Why Pega Is Ditching Token Costs and Betting on Business Outcomes Instead

 The Director of Pega's AI Lab breaks down the real mechanics behind agentic marketing operations and why "magical thinking" is killing AI projects before they start. 

In this CX Today discussion, Associate Editor Rhys Fisher sits down with Peter van der Putten, Director of the AI Lab and Lead Scientist at Pegasystems. 

With agentic AI dominating the conversation across the CX space, Peter cuts through the noise to explain what it actually takes to make it work in enterprise marketing, and why orchestration, not individual agent capability, is the real differentiator. 

If you're trying to separate genuine AI progress from hype, this one's worth your time. 

Agentic AI is everywhere right now, but most of the conversation stays frustratingly shallow. 

Peter van der Putten goes deeper, explaining how Pega's newly launched Customer Engagement Studio works alongside Customer Decision Hub to give marketers a governed, agent-powered path from brief to live campaign in minutes. 

Outcome-based pricing: Pega is making a bold move away from token-based costs, betting they can tie agentic AI tightly enough to business outcomes to charge on results instead. 

Left brain meets right brain: Peter explains how CDH has handled analytical decisioning for over 30 years, with Wells Fargo running 6 billion next best actions a month through it, and how Customer Engagement Studio layers in generative and agentic AI to solve the content and marketing ops bottleneck, not the decisioning one. 

How the agents actually work together: A conversational agent orchestrates specialized agents across marketing strategy, creative, data science, compliance, and performance, each with a defined role, none operating in a silo. 

Why 40% of agentic AI projects fail: Peter points to "magical thinking", the assumption that throwing agents at a problem will sort itself out. The fix is embedding agents into real workflows tied to measurable business and customer outcomes. 

Transcript

SPEAKER_00 0:08

Hello and welcome to CX Today. I'm Reese Fisher, Associate Editor, and today I'm delighted to be joined by Peter Vanderputten, the Director of AI Labs at Pega Systems. Peter, thanks for joining me. How are you doing today?

SPEAKER_01 0:23

Yeah, I'm great. Thanks for having me.

SPEAKER_00 0:26

Yeah, no, I'm delighted you could join us. You know, we're going to be talking, I guess the conversation is going to be based in this latest release from Pega, you know, the new cluster engagement studio, and we're going to kind of unpack that a little bit, I guess. So I suppose just from the jump, one of the things that stood out to me about the release is the claim that marketers can go from brief to live campaign in minutes while still maintaining that government. What does that workflow actually look like in practice? Where do the humans stay in the loop, perhaps, and where do the agents maybe uh genuinely take the wheel?

SPEAKER_01 1:03

Yeah, no, I think that's a that's a great question. And indeed, it's almost that what we're adding here is an agentic approach to uh marketing operations and one-to-one customer engagement. Um, but maybe before I even kind of uh jump into the solution, maybe I can set up the problem a little bit more as well. Is that okay with you? Yeah, yeah, absolutely. Awesome. And so uh I I think we as customers uh we we almost kind of enter morphized, you know, the enterprises that we engage with. We we kind of expect them to understand our needs and our emotion and where are we in our in our journeys. Uh, whereas if you look inside these companies, they're incredibly siloed uh across products, but also across channels, and marketing is run separately across all of those channels. So it's it's very hard to deliver on that expectation that customers really have without any central orchestration, right? So, and we just had our big customer event, and Wills Fargo was talking there about how they implemented what we call customer decision up at Pega to really have a centralized approach to one-to-one customer engagement. And that that started with um actually with control, not not with efficiency or marketing effectiveness, but with control for a bank, it's really important uh to make sure that you know anything you talk about is something within kind of the regulatory boundaries. And you are kind of referring to it a little bit, uh, but then also uh by by virtue of having kind of a central brain that kind of listens to customer signals and uh then looks across an entire library of next best actions, as we call it, to figure out uh what is the most relevant thing to talk about here in the moment, then serving it out across all of those channels and learn from interaction, right? So that that's basically the core approach of next best action, one-to-one customer engagement and customer decision up. And really have uh, you know, the analytical left brain AI figuring out uh what is the right thing to talk about, and that's where Wells Fargo ended up. Uh they're doing six billion of those interactions every single month across a myriad of channels, across a myriad of propositions, right? So yeah, um, and that's delivering you know double-digit MPV uh improvements um uh for their for their customers. Now, um then they wanted to take it a little bit to the next level, and so because it's a very powerful brain, it allows for those consistent journeys and interactions, but you need to feed the brain with content, yeah. And that's of course where Gen AI and agentic approaches can actually come into play. Yeah, uh, and that's this approach with customer engagement studio, we have where you have this agentic approach uh to to marketing ops and to customer interactions, uh, to to really uh uh um to really indeed start from a short marketing brief and uh translate that into kind of live personalized actions and minutes by by using like uh a set of agents that are collaborating together. I can delve into that a little bit more, uh, but to to really extract the right information and and to um generate the right content and actions.

SPEAKER_00 4:27

Yeah, when I was reading around kind of the the your use of a gentic AI here, I thought it was quite interesting. I think quite often it's focused on what individual agents can do, whereas with you it seemed to be better on how you can actually coordinate them. That seems to be where the real value is. What do you think made that orchestration design a priority for you guys?

SPEAKER_01 4:48

Yeah, so uh so so I was kind of referring a little bit also to left brain and right brain. So we already had the left brain, and we've been doing this kind of next best action omni-channel interaction and learning from all those propositions already for uh well, even 30 years or more, right? And uh, but then Gen AI came around the corner, and of course, uh Gen AI is uh there's a big opportunity there to generate content or to understand um uh which parts of your uh marketing are working and which aren't. Yeah, so we were looking for this combination of how to combine the left brain, the analytical brain, with the right brain, you know, with the creative power of Gen AI, and then ultimately also with uh with the gentic AI. And how we approach that is indeed uh uh by having kind of a conversational agent that's fronting it all, and the conversational agents is interacting with other agents, pegger agents and non-pegger agents, uh, um, that are guiding you to this process of how to translate your business objectives into actual uh into actual actions that have been set up in the system. So the conversational agent is this, let's say, the central orchestration of it all. Uh but then uh you get into well, it's it's leveraging marketing agents, it's leveraging um uh creative agents, it's leveraging uh performance, marketing performance agents, data science, machine learning agents, uh, but also compliance agents to really um to really get everything set up. And each of those agents will carry out a specific role in this overall process.

SPEAKER_00 6:33

Yeah, yeah, no, it's really interesting stuff. I want to draw back to something you mentioned around kind of the the decision making. Obviously, uh Peggy, you've had the I think you mentioned the customer decision hub there for quite some time. How does the customer engagement studio, I guess, differ to that or build on that? What's it what does it provide to customers who perhaps are already customer decision hub customers now they're interested in the customer engagement studio?

SPEAKER_01 6:58

Yeah, that's a great point because um, of course, with Genai arriving, people kind of uh naively thought like we can just put an LLM in the middle and it will figure out what to talk about to a customer. But yeah, that's that's not gonna fly. Yeah. When I uh talk about Wells Fargo and they were talking about um, you know, their their approach, uh, they're doing these six billion next best actions, but also they need to come back uh in well under 250 milliseconds, right? So when you open up your mobile app, it needs to decision across hundreds to thousands of actions, and it needs to be able to do that incredibly fast. Yeah, um, so you can't have the LLMs there in the moment. Well, how we how we kind of leverage generative and the Gen Tic AI is indeed in this marketing operations the let's say the design time process of configuring new actions and treatments into this into the system, and so that's how it actually complements uh what was already there with customer decision. And I spoke about these various uh agents and the the conversational agent is kind of fronting it all. Yeah, so the whole process could start either you know with the system proactively detecting there's performance gaps or relevance gaps for customers, but it can also start with some high-level briefing IDs you may have as a marketer. The conversational agent will actually take that and and maybe hand it off to a marketing agent uh that is going to tease out you know, what are the various um actions to talk about, the various channels that are involved, the various sets of creative assets like treatments uh that could be there. And it's maybe also conversing with, let's say, a marketing performance agent or even a data science machine learning agent, figuring out, yeah, but where are the the main gaps really um in your uh in your current marketing, uh, in your current marketing approach. Um, yeah, so um and based on that, uh there's a more than kind of you you get to a point where there's a more kind of fine-grained idea of what those different actions, treatments, policies are, maybe also reusing uh compliance policies that need to be taken into account or making sure that they're already pre-approved uh sets of content or policy rules that would speed up the process. Uh, it could call out to a uh, let's say a creative agent. Yeah, could be a pegger agent, could be a non-pegger agent that is going to test and create all kinds of different treatments. And then conversationally we can say, well, we like these pieces of content, we don't like those, run a simulation. Yeah, maybe these creative agents can be smart in our RD lab, in our AI lab, we're also looking at how can we actually generate treatments that at the same time we also looked at can we without any feedback from customers already predict which ones are most likely to work better? Yeah, so that's a little bit of uh AI magic there. But then the creative agent comes back and makes proposals, and then you can say, Well, I like these, let's tweak those. Can you make them a little bit different? And now run a simulation. Yeah, uh, run a simulation that will actually give me some indication based on these treatments and these policies, uh, what the reach is going to be for those treatments and what kind of response rates can we actually expect. Um now, and once you're happy, you can say, hey, you know, I want to deploy this, but the system is also smarter smart enough to figure out then based on the types of actions and treatments you're talking about, which which people actually need to provide sign-off. Is it a simple marketing sign-off? Does brand police come into play? Of course, we build all the the brand police guidelines already into the agent, but there's still a human that in certain cases needs to approve. Is it something that can simply be released? The compliance agent already signed it off, or is there a compliance person uh in legal that needs to sign these off? Maybe it's a regulated treatment or something like that, and then uh it will arrange those approvals and it will actually kind of deploy. And then post-deployment, you can also use the agency system to figure out hey, you know, uh what's the current performance, what's working, what's not working, how should I potentially tweak things, yeah, yeah, particular treatments that we actually should remove, should change, um, those type of things. Yeah, so it's it's really in that sense an agentic conversational way to engage and to really provide um speed up um speed up the delivery of new pieces of content uh and actions into the the the customer decision app. Get this 10x uh amount of of content into the system. It's no longer, it was no longer the left-brain customer decision app, which was on the critical path, but it was actually on the marketing side, on the organizational side. How do we figure out uh how do we bring new content into the system? How do we make sure we smoothen the compliance process? How do we make sure it's it's directed into in yeah, into the direction of treatment that's most likely to work well? And that's where you know agents that work alongside a marketer can can really speed up that process.

SPEAKER_00 12:20

Yeah, I really enjoy listening to you kind of explain how these agents sort of interact with each other, and it feels like a really kind of a seamless, kind of fluid process. So just wondering what what kind of feedback you're hearing from the human agents who are working with these. You know, we're learning a lot about tool fatigue lately. I was just wondering what is, yeah, what do they respond, how are they responding to these to these tools.

SPEAKER_01 12:43

And so uh from day one, uh, we actually also co-developed this with our with with uh some select clients, uh uh uh including uh Wells Fargo also spoke at the event uh what their experience was, and they they truly can see that um yeah that they can they can they with this solution they will be able to kind of speed up um the um the when the end-to-end marketing operation process uh really speed up um yeah on one side the amount of creatives and variants that you can um that you can release. And what they spoke about, for example, the three times three times three. So rather than having one treatment or action, and that you still where you still need to have all kinds of variants for different channels, and maybe you have different main audiences where you want to vary the treatment. There's maybe it's maybe not just a simple offer if you buy a mortgage, right? So it's more like oh, I'm expressing interest versus I'm in the process versus I asked for a quote, so it's an entire journey, and so you already have these, yeah, maybe three different main audiences. They maybe have journey stages, you have three there, then you have all the channels, and but also different variants and treatments you want to try out, and so yeah, if you say if you take three in every stage, that's already nine X that you need to deliver, right? But on the flip side, you don't want to get into um using Genai to yeah, create just even more stuff without making sure it's regulated, that's compliant, that's likely to be performant. Yeah, so and that's where also the agents can actually watch over the shoulders and make sure um uh the the extra content that's being generated is essentially also driving towards more relevant experiences for your end customers and and better marketing performance. Right. So and that's that's a little that's that's also in our co-development of uh Customer Engagement Studio uh um jointly with our our key clients. Uh that was the feedback we got. That was really helping them with that kind of organizational challenge.

SPEAKER_00 14:55

Yeah, listen to you speak, it's clearly like a really detailed, really thorough process. I thought it was interesting within the release you mentioned that the the Gartner statistic about I think it's 40% of agentic AI projects will get cancelled or will fail. Do you think maybe that's part of the reason why the other organizations perhaps lack that that thoroughness, that attention to detail, or do you think there's other reasons why these why these projects will fail?

SPEAKER_01 15:21

Yeah, I think one of the main reasons is that that people have maybe some magical thinking. You know, you just throw an AI model at a problem and it will sort itself out. And oh, by the way, it's not just a model, we'll make it an agent, we'll give it some tools, and we'll let it chat to a hundred other agents, and all our problems will magically disappear. But that's not gonna work, you know, like you really need a predictable AI approach where you give them, where you really embed these agents uh into your workflows, give them right the right predictable tools, yeah, um, and and and also make sure that agents are tied to both customer and business outcomes. In this particular marketing case, that it's tied towards customer relevance and marketing performance. Yeah, but if you uh you can imagine you want to you could deploy uh agentic AI in all kinds of business processes, uh being it customer service or um business operations or maybe even legacy transformation, uh it's really important to to embed really tie these agents into these into these workflows, and on the flip side, also uh use agentic AI to re reimagine these workflows at design time. Yeah, and that's our predictable approach to um to agentic AI. What I spoke about here was more from the marketing angle, but in these other kind of workflow-related problems, uh uh we gone as far as saying, well, we we think that we will be better able to tie agentic AI to your business outcomes. So you're no longer paying for token cost. Yeah, we're just going to charge you on business outcomes, and we'll take the responsibility of making sure that the agentic AI works in a predictable fashion, that it's really tied to your uh to your workflows. Yeah, I think that's a different in this day of age where I don't know, a month ago token maxing was a badge of odd, you know, like you were really proud that you were spending so many tokens, and then guess what? The entropics of the world said, like, okay, we're gonna charge you by token. People go like, maybe that's not such a good idea anymore. Yeah, and and we made this kind of bold move to say, well, we think that we can actually tie a genetic AI to your workflows and to your business outcomes, but also uh are able to do that in a very cost-effective manner. And so uh when when I think more of these non-marketing type applications, uh, but same for marketing, by the way, uh, you don't pay for the tokens, you actually pay for the business outcomes.

SPEAKER_00 17:56

Yeah, on that anthropic, anthropic point, I think it's quite interesting. Obviously, like you said, it's one big company, it makes a move, it has a massive impact across the space. I think we've seen that quite a lot with the Gentec AI because it is such a new technology, it is evolving, is changing all the time. What do you guys do to kind of make sure that you stay you stay abreast of all those changes?

SPEAKER_01 18:17

Yeah, well, um uh yeah, for scientists, like in the intro, uh we explained uh um I'm heading up the AI lab, so uh, and also the lead scientist for Pega. Um, so um yeah, we have many, many uh people in the company who are trying to stay ahead of where marketing technology is is moving, but specifically also the AI lab. Of course, we look at uh the latest uh in innovations, uh not just from a big AI lab point of view, but also from a research point of view, but being really close to to business operations, really making sure that we always look through the lens of how how is this delivering business uh business outcomes. So, in that sense, we're like uh on top of the market. Also, AI for something for us is not something new, you know. We've been in in this space for uh, like I said in the intro, over 30 years. Um, so so in that sense, um, yeah, we're also trying to kind of puncture some holes into the uh, you know, there's a lot of hype around AI, you know, like uh remember that everyone was talking about AGI, and suddenly that all went away a little bit. And so we're we we we try to have like a healthy view on what is this AI technology really, also from a research point of view, but also tying it back to yeah, but what are the business and customer outcomes that we're trying to achieve here, right? So, and that's that's keeping ourselves honest as well, uh sticking really close to uh these key um uh flagship uh and lighthouse accounts that we work with. Um, I gave the examples Wells Fargo, but we have more key clients that we we stick stick with like really closely, and we work alongside so that we also get feedback really quickly in terms of uh which IDs uh I you know I might come up with a nice ID in uh here in my AI uh in my volcano where I'm petting my uh James Bond cat, you know, like uh and trying to uh come up with uh AI uh um uh world dominance, but you really need to test these things really quickly out in the market and see what works and what doesn't work, and face up to bitter enterprise reality, right? So, and that that's also what we're trying to do.

SPEAKER_00 20:34

Yeah, thanks, Peter. I think uh James Bondcart is probably the perfect place to end things today. Um, thank you very much. Really, really enjoyed that chat. You know, it's a lot, you know, Igenic AI, we all know it's it's the biggest topic in the space right now, so there's plenty of talk about it, but I think it's quite often surface level. So I was introducing opportunities to speak to people like you, the real experts, and kind of get under the bonnet a little bit and find out you know what is what it is about this tech that is really so fascinating.

SPEAKER_01 20:59

So yeah, like I'm an AI nerd, so for me, Igenic AI is really interesting, but you can see the conversation shifting towards how do we make this real and how do we make sure that we can face up to the enterprise reality. And I think that's the real interesting discussion, indeed.

unknown 21:13

Yeah.

SPEAKER_00 21:15

Yeah, absolutely, completely agree. Uh so like I said, thanks very much, Peter. Thanks for joining me. Uh, really enjoyed the chat.

SPEAKER_01 21:21

Thanks for having me.

SPEAKER_00 21:23

Yeah, anytime. Anytime. Uh, I did also want to just quickly thank our audience as well for tuning in. If you enjoyed this chat, and I'm sure you did, uh, please remember to like and subscribe to the channel and head on over to cxthoday.com for more stories like this. Until next time, thanks for watching.