Hey, everyone. Welcome

back to AI in Action I’m Eric with Fast Slow Motion.

Got Zack Terry Fast Slow Motion’s Director of AI with us here today. We’re recording this on May the 16th, 2025. And Zack, some, some news dropped yesterday, Yeah, I’d say so. Two big changes coming to Agentforce, that’s for sure. The big news that they announced, on May 15th, yesterday, was they’re changing the pricing model of Agentforce.

We kind of saw this coming, didn’t we? Yeah. We are not surprised by this. This has been sort of a rumor that’s been speculated for quite some time. So I think this is a natural progression from the initial offering of how Agentforce is priced, transforming a little bit after, you know, we’ve had enough time for the market to see other AI tools and how they’ve been priced.

Do I think this is a natural progression, but looking forward to kind of diving into what the difference is between the old model? Well, let’s start with the old model. Prior to tomorrow, there was one way Agentforce was priced. Unpack that for us. Actually there’s kind of two ways it was priced beforehand, and it depended entirely on if you were using that agent internally or if you were using it externally.

And by internally, I mean, were you using the Salesforce Copilot product that used to be called copilot? Now it’s called Agentforce default, and I think it’s going to be changed again to be called the Agentforce employee agent. But basically all that is is the agent that you would use if you’re logged in to Salesforce, you click a little button at the top.

You get a conversation pane that comes out. It’s essentially your AI copilot, right where you’re able to ask it questions. It can do things on your behalf. That is a license based model. It is not a conversation based model. And so in order for your employees to use that, they have to be granted a license. That license cost some extra amount per month for each of those users to be able to use.

But you’re not consuming conversations when you’re using that internally on the other hand, if you take an agent and you deploy it externally, so let’s say you have a customer service agent that you’ve deployed to your website, it’s going to be answering questions for your knowledge base. Maybe it’s creating cases, managing cases, things like that. Every single time that you have a customer that comes in and opens a conversation with that agent, and they start conversing back and forth, the pricing model was $2 per conversation.

Now, there are some other parameters around that like, hey, how many hours have passed since they actually started that conversation? And did they close the conversation and then restart that conversation? And it was, you know, 24 hours would be one conversation, right? So if I came in today and I started a conversation and then I came back at a different time tomorrow and more than 24 hours in elapsed, that’s technically two conversations.

And so now I’m getting billed $4 for that. So those are essentially the two ways that you would price it internally. You have a license externally you pay by conversation. And then there’s an entire other sort of component to this, which is what’s called Einstein requests. And that’s any interaction with a large language model in Salesforce whatsoever, whether that is you are chatting with an agent internally or externally, or you’re using a prompt template that is invoked in the background to hit an LLM and interpret the results and generate a response that always consumes what’s called an Einstein request.

And those are billed in a little bit of a different way. It’s essentially a shared pool for the entire organization. You get a pool of Einstein requests, and it’s basically based on the token size. So if I have 10,000, words versus 20,000 words, the number of requests is going to be larger, depending on how large the prompts that goes into that LLM is going to be.

And then you’re going to consume more requests that way, and then you get 200 requests per single conversation that you purchase. So it kind of comes bundled in with some of the products that you buy. And generally speaking, it’s bundled in a way where it’s unlikely you’re going to exceed them. But if you ever do need more, you can always purchase more.

And so, Zack, as you were talking about that, what I’m hearing is, is a little bit of complexity. And that probably drove some uncertainty. Salesforce said that they quoted a second report that said close to two thirds of CIOs just had some concerns around the cost of implementing AI. Salesforce, his own research had that at 90% of these CEOs, CIOs that we’re talking to.

So it seems like this new pricing model is meant to address that uncertainty and tie AI closer to business outcomes. Am I right there? I think so, yeah. So this new model, I would not call it a fully outcome-based model because it really kind of sits in the middle. It’s not purely conversation based, but it’s not. You’re not really getting billed based on an outcome.

So the success of what the agent does doesn’t really go into the pricing model. Instead, this new model is based more on the actions that the agent takes. So in the previous model, if I have a conversation with an agent and I just have some small talk back and forth, you know, let’s say someone was malicious and just went to your website and just started typing in a bunch of nonsense just to, you know, Bill, you $2 for that conversation.

Right? Well, it’s still going to cost you $2. This new model, it’s only going to cost you that agent actually takes actions. So if I ask a question and the agent has to go and retrieve something from the knowledge base, that’s an action. And the way that that’s going to work now is that’s essentially going to cost you the equivalent of $0.10 per action that the agent takes.

And what’s really happening is you have a shared pool of credits, which are called flex credits. So if you’ve worked with Data Cloud, you’ll be familiar with this concept of paying for credits and consuming credits and different credits and different actions have different multipliers on how it affects, and it is all very confusing. This is a little bit simpler than Data Cloud, where essentially it’s instance for 20 credits.

So when you have an agent call an action, it’s going to consume 20 credits, which comes down to about $0.10. Because when you buy these credits, you are paying $500 for 100,000 credits. So the math kind of works out to where 20 credits for a single action is about $0.10. So all that to say, it’s not quite an outcome based model, but if you have a a system or a business process with an agent and you don’t expect to have these long, ongoing conversations, it makes much more economical sense to base your pricing per action, because you may have a lot lower, number of actions that are called per interaction with a customer, as

opposed to if you had a very long, ongoing conversation and you tend to have these much longer transcripts, it may make sense to stay on that $2 per conversation, because on flex credit side, let’s say you had someone who was sitting there for an hour and conversing with this agent, and that agent is doing 40, 60, 80 actions.

But you may end up paying for 6 or $8 for that conversation as opposed to just the flat $2 rate. So it really does depend on the specific business process that this agent is actually accomplishing, as well as the length of time you expect the customer to interact with that agent complex. That interaction is going to be. There’s definitely some considerations to think through as far as which model is right for you.

That’s an important delineation there because you can’t have both pricing models, right. You’re either going to be paying for conversations or you’re going to be paying for flex credits. Is that correct? That’s correct. So it is one model per org. So you can either choose to use the $2 per conversation model, or you can choose to use the flex credit model.

But you cannot combine the two models within a single Salesforce. Yeah. So knowing the data of how your agents, your your human agents typically interact with your customers is pretty important as you decide kind of which of these pricing models drive. This is admittedly relatively hard to estimate, right? You kind of have to do a little bit of guesswork around, well, especially if you don’t already have an AI agent, right?

You have to do a little bit of analysis around how often do we think customers are going to interact with this agent? How long do we think these conversations are going to be? How complex do we think the actions that this agent needs to take are going to be? And based on your answers to kind of those questions, doing that initial estimate that can help inform not only which model you need, but also how many of each you should purchase.

But again, I do think it is relatively hard to do that without getting something plugged in. But just like with Salesforce Foundations, you had, I think, a thousand conversations included for free. You can also do that with foundations with this new flex credit model, where you can get 100,000 flex credits for free, which again, it’s about $500 worth of value.

But if you want to get started with essentially no commitment and very low risk, you can get Salesforce Foundations, choose that flex credit model, and then you have 100,000 flex credits to play with. If you just want to test the ROI of a simple agent implementation and forgive my ignorance here, but Salesforce Foundations is that a different license type is like unpack for the audience.

Kind of what that means. Foundations is it’s not a separate license type, but it is an add on feature. It is available to all enterprise customers and above, and it is essentially a package of common features that you might get with other clouds and Salesforce. So you get some sales cloud functionality, some service cloud functionality, some marketing functionality.

You get access to Agentforce. And previously it was a thousand conversations. Now if you use Flex credits, it’s 100,000 flex credits. So it just is a way to bundle some common features that Salesforce offers with a bunch of different cloud without having to commit to buying each of those cloud separately. You don’t get every feature from Service Cloud, for example, but you might get some case management, Email-to-Case, some of the basics that allow you to run some of your service on top of, let’s say, the sales cloud org you already had.

And then of course, if you demand something that’s a little bit more robust, you can always get the full service cloud license. But this is a way for you to get some of those features and functionality without having to commit to a full license. Very cool. So if you’re kind of wondering, you’re listening to this, you’re going, okay, I really have no idea what level of AI we have in my Salesforce org right now.

You can give us a call, we can help you look into it or contact obviously your Salesforce sales team. They’ll kind of show you around and what the possibilities are. Absolutely. And with foundations, especially if you do not have foundations and you want that a simple call to your API is all you need to get that enabled. And then you just click a couple of buttons, sign an agreement.

Because when you get foundations, you’re agreeing to certain consumption based models like data cloud and Agentforce. But once you sign that agreement, you’re able to start experimenting and building out an agent or any of the other features that you get on Foundation. And Salesforce does have an ROI calculator online. We’ll link to it in the description on the YouTube so you can go on their input.

Different things like the number of human agents, the number of cases they work every day, how much you pay those agents, what percentage of those cases do you think you can deflect to Agentforce? And it’ll give you a kind of a ballpark of of what you can expect to pay, at least in this flex model. And the ROI associated with that.

So it can be pretty helpful. We did an internal webinar a couple of weeks ago. And you know, with this pricing model, again, the intention, I think, is to give people a path to sync their team and without committing to any sort of long term investment, I think before with the pricing model, with the conversations you would consume, those conversations, even in the testing phase.

And correct me if I’m wrong there. So this gives potential people that want to use Agentforce a path again to try out to see what these agents can do for them. So in our internal webinar a couple of weeks ago, you hit on kind of four points when we talk about bringing Agentforce into an organization that first point was start small.

What do you mean by that? Absolutely. So when I say start small, I’ll use a typical business term, which is don’t boil the ocean, right. You can’t take everything you want to do and tackle it all at once. And there’s a lot of reasons that doesn’t make sense, but not the least of which is it takes a lot of resources and a lot of time and probably a lot of money to do everything you want in the short term.

So the best thing that we can do here is, okay, I don’t have AI, but I want to get started. I’m not quite sure how or where or what I want to do. So the best thing is to identify something small, find something you know you can tackle, something you know is an inefficiency in business and something that an agent can help either facilitate, orchestrate, or provide answers quickly.

And then you can grow. Right. But I think the main thing is start with a simple problem, plug an agent into that problem and then see how that goes. Right. And you’ll probably find examples of things that you want to change and iterate on it. You’ll change some of the instructions, but the main thing is, just starting with something small is a great way to what we call it a crawl, walk, run approach.

Right? You don’t want to start running or sprinting when you’re training for a marathon, or you’re training for a race, you want to start building up those capabilities. So we’re going to start with a crawl. We’re going to do just the basics. That’s going to transform into a walk over time, where we’re moving a little bit more quickly, and then we’re going to kind of go off running right.

We’re going to build more features and more functionality onto the system. So start small, find a simple use case, just start building and then I think things will grow from there. Yeah. Your first two points start small. Find a problem right. One problem. Let’s see if we can solve that problem with an agent. Your third point in that webinar was focus on inefficiencies and bottlenecks.

Yes. So that is one of the main areas that you can plug an agent in to start to achieve some ROI. So really you want to evaluate your current business processes and see where time is disproportionately spent and say, hey, well, if I plug an agent into this, can that make it better? Can they make it faster? Can that free up some time for our human agents to focus on more high value work?

So I think the implication there is in order for you to do that, you have to be clear on what your business processes are, because if you are going to articulate what the business process is to an AI agent, that implies that you understand that process well enough to document it and articulate it to another human being, for example.

So it’s a good opportunity to see if your business processes are well documented. Are they followed? Do people deviate from those processes? Where are the gaps in that process that can allow us to a just look at the process and figure out if we’re actually doing this in a repeatable way, and then B take that repeatable process and explain that to an agent so that they understand exactly what their job is to be done and can go out and execute on that job.

It’s the Scalable Business Framework, right? Start with a process, and we look for opportunities to come in and automate it. Love it. And then the fourth point you made was focusing on conversational touchpoints with the customer. And what do you mean by that? So for conversational touchpoints, the reason that I throw that in there is because most interactions with AI agents today are through a chat interface.

Now, that doesn’t mean that that’s the only way that you can interact with an agent. That is just the most immediately available. That is what most of us are familiar with. So that is essentially the easiest use case to get started with. Now with Agentforce, you have you have a whole API where you can call these LLMs in the background.

You can use it programmatically so you don’t have to have this literal conversational chat bot in this window where you’re interacting with it. But the most common use cases are typically going to involve putting a chat bot on your website or maybe you’re enabling an agent to respond to SMS messages or emails for you, right? So all of that really does involve a conversation.

And so thinking through, hey, we’re in our business process. Does it help to have something handling that conversation. And in a service use case it’s that initial touchpoint right. It’s I want to be able to deflect this from cases. And so if I can have an agent have that conversation with the customer and handle, you know, 80% of what’s coming in, that’s going to save us a ton of time and allow us to focus only on the complex tasks or the work.

High value use cases, right. So that can be extrapolated across sales for SDR use cases, making sure that’s we’re only getting those warmed up leads that are actually coming into the pipeline. So we’re having that SDR agent handle that up front. So a lot of different ways think about that. But really the conversation component is I think the first place to start.

Right. That’s where you can go in and you can understand this is what it’s going to look like. Someone’s going to come to the website, they’re going to start talking with the agent, and the agent is going to perform actions on their behalf that we’re almost out of time. Fun as always. Everybody’s talking. I everybody’s wanting to stay ahead of this, stay on the leading edge of it.

Everybody’s curious. What resources do you subscribe to or do you have that have your attention that maybe our audience could tune in to, to kind of stay on top of AI? What’s developing there? I’d say there’s probably more resources out there than your inbox has capacity to handle. But, a few things that I, that I pay attention to.

One is a guy named Ethan Mollick. He’s a I believe he’s a professor at the Wharton College of Business, but he’s been in sort of the AI space for a long time. He’s got a book out there called Co-Intelligence that walks through really good and just sort of basic understanding of what our LLMs, what is ChatGPT, why does it work?

How does it work? What are the business applications that we need to be thinking about, and what are some of the future possibilities? Everything from the apocalyptic scenario to this means nothing and everywhere in between, right? So he’s a really great resource, sort of in the ecosystem. And then, there’s Dharmesh over at HubSpot. He has a platform called Agents.ai.

And they put out a newsletter, and they’re always putting a ton of different functionality onto that platform. So he’s a really good person to follow, but also sign up for his newsletter, and you’ll get a bunch of AI related information in your inbox on a regular basis.

Thanks again for joining us today on AI action. Thanks for all the wonderful, information there. Those two,

resources that I mentioned will be linked in the description. Will also include a link to that article from Salesforce covering the pricing changes and a link to that ROI calculator.

If you have questions about how I can impact your business.

Go to fastslowmotion.com/aiforyourbusiness – Hit us up there and we will contact you and have a conversation or hit us in the comments. Let us know how you’re using AI. Any questions that you might have. That link in the description is fastslowmotion.com/aiforyourbusiness. We’ll see you next time.