Wolf&Grayson EditorialBoards have stopped asking whether their companies should do something about AI and started asking what the spending has produced. That shift has exposed an uncomfortable truth across the enterprise market: a great many AI programs were approved on the strength of a demonstration rather than a thesis, and the people who approved them never defined what success would look like. Pankaj Kumar, a technology executive who has held profit and loss (P&L) responsibility and led global delivery at scale, argues that the discipline problem sits at the top of the organization, not in the engineering team. The question is not whether the technology works. It is whether the executives funding it can explain, in business terms, why this capability, in this workflow, is worth the money and the risk.
The Question That Should Open Every Investment Review
Kumar’s opening test is deliberately blunt: “What business outcome are we prepared to own and why is AI the right way to achieve it?” The phrasing does two jobs at once. It demands a concrete outcome, and it assigns a name to it. “We need an AI strategy” is intent, not a decision. Reducing the time it takes to launch a product, improving the quality of a customer decision, or serving a market that was previously uneconomic is something a finance team can evaluate and an executive can be held to.
What follows from that question is a short list most proposals never contain: the current baseline, the improvement that would matter, the executive responsible for delivering it, and an honest answer to what happens if the company does nothing. That last item is where Kumar separates two very different kinds of spending. “Some investments improve today’s performance; others build capabilities that could determine tomorrow’s competitive position.” Conflating the two is how companies end up funding a strategic bet on an efficiency timeline, then killing it when the quarterly numbers do not move. Certainty is not the standard. “You need a clear reason to invest and a credible way to learn whether that reason holds.”
Where The Narrative Pulls Teams Off Course
The failure mode Kumar sees most often is simple and seductive. “The easiest mistake is allowing a compelling demonstration to substitute for an investment thesis.” A demo shows what is possible under conditions someone else selected. An investment decision requires confidence that the capability produces value repeatedly, inside your business, at an acceptable cost and level of risk. Those are not the same claim, and the gap between them is where budgets disappear.
Competitive anxiety makes it worse. When a board asks what the company is doing about AI, the instinct is to answer with volume: a portfolio of initiatives that proves activity. Kumar would reverse that. “I would rather see a small number of consequential opportunities with explicit assumptions.” Disciplined evaluation tests those assumptions against real workflows, representative data, and difficult cases, and it prices integration, human review, adoption, and operating costs alongside model performance. It also sets, in advance, the conditions for expanding, changing, or stopping the investment. “The strongest executive teams can be ambitious about AI while remaining precise about what the evidence actually supports.” Ambition and precision are not in tension. Treating them as if they were is what produces both the overfunded pilot and the paralyzed competitor.
The Gap Between A Business Case And Realized Value
The distance between a promising case and working execution rarely shows up as one large failure. It accumulates. “In my experience, the gap appears in the dependencies that the business case assumes somebody else will resolve,” Kumar says. The data exists, but access is fragmented. The system produces useful outputs, but employees still verify every one. A workflow gets faster while the next step stays a bottleneck. Individually, each looks manageable. Together, they can undermine the economics entirely.
Then there is the accounting problem that quietly inflates most AI business cases. “Hours saved are not automatically financial value realized,” Kumar says. Released capacity has to be converted into greater throughput, better service, additional revenue, or lower cost, and someone has to own that conversion. Without a plan, “the benefit stays in the spreadsheet.” Enterprise adoption, in his read, depends on a chain of capabilities running from models through data, integration, domain expertise, and organizational change, with clear ownership at every link. His test for credibility is practical: “The business case becomes credible when the people responsible for operating the change help design it.”
That same logic reshapes how proposals should be framed under CFO and board pressure. Kumar separates investments by the return they are meant to produce. Efficiency investments demonstrate improvements in cost, speed, or quality. Growth investments test customer adoption, willingness to pay, or commercial performance. New capability investments need a defined learning objective, a capped budget, and a decision date. Different criteria, same financial discipline: visible assumptions, the full cost of implementation and operation, and a stated threshold of evidence that unlocks the next tranche. He warns against the obvious trap of making cost reduction the only acceptable return, which “can lead a company to automate its existing business while underinvesting in how that business could evolve.”
The next shift is already visible. Kumar expects leaders to stop treating approval as the finish line and start managing AI as an ongoing operating capability, with performance, economics, and permitted scope under continuing review. That matters most as systems gain authority to act. “An assistant that recommends a decision and an agent that executes it require different controls.” Executives should be explicit about what a system can do, when a person must approve an action, how errors are detected, and who can intervene, with business outcomes, reliability, cost, and human oversight examined in the same management review. Governance, on this view, is not a brake. It determines where greater autonomy is justified by evidence. Underneath it sits a design choice with real strategic weight: preserve the ability to adopt better models and tools without rebuilding the business process around them. Kumar’s bias is toward what survives the churn, “evaluation capabilities, usable data, well-designed workflows and customer trust that can become more valuable as AI improves.”
Follow Pankaj Kumar on LinkedIn for more insights on AI investment strategy, enterprise execution, and technology governance.
