and we saw exponential uh upside to that. Such a great tip. >> Yeah. What about for you? Feedback. When I hear from people that, you know, the agent gave them a wrong answer or I wish the agent could do this, then it's time to dive in and that's when you start that process. Like Sarah said, it's a cycle. So, you ideate and you start implementing and testing and deploying again. Amazing. And uh so, how much customization did you need to do for the employee agent versus out of the box? So, my employee agent was mostly out of the box. I spent most of my time with the instructions and the scope and all of those written things that you have to do. I did have to do some flows. Kind of like Sarah said, you have to think about should this be deterministic or should I let the agent do its thing. And there were a couple of times when we found, you know, the agent continues to hallucinate or it's pulling the wrong information, so then we would build a flow and add it in there. What about you? Yeah, so I I absolutely agree with Brittany what Brittany said, and I just think that the evolution of going from like that single agent to multiple agent architecture is something that we have found to to really implement into our organization. Okay, I think we have time for one last question. Were you really able to do this with all clicks and no developer/Apex? I was. It's is all clicks, no code, everything is out of Agent Builder and Prompts Builder with flow for me. Wow, what about you, Sarah? >> bit more customizations because we are calling some API endpoints, but it's amazing what Salesforce has to offer you with the foundational blocks. So, you're actually able to use a lot of the out-of-the-box features, and as you want to build those customizations or use something other than clicks and use some code pro code tools, you're able to do that, but a lot quicker because you can already take what they've offered you and just build on top of it. >> Yeah. Yeah. So, my last question is, what do you do with all this time that you saved? Do you end up building more agents? Yeah. I'm trying to help the agents build some agents, right? >> now agents building agents. >> Well, Brittany and Sarah, thank you so much for joining us. Two amazing stories of Agent Force in the Wild, and thank you so much for using your experience to answer our Agent Blazer Community questions. If you'd like to learn more about what you can build with Agent Force, scan the code on screen or head to the link salesforce.com/agentforce. >> [music] >> Hey! Hey!