This article was based on the interview with Pega CFO and COO Ken Stillwell on what’s actually inside your AI budget by Greg Kihlström, AI and MarTech keynote speaker for The Agile Brand with Greg Kihlström podcast. Listen to the original episode here:
It’s a familiar scene in budget meetings across the enterprise. There it is, a single line item, deceptively simple: “AI.” It sits there, pregnant with promise and, if we’re being honest, a fair amount of ambiguity. As a marketing leader, you’re under immense pressure from the board, your C-suite peers, and your own team to invest in it, to innovate, to *do something* with AI. But what does that line item truly represent? Is it a sophisticated predictive model to optimize next-best-action? A simple rules-based engine that has been rebranded for the current moment? Or is it a series of generative AI pilots that, while interesting, have no clear path to the P&L?
The truth is, “AI” is not a monolith. It’s a catch-all for a vast spectrum of technologies, each with a wildly different cost, risk profile, and potential for return. The challenge for today’s leader isn’t a lack of opportunity; it’s a lack of discipline. The real agility isn’t about moving faster on AI, but about being able to discern what’s truly being proposed, apply rigorous financial and operational thinking, and have the courage to shut down the experiments that aren’t working. In a recent conversation with Ken Stillwell, the CFO and COO of Pega, we explored what it takes to move from the hype cycle to building a durable, value-driven AI strategy. His perspective, forged at the intersection of finance and operations, provides a much-needed dose of pragmatism for any leader navigating this landscape.
AI Is Not a Monolith: The Case for a Central Nervous System
Before you can measure ROI, you have to know what you’re measuring. When a request lands on your desk for an “AI” investment, the first step is to deconstruct what that actually means. A generative AI tool for copywriting has a different cost structure (token-based) and risk profile (brand voice, hallucinations) than a predictive AI model for churn, which requires clean data and statistical expertise. One of the first breakdowns in many organizations is that these disparate initiatives are happening in silos, leading to fragmented efforts, redundant spending, and a Swiss cheese approach to governance.
Stillwell advocates for a centralized approach, not to stifle innovation, but to channel it effectively. This is less about top-down control and more about creating a shared intelligence hub.
“One of the things that we’ve done at Pega is we’ve tried to create some centralization of anything that comes through around AI… we try to channel that through this AI steering committee, so to speak, that we have… We kind of started from the very beginning with AI. Very similar to the way it happened when the cloud came on, where people would try to get credit cards and get an AWS instance, and then you get to some point where you say, ‘Well, hold on a second. We can’t have cloud environments everywhere. We have to centralize that.’”
The parallel to the early days of cloud adoption is apt. Just as “shadow IT” for cloud services created security risks and budget overruns, “shadow AI” can lead to inconsistent customer experiences, data privacy issues, and a portfolio of pilots that never scale. For marketing leaders, establishing a cross-functional steering committee—with representation from IT, finance, legal, and other business units—is a critical first step. It ensures that when your team wants to pilot a new AI-powered personalization engine, the conversation starts with a common understanding of the technology, its potential risks, and how it aligns with the enterprise’s broader data and governance strategy.
The Most Expensive Tool Is the Wrong One
There’s a temptation, fueled by an endless stream of product demos, to believe that the newest, most powerful generative AI model is the answer to every problem. Need to send a follow-up email? Have GenAI write it. Need to populate a field in your CRM? GenAI. But this approach is not only inefficient; it can be wildly expensive and unnecessarily risky. A core tenet of operational excellence is using the right tool for the job, and this has never been more true than with AI.
Stillwell offers a simple, powerful thought experiment that should be a litmus test for any proposed AI use case.
“When you have things that a CPU could do or a task or a rule or a workflow could do incredibly inexpensive, why would you use AI to do that? You wanna use AI for the things that can’t be done through a workflow… people using AI to do routine, structured, rules-based or workflow-based tasks, where you… set up the workflow and it’s incredibly cheap to do billions of transactions afterwards. Why would you wanna do that with AI?”
This is a critical insight for marketers. Your marketing automation platform is already excellent at executing rules-based campaigns. Your decisioning engine is built to handle next-best-action logic at scale and at a low cost per transaction. Using a large language model to perform these tasks is like using a sledgehammer to hang a picture frame. It’s not just about the direct cost of tokens; it’s about the inherent risk of introducing variability where you need predictability. In highly regulated industries like finance or healthcare, using generative AI to create communications that could easily be handled by a template-driven, rules-based system introduces compliance risk that is entirely avoidable. The smart play is to reserve the power—and expense—of generative AI for tasks that truly require it: summarizing unstructured data, generating novel creative concepts, or powering sophisticated conversational interfaces.
Adopt a Product Manager’s Mindset
The initial frenzy around generative AI led to a proliferation of pilots. Everyone, it seemed, was experimenting. While that spirit of discovery is valuable, it can quickly devolve into a series of “science projects” that look impressive in a slide deck but deliver no tangible business value. The difference between a successful pilot and a dead end often comes down to the mindset with which it was initiated. It’s the difference between being a tinkerer and being a product manager.
Stillwell argues that every AI initiative, no matter how small, should be approached with the same rigor as developing a new product.
“If you have people that are not thinking like a product manager around the part of their business, they will just do things in AI because they can get some whiz bang thing… I would advise companies to think more around what problems are you trying to solve, and can AI be helpful in that? And if so, what would the pilot look like to prove that out? I think you’ll get a lot more value from your spend and, quite frankly, your management focus.”
For marketing leaders, this is a call to action. Before approving any AI pilot, ask your team the fundamental product management questions: What specific customer or business problem are we trying to solve? Is AI the best way to solve it? What does success look like, and how will we measure it? What is the path from a successful pilot to an at-scale, operationalized solution? This shifts the focus from “Look at this cool thing AI can do” to “Here is how this AI-powered solution will reduce customer churn by X%” or “Here is how this will improve our lead qualification process and increase MQL-to-SQL conversion by Y%.” It grounds experimentation in business outcomes and ensures that your team’s valuable time—and your budget—is focused on initiatives that matter.
Rewarding the Courage to Stop
In the current climate, there is immense pressure to demonstrate progress on AI. No one wants to be the executive who tells the board their AI initiatives aren’t panning out. This creates a dangerous incentive structure where projects are kept alive long past their expiration date, consuming resources that could be better deployed elsewhere. True leadership requires not only the vision to start new things but also the discipline to stop them.
Stillwell suggests that organizations need to reframe their perspective on what constitutes success, creating a culture where shutting down a failed experiment is seen as a smart business decision, not a personal failure.
“I give a lot of credit to people that actually can acknowledge, you know, I thought AI was gonna help me here, but it isn’t actually going to be the case. I’m gonna shut that down and shift resources over to another project… I think boards and executives should reward team members for having that thought process of, like, where’s the real value, and not view it as, ‘Well, you couldn’t make every use case work, so you must have failed.’ Of course we were gonna guess wrong on some things.”
This is perhaps the most difficult—and most important—piece of the puzzle. As a leader, you set the tone. If your team sees that the only path to recognition is launching a splashy AI project, that’s what they will do. But if you actively celebrate and reward the team that runs a tight, well-defined experiment, honestly concludes it won’t deliver the expected ROI, and quickly reallocates their efforts to a more promising venture, you cultivate a culture of intellectual honesty and financial discipline. It sends a clear message that the goal isn’t just to “do AI,” but to create value. Sometimes, the most valuable action you can take is to free up resources from a losing bet to double down on a winner.
From AI Spending to AI Investing
The journey with enterprise AI is moving into its next, more mature phase. The initial period of boundless experimentation is giving way to a necessary period of rationalization and accountability. The organizations that thrive will be those that treat AI not as a magic wand or a budget category to be checked off, but as a powerful set of tools to be deployed with precision, purpose, and a relentless focus on the bottom line. It’s about shifting from simply spending on AI to truly investing in it, with all the rigor and expectation of return that the term implies.
For marketing leaders, this is your moment to lead. By establishing centralized governance, demanding clarity on the right tool for the job, instilling a product manager’s mindset, and building a culture that has the courage to stop what isn’t working, you can transform that ambiguous line item in your budget into a strategic portfolio of high-impact initiatives. The future doesn’t belong to the companies that do the most with AI; it belongs to the ones that do the smartest things with it. And that requires a leader who knows what’s actually inside their AI budget.