Speaker 0: Alright. Hello, MarDreamin’, and welcome, everyone. Uh, we’re super excited to have you all joining us today. My name is Cecil Natulli from Sercante, and I’ll be monitoring today’s session. Uh, before we get started, I just have a few housekeeping items to cover. Uh, yes, of course, these sessions are going to be recorded and will be available on demand directly after the event. Uh, we’ll also be following up with them via email as well. During the session, if you have any questions, please feel free to post in the q and a tab above. I’ll be monitoring that and answering as we go along. Um, and we may even have time for q and a at the end. Lastly, use the chat. We’re here. There’s emojis, gifts, and so much more. We wanna hear from you about the content you’re receiving. Alright? So now let’s go ahead and get started. I’d like to in introduce you all to our amazing speaker, Heather Rinke, uh, who has an awesome session today ready for us about the five tips to get started with Agent Force. So Heather, handing it off to you.

Speaker 1: Thanks, Cecil. So I’m really excited to be here. I’m Heather Rinke, um, Salesforce product director at Circante, and, um, I’m really excited to be talking to you a little bit about Agentforce. Before we get started, um, I wanna, uh, take a moment to thank our awesome amazing sponsors, um, without which this session would not be available or would not be, um, would not be here. And, um, I would encourage you to, um, check out these sponsors, um, throughout the conference. You’ll see, like, a banner, um, at the top of the event, which allows you to go and, um, check them out. So I really encourage you to to do that.

So we’re gonna take a a brief step back in time a little bit, um, to talk about, um, AI and the and how it’s changed over the last few years. Um, so first, we, um, when we saw Sales Force talk a lot about AI functionality, they did it mostly, um, talking around predictive AI. And that allowed us allowed for us to have data driven recommendations and insights that helped us focus and prioritize on what to do. So things like scoring, predictions, and bots. Then, um, a couple years ago, we saw g chat gbt take off and the influx of generative AI capabilities. Um, these were, um, AI assistant features that helped to, um, helped people to flow work, um, to take on, um, individual tasks when needed. And now we’re starting with, um, autonomous AI, and this is where, um, agents act as digital workers doing tasks without human intervention. And at Dreamforce this year that that was the big theme, autonomous AI and the introduction of Agent Force.

So a just to level set a little bit, I’m gonna talk a couple minutes about what Agent Force is before we start getting into the tips. What Agent Force is is, um, it enables autonomous AI agents, uh, to help your business get more work done. So whether it’s launched from a customer facing channel or being in the CRM, helping employees in the flow of work, Um, the idea is to, um, they in to have they increase productivity, reduce team workload by automating routine tasks, and assisting with complex ones. There’s out of the box agents, or you can create your own. You can connect agents to enterprise data, making those, um, interactions relevant and accurate. And, um, you can utilize some of the automated processes that you probably already have in your organization like Flow, Apex, APIs, or prompts. An agent force is integrated in the Salesforce platform, which is a, um, a real, uh, which makes it so effective. And this view really encompasses how it it layers across, um, across all the clouds. Those out of the box agents, I’ll mention those, um, uh, coming up, and custom agents you build, they work across those clouds to carry out relevant tasks, utilize data they have access to across the CRM and even in data cloud. And data cloud also supports that trust layer. So fueling the the reasoning and the feedback loop that, um, helps Agent Force, um, uh, continue to be relevant and continue to learn. And all of that is is grounded and is based on the Salesforce platform.

So taking a look a closer look at what’s included in Agent Force, um, you can sort of take sort of look at it in two ways. One, you have out of the box agents, and so these could be autonomous. So, um, agents that work independently and hand off to a human when needed. So we saw this, um, uh, GA, the first, uh, agent to be available was the service agent that can help with case deflection or order management. And there are many more that are, um, that are targeted for the road map, um, coming out, like SDR or personal shopper. And then you have ones that can assist and augment, so helping employees by helping them, um, take on tasks and perform tasks more efficiently and quickly. Um, and there there are several that are on on the road map like sales coach or campaign generation, um, that are coming in the the, um, uh, in the near future. And then you have a set of tools that can help you build your own. So within agent builder, you can, um, create, test, and deploy custom agents using building blocks called topics, which allow you to sort of categorize, um, your actions based on a particular job that needs to get done. So something like deal management can maybe contain some actions that help a sales rep with with their day to day tasks, like maybe finding relevant opportunities or creating to do items. Um, your data, obviously, is a big factor of this and then the actions themselves. So using prompts, flows, Apex, you can do you can pull all this together, um, so that you have those set of actions that an agent can use in order to take take, um, uh, complete tasks. And there are standard actions as I’ve mentioned, and you can create custom, and you can combine these. So you can create you can also customize, um, standard actions as well.

So let’s look at how this all works together. When an when a user or a customer makes a request to an agent, might be on an external website or in the CRM, that agent matches that request to a topic that, um, in order to find the the most relevant job that’s being requested. So the best topic then is selected based on the instructions within that topic. Those those are typically made of natural language that helps guide that agent on how it uses which actions to run, which actions to take, or which clarifying questions to ask. And this is also where you can define guardrails on what the agent should not do or what when it should involve a, um, a human. And then once a topic the best topic is selected, uh, you the agent will then look at the available actions and the instructions available to help determine what’s the action it should take. This could be, again, standard or a custom action. And, um, there the action includes its own instructions both on how guiding what it should and shouldn’t do, and also the inputs and outputs, um, maybe required information that it needs to execute or, uh, what may be needed in order to respond. All of that is done and then, uh, pulled together in order for that agent to generate a response.

So I mentioned standard topics. There are a lot, um, actually, um, looking through even just over the last few weeks, I feel like they keep there are more and more that keep getting added. Um, so there’s there are a number that are sort of standard across the clouds and then there are some that are going to be more that are gonna be that are gonna come with, um, different different clouds, whether it be sales cloud, service cloud, marketing cloud, or commerce. So I mentioned a few. So content creation, this, you know, you can use it to help help users draft or revise text. Um, so things like, you know, helping me draft a new subject line. Um, you can help with generating campaign briefs, helping sales reps to get recommendations on the best approach to close deals based on past data, um, or using something like general CRM, which allows people to allows users to maybe query for, um, for all of the cases related to an account or to summarize an account, um, account based on the data that’s in the system. So there’s a lot of capabilities even in topics, which are comprises of a set of of number of actions. Then you also have standard actions that you can use. So things like creating a close plan, so helping helping an agent define or helping a sales rep define what are the things they need to do in order to close a deal. Or maybe they want to get an activity timeline of maybe all the things that they have on the go today, or the the last activities related to a prospect. There are a lot of capabilities available, and there are a number here here that are included with the other clouds that that, um, that add to this list.

I’ll also mention another major component of Agent Force, which is the reasoning engine. And this is the brains of the agent, helps it reason, learn, and take action. So when it when somebody, um, adds a request, whether it’s a user or a customer, um, it’ll take that request and match it with a topic based on the instructions. It finds and launches the right actions that are relevant in order to respond, and then then it continues and it learns based on the outcomes of the interactions in order to determine what to do next. So when you hear me say the agent does this and it and it looks at this, it’s really Atlas that’s doing all this, the brains.

So what can it do? Um, so here’s a number of different things that it can do. I’m not gonna touch on every single one, but, um, some highlights, um, augmenting and, um, automating, which I’ve already mentioned, helping helping humans do their job faster and more efficiently, um, reducing manual workarounds, um, just helping them get things done faster, um, or things that are repetitive. Um, be able to personalize. So using the data in the CRM, um, helping to craft messages that is relevant to to the customer or, um, to an individual in order to get that right message across. Um, being able to summarize. So I mentioned account summaries, but even just like conversation summaries, knowledge article summarization, um, there’s a lot of capability there. And then also, um, I’ll also mention search and learn. So this is something that, um, I’ve learned about recently about, um, just the ability to search for structured and unstructured data or search within, um, structured and unstructured data can open up a lot of capabilities. Like, imagine, um, searching PDFs or contracts for mention of specific terms or clauses. Um, there’s a lot of things you can do.

What it can’t do though, um, but there are some things I can’t do. It’s not gonna do all the things. Um, so it’s not going to fix all of your data problems. Um, there are tools that can help. This can even provide some assistance with some data data, um, uh, tasks, but it’s not gonna do all it’s not gonna do all of the, um, fixing and all of the cleanup without requiring work or human intervention. And, um, it’s not gonna automate out the things. Um, you know, it’s it’s, you know, it’s there’s still gonna need to be humans involved. Um, they are the most powerful piece of training for AI and, um, will always have a role in that process. And then, um, it’s not gonna possibly solve business problems. It knows what you teach, um, and it must be trained. So, um, it’s it’s constantly gonna rely on that training, um, and because scenarios are always changing. And it’s not gonna maintain itself. Um, you don’t set it and forget it. Um, they will require maintenance. Again, scenarios change, processes change, data changes, so agents must be maintained in order to adapt.

So quick check check on just what’s what’s available today. Um, so if you have Einstein Einstein one edition or Einstein licenses like Einstein for sales, for service, for marketing, you you will have access to this. Um, as of October, the, um, service agent was the first out of the box agent to be available. This is there are more agents on the roadmap to help sales, service, and marketing commerce. Um, but then also the builder and the tools, uh, which helps you, you know, to build your own, to use some of those standard topics and actions, um, and then the testing ability, um, that you see, um, just in this screenshot of the builder helps you to test, um, scenarios in order to make sure that you’re, um, it’s doing what you need it to. This is also available if you’ve enabled the foundations in your org. Um, this should always also be available to you there as well.

Okay. So that’s a little bit about agent force, just in general, just to level set. So now here’s a few tips for getting started.

So first, um, think about your use cases. I mean, I mentioned a number of them, you know, potential general use cases. But now you need to think about how where where it can work in your organization. So a few things to ask. You know, first of all, how are you using your CRM today? Um, what what our processes are is your org currently supporting? Where are you seeing some pain points in your processes? Um, are there areas that you could be automating tasks so your team can function on more valuable things or focus on more valuable things? Are you seeing friction points uh, where or hearing frustrations about, um, employer or employee or customer experience? Or are there things that you wanted to introduce, um, in terms of processes or tasks but haven’t been able to because of team capacity or capability. Um, so those are may hope hopefully, will help tease out some areas to tap into.

Then for each use case, you wanna think about what will be needed to transition to an agent. What do they what do they need to be doing? What is the job they’ll be doing? What are the actions they need to take? And also importantly, what should they not take? So this is important to define so that it keeps the agent in their lane and operating within the use case that you want it to. You wanna then prioritize those use cases based on, you know, what are after looking at all the things involved, um, what is the lowest effort of lowest effort of risk, and that’s gonna bring the highest value. You break that down even. Um, so if it’s something that maybe will take a little bit of time and effort, but maybe there’s something a sub a sub process or subtask within there that, um, where an agent could support that, um, that could that you could tap into earlier. So a few use cases here. Um, so I’ll just touch on these, you know, a couple of these, you know, maybe giving sales reps deal insights and guidance or assisting them with lead nurturing tasks on service, maybe automatic automatically answering common questions with, um, some knowledge or FAQ management or automate returns processing. And then marketing, right, um, helping to generate email content or generate campaign briefs. These are just a few examples, but I’m sure there are others that you can probably think about in your org.

Then you’ve defined yours once you’ve defined your use cases, you’ve, um, you wanna think about, you know, how are you gonna measure success? So things like, what are you know, what’s the actual problem we’re trying to solve? Maybe reduce cases that a service team can handle. Uh, what’s that level of success mean? Maybe reducing agent handling time, um, you know, in terms of, like, KPIs. What are the KPIs that we need to track that goal and how are can we measure that now? So things like average handling time or first response time. Do we have that data in place to measure? Um, and you also wanna make sure that you can measure so you can get a baseline, um, so that you can do a before and after to compare and measure success. So a few examples, um, on things that you can, uh, based on, you know, your use case, maybe you’re checking lead response time or deal velocity, or maybe for marketing, your check your your um, tracking campaign ROI or lead conversion. These are just a handful, but some things to get you thinking.

Next, you wanna assess your data. Your AI is only as good as your data. So for the use cases that you’re thinking about, um, asking what data is needed, um, you know, if you’re doing if you’re focusing on case deflection, what what where would you have knowledge base? Do you have a knowledge base that already in place that can help? Um, what customer data do you need to make sure knowledge is relevant? Um, where is it located? Is it somewhere that an agent could access it? And is it reliable? Uh, or do you have lots of duplicates in there or do you have that single view of the customer? Is your is your data clean and accurate? Is it even updated regularly? Are job titles and company information up to date, for example? And I’ve just added a little placeholder here. There is a workshop coming up on Thursday that tackles this this topic in a lot more detail, um, on data management and thinking about how you how you wanna craft your data in order to prepare for AI. So I encourage you to check that out.

And then you also wanna assess your metadata. I mean, data is really important, but for agents, um, metadata is also important. Uh, the descriptions that you’re putting in your flows and in your variables that your resources that you’re using, those are directly used, um, to populate when you’re create when you’re building, um, agent actions. So you want those descriptions to be clear and concise. You want them to describe what’s what what action is going to be taken. And using naming conventions of your resources will also make it deter easier to determine what the inputs and outputs need to be.

Okay. So tip number four, start small. So this is, uh, uh, probably a no brainer maybe, but I wanted to just call it out. Um, you wanna pick one thing to pilot. Um, it really is the the idea that you wanna focus on what what can give in the most impact in that shortest amount of time. Um, it’s really easy to get caught up in all these things that can be solved with, um, with agents, um, but you may wanna avoid or I would encourage you to avoid, um, starting with complex processes. First, it takes it would take more effort to get off the ground, and then it could potentially the more complexity, the more it’s prone to, um, issues or disruption. Um, so you wanna really think about those quick wins that can move the needle that you can expand on over time. So consider the use cases, the success metrics, the data needed. Um, you know, think about those out of the box agents, um, the topics or actions that available, or your existing automations you can utilize to create your own. And maybe even think about if you were to roll this out, like, is there a way to be able to do this in a small scale, um, within a single team or within a single region? Um, so something to think about there. It really by starting small, you’re more likely to be measure success, um, and this is also better for building excitement, building trust and confidence, um, creating momentum, and getting buy in across the organization.

Tip number five. So let’s say you’ve picked your use case and you’re ready to design and build. This is more tactical but important. You wanna make sure your instructions are clear and solid. So when an agent is triggered, um, your your user and you know, or when your user enters a a question or a request, the agent is using that and comparing that with the instructions of the topics to determine the most relevant topic to respond. Those topics are defined of, um, different, um, um, I’m using the term instructions, but really there’s sort of three areas. One is the classification. This helps an agent determine when to use a topic in the conversation based on the user’s intent, um, the scope, which tells an agent what it’s able to do within a topic, and the instructions, helping an agent decide how to use those actions for different different use cases. And as you can see in the screenshot, like, you can add multiple instructions, which is, um, you know, can be a good idea in order to make sure that you’ve got a clear view of, um, what the agent can and can’t do.

And you also think about your actions. So once that agent selects a topic, uh, the available actions, um, it will check those next and take a look at the instructions within those actions to determine what is the most relevant action to take. And this is done through the action instructions that help determine what your action does and when to use it, and then those input and output instructions, which helps to define what information may be required. So what information that agent needs to make sure it has, maybe get from the customer before before acting, and then what the results of the action is or what the response should be.

So some in so instructions are really foundational for grounding agents. Um, they set the guardrails for how agents should behave, um, and set that context for what they need to do to do their job. Um, a few best practices for writing topic instructions or action instructions, um, would be start simple. Start with maybe the main use case, um, test it, then iterate and test again, um, making sure the instructions are um, or the agents are behaving the way you want them to, um, and there isn’t any conflict between your instructions. Um, you wanna use plain language. So natural language to describe what your action does. Maybe keep it between one and three sentences. Um, include the goal of the action and use cases, any objects or records. And, um, in general, just be more relevant in the detail. The more you vary the language, also something to keep in mind, um, use the stronger your agent can become. So if you’re if you have an action that’s gonna do some querying of records, maybe using, um, get, find, retrieve, identify, you know, multiple different versions in order to make sure that it’s learning, um, and it’s stronger. You wanna avoid industry company jargon. So, you know, pretend that you’re writing this for somebody that doesn’t know your business. So if you’re writing something in the where the act action is to act on a qualified lead, that qualified lead definition is gonna depend, um, on your company. Everybody many companies have different versions of what qualified lead means. So instead, maybe you wanna focus on the data or what the what the actual data says. The lead status is MQL, for example. Um, and you wanna give the agent, um, examples to help them understand. And it’s also not gonna know your business process. So be if if a specific sequence has to happen, call that out. Describe it. Um, call out any conditions that need to happen before an action an action is taken. Um, then you wanna think of all the paths. You wanna go through all the possible permutations to make sure that you’ve got all your bases covered. So if you think about, like, a customer reaching out because they didn’t receive their order, well, you know, you have different order statuses, order shipped, order delayed, etcetera. So if the status is shipped, then what are the then is it in transit or is it delivered? If it’s delivered, well, the customer said they didn’t get it. So do we have the right address or is it stolen? So there’s you can you can see how based on one specific utterance from the customer or a request, um, there could be a number of different paths. You wanna also remember the guardrails. What are the things the agent can’t do? What are the what are the things that you wanna make sure that you prevent, um, any unwanted unwanted responses? Um, and you want a good clear direction on when the agent should route to a human. And then lastly, if you didn’t call out the caps there, you wanna make sure you test. That’s really important. Agent builder gives you the ability to do that, um, to do testing before you activate anything. So that’s you really wanna make sure you test all possible, um, uh, scenarios to make sure it’s behaving the way you expect it.

And there that is five tips for getting started. Um, you wanna identify use cases. You wanna define your success metrics. You wanna assess your data and your metadata. You wanna start small, start with those quick wins, and then you wanna make sure your instructions and, um, your descriptions and your, um, all of your details of your actions and your topics are sound. So obviously, I mean, that’s just a that’s just a highlight, but you wanna get your hands on it. If you have not already, um, I gotta say Salesforce has been really good about making this, um, accessible to people, um, to learn about it. So if you but if you have not already, um, there is a trail to help you get started, um, on it. And there are virtual events to help you workshop and build your first agent with with Salesforce. There’s world tours that will have some of this capability to get hands on as well as content on Salesforce plus. Um, and then there’s also a lot of a lot of AI related, um, sessions to check out at MarDreaming. So, um, today has a lot, um, in store. And, um, day two, I’ll call out that you also have the ability to burst build your first agent, um, at this conference. So that’s pretty exciting too. And with that, I will say thank you.

Speaker 0: Alright. Thank you so much, Heather. Uh, that was an amazing session. I know there was stuff I learned. Um, I hope the attendees have learned stuff too. Um, we do have about three minutes, so it looks like we do have some time for q and a. Um, if you continue to have questions, please post them. Um, we have about two minutes left in in in our session, so please feel free to post them. Um, there were a couple of sessions, Heather. Um, I provided some answers, but if you wanna provide some insights, um, please feel free. So one question that I have, um, from, uh, let’s see, Audrey. Um, Audrey asked, uh, where is it? Do we need to have data cloud in order to leverage agent force effectively, or can you just have CRM and marketing cloud?

Speaker 1: So that’s a that’s gonna depend a little bit on your use case. I mean, the typical question, it depends. Right? So, um, from what we’re hearing from from Salesforce, um, that you you don’t necessarily need to have data cloud implemented to use agent force. I’m just gonna say that as a bare bones. You do need it enabled because Trustlayer needs it in order to be able to capture feedback. That being said, if your use case requires a lot of, um, access to, you know, um, unified data, um, harmonized data, maybe a, um, a a true view of the customer based on lots of different, um, uh, different sources and information, it’s really recommended. Um, but, again, that depends on your use case as well. So, uh, I would I would encourage you to take to give it consideration. It certainly does help make your agents more powerful.

Speaker 0: Yep. And I I think that was along the lines of what I what I said too. And and my apologies. It’s Audra with the a. My apologies about that. I’m just

Speaker 1: gonna finish with you.

Speaker 0: Um, we do have another quick question. Um, is there a way to test Agent Force before deploying into production?

Speaker 1: Oh, yes. Absolutely. So, um, I touched on that a little bit. I didn’t I had one little screenshot, which probably wasn’t very good is so Agent Builder gives you that, um, capability. So, um, when you’re building the agent, you have the ability to to add your topics, add your actions. But then there’s a there’s a whole builder that or a whole testing area where you can actually, you know, enter information or enter, um, uh, requests or queries as if you were that customer or that user, and you then you can see how it’s how the agent is actually looking at the topic, looking at the directions, and executing the actions. So then you can see if it’s doing what you expect and it will iterate it will give you a response. So you can continue to utilize that to test before you even activate.

Speaker 0: Okay. And that that concludes our session. So thank you so much, Heather. Thank you everyone for joining us. Special shout out to our sponsors for their support. Uh, without them, Marjorie, it wouldn’t be possible. So take care, everyone. Enjoy the the sessions that we have throughout the rest of the day.

Speaker 1: Thanks, Cecil. Thanks, everyone.

Speaker 0: Take care.