What I Told Harvard Business Review About Why Professional Services AI Is Stalling

I've had some version of the same conversation hundreds of times over the last two years. A services leader—smart, well-resourced, serious about the technology—tells me their organization has made meaningful AI investments and the returns still aren't showing up the way they expected.

What I've come to believe, and what pushed me to write a longer piece on this for Harvard Business Review, is that the return isn't hiding. The question organizations are asking is wrong.

Most organizations walked into AI asking, "How do we apply this to our business?" But that framing treats AI as a coating you apply over existing workflows. It sounds reasonable, but produces very little. 

The leaders I see breaking through asked something different: "What would we build if AI was the starting point?" Those questions lead to radically different places.

The stat that keeps me up: 95% of generative AI pilots in a recent MIT study failed to deliver measurable bottom-line impact. When I saw that alongside McKinsey's finding that only 6% of organizations qualify as genuine AI high-performers, I felt the industry needed a more honest diagnosis of what's going wrong structurally, not more speculation about what AI will eventually do.

Where the Money Is Actually Going

There's a striking consistency to where AI investments stall across the firms I talk to. It's rarely a talent problem. Rarely a vendor problem. Almost always, the stall point looks the same. The organization deployed one kind of AI and expected it to do two fundamentally different jobs.

Services delivery and services management run on different data, tolerate different error rates, and benefit from different AI architectures. Delivery work—consulting, analysis, client-facing output—is a natural fit for generative AI. Human judgment stays in the loop, probabilistic outputs are fine, and speed is the win.

Management work has no room for approximation. There is no "close enough" in revenue recognition or project margin calculation. When AI gets it wrong and a project manager has to manually audit the output to find the error, you haven't saved time. You've added a step. This verification tax is quietly devouring the ROI from hundreds of AI investments right now.

The Next Frontier

The part of this conversation I find most interesting is what happens when you build both sides well and then start connecting them.

The real unlock in professional services AI comes when a single intelligence layer can draw on delivery and management data simultaneously. Autonomous staffing decisions that factor in availability, client engagement needs, and the margin constraints that only live on the operations side. 

That's agentic AI done right. 

Getting there requires clean data and integration intentionality on both sides now, even while the two domains still operate differently. The organizations treating delivery and management as permanently separate will hit a ceiling they won't see coming.

We're building toward that convergence with Veda, Certinia's AI engine for services operations, but the architectural thinking applies regardless of which system you're on.  

If you want the full framework—including where the verification tax comes from, what high-performing organizations are doing differently, and how to think about the convergence layer—the HBR piece has it all.

Read it here: Why Professional Services Organizations Keep Solving the Wrong AI Problem