PegaWorld 2026: AI Decisioning in an Agentic World: What It Really Takes for AI to Act at Enterprise Scale

Well. Hi everybody. Welcome to our session on AI decisioning. I'm Stephanie Hawkins. I'm on the market intelligence team at Pega. And I'm joined today by Mark Mike Gualtieri from Forrester. He's a VP analyst who covers AI decisioning platforms for Forrester, and he recently wrote the research for the AI Decisioning Wave. Yes. So we're super excited to have him here today. Um, and we're going to start off with a question about Mike's general kind of understanding of the market. Because I know, Mike, you talk to customers every day who are using Agentic AI thinking about using Agentic AI. So what do you hear and where, where is the market in reality? So the market in reality, everyone's talking about at this conference to everyone's talking about it. But the reality of, of agent rollout. Um, it's mostly proof of concept, but there are production use cases.

I have some data slides on this. So we'll show that we surveyed. This is actually a little bit of an older survey. It's about six months old now. But we said what stage is your organization currently with the Gentac AI. And you can see a lot are piloting. We've got some production there. Um, you know, it's really like as an analyst, data has never been, um, less valuable than it is like now because things are moving so fast. Right? So, so I would say this data shows, you know, a lot of piloting, but as the conversations evolve over the months, it's like more and more going into production. So, you know, a lot of the data that you see is not really predictive of where it's going. So we think that this will be a major trend similar to when companies changed to service oriented architecture.

They're going to have a system of agents. It's going to be a layer over everything. So we also asked about the use cases that they're interested in. It's all the same use cases. That software you try to do is productivity automation. You see last on the list is is as a goal to reduce the need for staffing. I mean, that's going to occur in some cases when you do automate, when you do make teams more productive, maybe a team of seven, you know, requires a team of five, or maybe that team of seven goes to ten because you can grow with that sort of productivity. So, so it's all the usual suspects in terms of, of what the, what the business goals are. Um, but, um, the conversations that I find most interesting lately is with enterprise architects who. Are trying to think about, well, what sort of architecture do we need to roll out hundreds of agents and see.

To me, that's a signal that this is going to be a rollout that's going to happen very, very quickly. I still think we're in an experimental phase with some production, but I think it's going to it's going to go very, very quickly, because I think the intent is that we're going to have hundreds, we're going to have thousands of agents. And so when enterprise architects start thinking about what does that mean to our architecture? What should our architecture be and what platforms do we need to develop these agents? That is a signal that tells me that that this is going to go very fast. Interesting. I think we see the same thing with with Pega customers that there's a lot of experimentation going on, but people are struggling to kind of get to that large scale production. So maybe these numbers will look a lot different next year.

Yeah. Well, I think also, I mean, this is going to be part of our conversation. I think about platforms because a lot of the experimentation occurs at sort of a framework level, like an open source framework level. It's like, oh, it's agents and we're going to try to use Lang chain or we're going to use some, some, you know, a framework that you can use to build agents. But that is not yet a complete platform. And what what you know, everyone knows the big difference between gen AI use cases and energetic use cases is the process orientation, the action orientation and the decisioning orientation. And that requires a whole different type of platform. And that's one of the things that we look at and try to cover in the AI decisioning platform wave. Okay. So for these for this data and when you're talking to customers, are you seeing these initiatives?

Are they being led more by the business users or by the IT leaders. I mean its lead is a is a is is a. It's tough to answer when you say who's, who's, who's leading it, because it's a bit of both. It's a bit of both. I mean, business definitely wants to, you know, use AI and drive agentic use cases, but it wants to be there to deliver it as well. So I'd say it's it's led by both, right? One's kind of coming from a technology firm. Well, they're both coming from technology first perspective because it's AI, we have to do AI, but agentic AI, there's a there's a much better discipline about it because it does look a little bit more like software, right? With gen AI, it was just kind of like, let's give copilot to 50,000 people and magic will happen. Um, but when you start thinking about business processes, the decision, the repeated decisions, the consequential use cases and current business processes.

Well, then enterprise architects and business people, they look at that and they say, oh, this looks very familiar. This is we're talking about software here still, aren't we? We're talking about a new type of fancy software. AI magnetic AI. But we've done this before. So and what's good about that is now they understand the capabilities more so that they need in a Platform things like built in governance and some of the other stuff that we'll probably talk about later. Okay, interesting. So I'm sure all of you have come from 1 or 100 sessions where you talked about AI agents, Agentic AI. And what we want to talk about today really is something that's kind of underneath the AI agent, and that is the decision. Yeah. And I think, you know, when you think about a decision in a business context, there are a lot of ways to think about it.

What types of decisions, Mike, are we talking about here in the context of Agentic AI? Well, I had 27 slides showing all the use cases, but we only put one slide in the deck. Okay. And this is just a few categories of the decisions. But if you think about all functions of a business and we don't have like lending use cases, we don't have like fraud use case. I forget the slide we picked. But but but we create, we literally created tons of slides on every decision that gets made in a business. And yes, there's literally thousands of these decisions. It's not the type of decision that's like, who should we hire? It's more the type of decision that's like, is this person a credit risk? Oh, interesting. Yeah, yeah, yeah. No, that's a very good, very good point because we're not talking about sort of business intelligence.

Like we're not talking about the types of well, we're not talking about these decisions yet, but we're not talking about like, should we open a new factory? Should we open a new retail outlet? I mean, it could get to that probably will, but that's probably going to be more decision support, but we're talking about those repeated decisions and things that you might not think are a decision. Like, I spend a lot of time talking to financial institutions about fraud use cases, because when a transaction comes in, you're making a decision. What is the likelihood of fraud? Right? That's a decision. When you get a loan application, you're making a decision. Should I, um, should I approve this loan? When you're getting a claim in you're making adjudication decisions. So there's decisions that are, that are embedded in in software everywhere. And the question is when we start moving to agents and when we're using LMS as sort of the brain to make those decisions, how can we make those decisions predictably?

Um, how can those decisions be optimized and even self optimized? So it's very easy to find use cases for decisioning. It's kind of the same way you find a decision for an agent. You look in your business processes and you will find where decisions get made in an automated fashion. And that is probably where an agent can also make that same decision. Okay. And in terms of how you're making the decision, I know you write a lot about something called blended intelligence. So what is blended intelligence? Yeah. So blended intelligence. So so if we back up a second and talk about decision intelligence. So decision intelligence is this discipline of how an organization arrives at decisions. And decision intelligence is broader because decision intelligence does um, cover. Well how do humans make decisions and how can we support those decisions, but for the purposes of an AI decisioning platform, the goal there is the automated decisions.

So. So decision intelligence is what you need to bring to an agent to make that decision. And I just talked about LMS, and you brought up one of my favorite topics, which is blended intelligence. And so the way we think about it is we think about if you're making a decision, meaning you, the agent, the AI agent, um, are you just going to make that decision with an LLM? No, you're not actually, you want to you want to take a blended intelligence approach. Now, you've probably been to a lot of sessions about agents and, and the, the unique characteristic of an agent over, say, regular gen AI, although it's kind of morphing into the same thing an AI agent can call a tool and a tool is, you know, some piece of software. Any other piece of software or it can call another agent. So the concept of blended intelligence is, let's use any and all and the best decision intelligence technologies to make that decision.

So it's not always just data coming into an LLM and the LLM making the decision. It might be a score card. It might be human decision logic. It might have a slide on on all of the different ways that a decisioning agent can make a decision. So, um, uh, optimization, often known as mixed integer programing, it's mathematical programing, regular old analytics data, business rules. So so all of these different ways to make a decision. You want to give those to your agent to be able to use those technologies. And then that brings up the question, well, how do I build those tools? Not very easily unless you're using a platform. So one of the things we look at at Platform is what tools are available. Now I happen to cover AI platforms as well. And the origin of AI platforms came from Data Science World where the output of the platform was a model.

But what's happened is even the AI platforms need need to support agents. They have to support all these other technologies. But a lot of those platforms just give you a way to write code and then call that code. The unique characteristic of AI decisioning platforms is they have ways to build and use tools by business experts that are already available. They're already in there, and they may have to do with integration to existing systems, but they might have to do with optimization scorecards, human decision logic via business rules. So the point is, is that an AI agent isn't just using an LLM, it's using any and all decisioning technologies that are relevant to that decision. We call that blended intelligence. Okay. So we have the regular AI agents we all use as part of our work life, whatever our work gives us. Copilot Claude. And then we have decisioning agents.

And those are different because the decisioning agent would be using all of these things on the screen. Is that correct? It might it might not be using all of these things. Okay. The point is, is that that it could or could use a subset of these things. The real point is it doesn't just have to use an LLM. OK. But it should use Other. We shouldn't throw out other decisioning technology like, for example, science or math, right? Sometimes, you know, even the LMS. Now, if they're if they're doing a math problem, they call a tool to do the math problem, right? So, so that's the same. It's enterprises need to think about tools in the same way to make a decision. Okay, so the research that you cover is on AI decisioning platforms. So we've talked about agents. We've talked about decisioning agents. What is an AI decisioning platform according to Forrester.

So. The focus is on there's two characteristics of this platform. It outputs an agent that makes a decision. But it provides design time tools that allow business experts to actually define and change those decisions. Now, sometimes those decisions just require a few rules and some logic and some analytics. Sometimes those decisions are a multi-step process. So the platform is essentially a development and deployment platform which provides design time tools And a deployment environment. So obviously Pega is one of the vendors that we'll see the we'll see all the vendors in our decisioning wave here in a. I don't know how long, but. OK. A number of minutes or seconds, I'm not sure. But um, but these platforms cover the, the, the full lifecycle. Now where did these platforms come from? Gee, I've been doing this for 19 years. And the first evaluation I did, it was called the Business Rules Management System wave.

And then it kind of morphed in recent years to the AI decisioning platform wave because, because it wasn't just about business rules. It's about all these other decision intelligence technologies. The problem with most agent builders that are not in AI decisioning platform is you go in there and they do a demo and it's like, oh, well, here's your big text box, right? Your big old prompt here and then add your tools. Okay. That's not, you know, writing a three page prompt is not necessarily a great way to define how you're making a critical decision. Business decision. It's much better to use design tools to define that decision the way humans think about it, or the way they need to express to the agent on what it should use to make that decision. So, so an AI decisioning platform covers that whole development lifecycle and is a deployment platform as well.

Okay. Are there any non-negotiable things that would absolutely need to be in an AI decisioning platform for it to be called one? So I just happen to have a slide on that. So. When we do the wave on this. Um, we scan the entire market for vendors that we think are AI decisioning platforms. And the first defining characteristic is it has to have, it has to be designed, it has to have design tools that are geared for business experts. Now, that doesn't mean it excludes technical developers or doesn't provide a way to program, but it has to have tools for business business experts that usability. That's something that most of the vendors have spent many, many years perfecting those tools. How do I express a decision? How do I bring in optimization? How do I allow it to self optimize? How can I put a whole bunch of business rules in there?

So, um, that's the first requirement is that, that it has to have those tools usable by business business experts. And the second one is, um, that it has to support AI agents. That's a new one. That's only become non-negotiable, like in a year. And every platform has stepped up in some way, right? There's a lot of varying degrees, but there's no platform that I've been covering for years that says, oh, you know what? We're just not going to do agents. There's no one there's no vendor that says that. So they have to be able to, to now compose these decisions. You know, before they were saying, well, yeah, you can, you can create this decision and then call it with the RESTful API or expose it behind a service. Now those have to be within the context of an agent and all vendors have done that.

Um, these decisions are consequential. So there have to be governance. And you know, you'll see later when we evaluate, we don't have a criteria called governance because it's not like one thing that you point to, it has to have, it has to have governance built in in different steps. And there's many different types of governance. There's governing life cycle governing IP, governing the quality of the decision. There's the security element of governance. So governance is a big topic. Um, but it has to be throughout the Platform it can't be bolted on. Um, and then what, what we think, what we think distinguishes the leaders from some of the other vendors is being able to optimize those decisions against KPIs, against business, business metrics. And we also think that that's the future of these platforms as well. So right now, a lot of these platforms focus on how do I make a decision and how do I use different Blend, different decision intelligence technologies to make that decision?

MM. I think that's going to go to how do I define KPIs? How do I measure it? And how do I tell the system how to optimize the decision? How is it going to optimize the decision? Well, it's going to find its own rules to optimize that decision. So so that's so business outcome optimization is important. But I think it's going to be even even more important. And then what I also think is non-negotiable is the vision the vendor has for how will agents be used in an organization. The reason for this is because while the current capabilities of of the platform matter, it's changing so fast, it's changing so fast. So you have to understand, well, how is that vendor going to keep up with these changes? Or am I going to be disappointed that they're not going to be able to keep up with the, with the innovation that's happening in AI?

Right, right now? That makes so much sense. Yeah. So you do this evaluation every year or every other year. Um, we do it usually on an 18 month cycle. But for fast, you know, fast changing markets will sometimes will do it in 12 months. Okay. And there are 20 vendors in the landscape and then 15 that actually make it into the wave. Yeah. So there were 20 and 2025. Okay. Our, our research methodology allows us to have up to 40 plus vendors in the landscape. And the landscape is where we don't evaluate the vendors. We just identify worldwide, the vendors that are relevant. Um, we're going to do that. We're going to do another one this year. And I expect that to be close to the 40 because there's a lot more vendors getting into the space. And then from that up to 40 vendors, we have a maximum of 15 that we put in the wave.

Um, and so, and then we evaluate those 15 vendors. Now, you know, people ask me, well, how do you determine who's who are the 15? Well, is that based upon their revenue? No, Not. No. Absolutely not. Because buyers, they want to see a new upstart as well. They want to see, well, how does this new company compare to a well-established company? So the inclusion criteria has more to do with market momentum. It has a lot to do with who clients are asking us about. Um, so sometimes vendors I've had in there for years. They're not in there anymore. They lost momentum and we needed to make room for maybe an up and coming vendor. Okay. And I know you have pretty rigorous evaluation criteria. So what does it look like for these vendors when you're evaluating them? So here's what the wave looks like. And you can see the um the y axis there, which I've labeled Platform capabilities.

So I come up with a set of criteria that I evaluate all of these platforms on. and how do I come up with that criteria? By listening to the buyers. I'm talking to the the enterprises who are buying these platforms and they're telling me what they're interested in. It's been more difficult, though, because AI becomes, uh, because it's changing so fast. Sometimes I have to anticipate what they might be asking for like next week. So it's, it's not just about listening to, to what they want, but it's also about what's coming. But anyway, we come up with this, this criteria of platform capabilities and that, um, positions of vendor on the up and down and then on the x axis. We also look at a company's strategy. That's their vision, their innovation, their roadmap, their partnerships, their, their ability to implement. And that's what gets a vendor on the, on that x axis.

And so we have, um, what do we have contenders? Strong performers, leaders. And we got it. And so all of these vendors have that that non-negotiable capability, which is that business friendly, um, user experience. So, so other vendors that came from, say the data science world, they don't have that sort of thing. So they don't appear here. There are some overlap vendors on here. Um, like. Uh, uh, IBM, for example, Palantir. So Pega is on here obviously as well. So, um, this when companies are asking about what should they start with, with AI, I tell them an AI decisioning platform over the regular AI platform, because the regular AI platforms for the most part, don't have life cycle governance, and they don't have these tools that allow that keep business people in control. They still there still has to be like a middle person who's interpreting what they want and then coding it.

And so it's much better if the business expert can can do it directly. Okay. For the, for the vision category. I mean, how do you, how are you able to determine what a, what a real strong, durable vision looks like versus a vision that's kind of all talk or hype. Yeah. So the way I look at vision, is it like, I'm not looking for one thing like, okay, I'm going to ask the vendor what their vision is and I'm waiting for them to say X. There's a little bit about that, but it's more how is the vision likely to help this vendor grow more adoption? Because that's like the enterprises that are voting, um, that they want to spend money on this vendor. And I think the most difficult thing for an enterprise right now is who is my AI partner, who I'm going to have hundreds of agents running critical business processes.

What Platform can I do that with? That is a very difficult question because it's changing so, so quickly. So the vision of that. So if you get a if you get a company that's saying, well, you know, agents are going to roll out over the next 3 or 4 years. It's like, that's probably a weak vision because most companies want to go a lot faster. Mhm. So. Um, but it's going to be a combination of the vision, uh, proven innovation and what they have on their roadmap. Okay. Because you can have a great vision, but if you can't execute on it, you know, if the client can't trust you to execute on it, that doesn't matter. So it's a combination of those, criteria. Okay, that makes sense. So I want to move on to the topic of decision lifecycle management, because it feels like that core set of capabilities are what distinguishes these platforms from something like an AI platform that might have some of the capabilities, but not all of them.

And decision lifecycle management is just being able to take every aspect of what goes into the decision and having a tool to help you shape that to be the best possible decision. Yeah. So what are those capabilities actually look like? So this is in fact, the criteria that we used up here on this slide to determine that y axis of a vendor. Um, and there's a couple of data categories here. Data modeling. Because if an agent is going to operate on the business, it has to have a model of the physical assets and digital assets of the business. Now you've got a lot of people now running around saying, oh, you need an ontology or you need a graph database. You need yeah, you need some of that. But, um, but it's really just a data model and there's different ways to implement that. Um, and so we look at, well, how quickly can I represent, uh, the, the objects that I'm actually making decisions on?

So we, so we look at that. The second thing is we look at the data integration, because a lot of that data has to come from existing systems. That's the difference between an enterprise and a mature platform than sort of a new naive platform, is that most of the most of our clients are large enterprises that have grown through huge acquisitions. Their IT environments are very complicated. They have like, you know, they may have four ERP systems because they made acquisitions. These are very complex environments. So so the integration to all of that data is critically important. So we look at that and then we look at, well, how can I actually author a, we call it authoring sort of a, an older term, but it means how do we develop these decisions? How do we develop the agent? How do we tell the agent, um, how do we give it the parameters in which it should make a decision?

How do we give it the guardrails and how do we give it access to those decision intelligence technologies that it can use as tools to essentially achieve blended intelligence on those decisions? So we look at that testing which which should be part of a lifecycle. But testing is a little bit different these days. Testing used to be, okay, here you go. Let's test it. Does it work yes or no? Now more and more companies are actually testing it in production. So a B testing type of thing, which is a common way to test a new feature. But with some types of decisions those decisions can be um. I don't want to say the word hallucinations, but they could hallucinate a decision. So you. So sometimes a company will will put out three versions of that decision against production. And they'll have a new one they think is fabulous and better, but they only want to put like 5% of the traffic against it to test it in the real world.

So, so the testing gets more complicated as the point. We look at the, the features for gathering business observability, telemetry. We definitely look at we have two, two criteria just for the user experience. Okay? One is cohesive user experience. Like do all of these things kind of hang together. And then how easy is it to use scalability and some of the other enterprise, uh, nonfunctional capabilities for deployment? Okay. So essentially, when you're using an AI decisioning platform, the decision you're able to make is going to be a lot more sophisticated than what you'd make with a platform that didn't have this breadth of capabilities. So. Well, what a decision. Yes. What a decisioning platform is. Competing with most of the time is hard coded in Java or COBOL. I mean, it's still the reality. So when when when a company invests in a decisioning platform, usually it's to pull out the decision logic from within.

You know, like if you're doing like a KYC, know your customer, you have to make a decision about, is this a risky customer that requires an investigation. You could have KYC logic in six, seven, eight different systems. So a company might get a decisioning, say, you know what, I'm going to centralize that. I want to I want to unify how I do that, and I want to be able to change how I make that decision very, very quickly. So a lot of times you're competing against hard coded OK now with agents, that actually adds a new use case because now if you're developing an agent, agents will make decisions. Um, and they shouldn't just make a decision based upon an LLM they need to give that agent Other decisioning tools. That's why I make a big deal about blended intelligence. Um, so this is like an entirely new use case that's also important for decision platforms.

Okay. So for, for a decision like, um, is this fraud, for example, uh, my understanding of the way you do it without a platform like this would be, for example, flag every purchase over $10,000. What would the equivalent be in an AI decisioning platform? How would you how would it be a more sophisticated decision? So. One of the things we have with gen AI is we can do it. So okay, let me, um, if you're doing credit card fraud, do you know how fast that has to be super fast. Um, but if you're doing like a claims fraud, you have a little bit more time, right? But, but some of the things that that current AI brings us is, is ability to do investigations of counterparties and, and read documents and so forth. So if there's time to do that, AI can bring a lot more sophistication to the type of information you can look at.

Um, if there's no little time to do it, um, you can start using, um, let's call them classical machine learning models. Um, and more than one and in concert with um, analytics and other forms of data. And if you need to do that at low latency, right, you should be thinking, oh, do I have time to call an LLM. MM. I don't think I do, I don't think I can call that LLM. So, so you might do like a first pass, deny the transaction or let the transaction through and then automate an investigation after the fact. If you have to look at sort of unstructured data. But the other thing to think about is that business processes can be including fraud, because that was your example can tend to be very regular, right. And the patterns tend to emerge. So everyone's always worried about the cost of the tokens.

It's going to go down. The reason it's going to go down for enterprise. I mean, it's going to go net up because you're going to use it. You're going to have hundreds of agents, but the cost per agent is going to go down because agents can identify these patterns. And what can LMS do? LMS know how to write code. So an LLM can look and say, oh, I've seen this. I've seen this a thousand times before. You know what I think? I think I'm going to write a little Python script. And when I see this pattern again, I'm just going to call the Python script. I'm not going to call multiple make multiple LLM calls. So there's all kinds of tricks like that and caching, just straight caching that are only just beginning to be employed. And that's going to have, you know, at least cut most of enterprise use cases in half how many tokens a transaction is going to need, and not to mention lower the latency of a call.

So there's lots of optimization tricks that. They're not going to have to reinvent the wheel every time, right? Yeah. So I'm curious what advice you would give to anybody who is sort of just getting started with decisioning, thinking about buying an AI decisioning platform. What are the first steps? How do you get started? Um, so the key. That first non-negotiable requirement of a platform is that business experts have to be able to participate in the development process of these decisioning of these decisions. Number one, they're the ones that have all the expertise. Number two, you don't get things lost in translation. Doesn't have to go through like a formal SDLC. It still still has to have a life cycle right to be tested and deployed safely, but it's just a much faster way to deploy decisions safely and within an agent. So. There's still an awful lot of pilots going on and people want to test it first.

I mean, my advice is this is going to happen. This is happening. Who thinks it's not going to happen? I don't know. So why mess around with, you know, a pilot? We recommend that our clients invest in a platform and just use the platform from from the get go because it includes the life cycle, it includes the tools. So even before a regular AI platform, which came from a, you know, producing models, we, we still recommend an AI decisioning platform because it has It has the ability to bring any and all decision and intelligence technologies into that agent. Now, the reality is most. Having said that, the reality is we don't think enterprises are going to standardize on one platform. No way. No way. They're going to have six. They're going to have seven. So interoperability between agents and tool reuse is going to be very important.

I'm saying going to be very important because it's interoperability among agents is very immature right now. Google put forth a standard called A to A. Most aren't using that. But the reality is enterprises will have a reason to to have multiple platforms just like they have multiple platforms. Now for for other things. But agents are also going to enter the enterprise through their business applications and through their SaaS applications. And all of those are going to have to interoperate with with this. So. An enterprise is going to have just like they might have thousands of services, they're going to have thousands of agents and they're all going to be interoperable, or they're going to have to be interoperable, but they're also going to have they're also going to have to be controlled. They're going to have to be changed very, very quickly, um, as the business environment change.

So you need those tools that you can do that quickly. You can't change a six page prompt. Right? Okay. I love this slide up here about business experts being in control. Are there certain, uh, capabilities off of that list of capabilities you showed earlier that are specifically would line up with the business expert being able to be in control? Yeah. So different ways of representing. So that list of decision intelligence technologies like analytics, rules, tables. So all the different ways to express what you how you want that agent to make the decision, right? So if it's, if it's an insurance company and it's auto insurance for 50 different states, right? You need a way to express the rules of those 50 different states. Um. Not like a 50 page document that that becomes part of a product. You can't debug that. Right. That's very difficult to debug.

So. So you're looking at tools that allow you to express how that decision gets made, uh, using the decision technologies that matter. Okay. And I know we touched on this a little, but I would love it if you would just talk a little bit more about governance as it relates to AI decisioning platforms, because you don't call it out as an explicit category, but you say it needs to be embedded. So what does that mean? Because there's too many different categories of governance. There's like business outcome governance, which really is all that matters actually. Well, doesn't know there's risk categories that matter equally, but but like business outcome governance is that observability, getting the telemetry, being able to define KPIs and business metrics and have them fed back into the platform where they can be monitored. In some sense, you need a decision agent for that.

It has to decide when things aren't working and when things need to be optimized or shut down. So that is not like a separate standalone thing that needs to be built into to every agent collecting that telemetry. The life cycle life cycle governance itself. So that's simple things like version control. Who changed what. That type of governance. There's IP governance. If we're calling models and we're. We have some stuff embedded in the prompts that we consider trade secrets. Um, are we sure that that that's being protected if we're, if we're making external model calls? So there's many different forms of governance that, uh, that matter here. So the idea that there's like one governance module, I don't know what would you do in that module? Because you have to do it. You have to, there's governance issues on every step in the lifecycle. Okay. That makes sense.

Um, I'll ask one last question and then we can move to questions. So for customers who, you know, are doing this at a mature level and are talking to you about it, are there ways that they're actually measuring the impact or their success with these platforms? What are the what are they talking about when they talk about ROI? Well, so. The ones that are furthest along actually converting decisions to decisioning agents, um, already had a mature decisioning process. This tends to be large financial institutions that are making lending and credit decisions. Those are very, very sophisticated. They've been doing that for a long time. Um so they essentially use the exact same metrics that they've been using all along to measure the effectiveness of those decisions. And many of them use a CHAMPVA challenger approach, which isn't new or an a B testing approach to see the impact on a decisioning agent versus, I don't know, some hard coded decision or, or even multiple agents.

So, so there's a lot of sophistication already in how that gets measured. That's one element of, of ROI. The other element of ROI is how quickly they can change the decision, right? Because you'll get, um, you know, the strategy changes. I don't want more customers. I want the same amount of customers, but I want to lend more to them. Okay. So as you start changing that, it changes how you make decisions or regulatory requirements that come in. So the other way to measure ROI is how quickly can I respond to either business strategy changes or, or, or compliance changes? And those those same approaches apply to, to any business, even if it's a manufacturing business. So when people ask me, well, how do I calculate the ROI? Same way you always calculate ROI with software. Um, it's just that there's increased risk. There are some, some different types of risks.

And sometimes the implementation of AI is longer because it's a newer. It's a newer technology, but I think that's going to turn into. I think as as the world and as enterprises learn how to how to build these agents, they're going to. It's going to make these projects much more predictable in the same way that they can predict how much, how long it would take to change software. Okay. All right. This is great, Mike. I mean, I think it's really interesting that this category of software is so important. Um, I love the thought about the, the business expert kind of being in the driver's seat and really needing to have that person drive what these decisions look like. Um, so thank you so much for being here, explaining this all to us. And I'm happy to take any questions. If anybody has any questions for Mike, feel free to jump up to the mic.

I can keep going and just ask more. I just wanted to give you all some time. Well, actually, I think we only have 30s left. That's not enough. Time. No, I don't know if that's enough time. Uh. Um. Well, let's let's do one more. Let's close with a hard one. So you have talked about, uh, AGI. How does that kind of relate to all this and what are the implications? Start prepping. Um, I mean, the thing about AGI is that. It also is going to depend upon, um, data. It's going to, it's going to depend upon how, how it can sense what's happening, right. So it's going to have that sensory and that and that's data. So so even if there is a is an AGI, it's not a universal AGI that knows everything, right? So it's going to be industry specific. It's going to be company specific.

So companies are going to have to feed it. They're going to. It doesn't have eyes. It doesn't it can't sense what's going on. So this is where companies are going to be able to take advantage of that, because they're going to have to go from like building software, because it's going to be able to build software to providing it data about what's happening in the real world. So so being a company is going to be more about, about building infrastructure to sense everything that's happening in a market, because that's the only way it's going to be able to differentiate from someone else's AGI. MM. That's really interesting. How close are we to this? I'd say well, so at Forrester we have four levels of AGI. Um, and I think the first level we call like competency, that would be like a person, a business person.

So we're, we think we're like 1 or 2 years away from that. Um, that's going to be able to do tasks. It's not going to be the level for super intelligence. Um, but already, um, it's a lot, uh, you know, the, the knowledge it has its vast and the reasoning capabilities are getting a lot better. It's probably going to have to have a new uninvented technology or a technology in a lab that we don't know about yet to really become AGI. But I think before people were thinking it was 50 years away, but it feels like it's much closer now. Okay. Very interesting. All right. Well, thank you so much, Mike. Thanks for being here with us. And thanks for being such a great partner. And yeah, thanks, everyone for coming. Thank you.