Before the Budget Vote: What Directors Should Ask About AI Spending This Year
Photo: LA MISERICORDIA ASILO NACIONAL MASONICO, Public domain, via Wikimedia Commons
Across corporate America, board agendas are filling with AI proposals. Some are thoughtful. Many are vague. Nearly all arrive with the implicit message that delaying approval means falling behind. Yet the questions that matter most—about lock-in, readiness, accountability, and realistic returns—rarely surface before the vote. They surface later, in the form of write-offs.
Directors do not need to become technologists to govern these investments well. They need to insist on five conversations before budget season closes.
1. What Happens If We Want to Leave?
Vendor lock-in has returned with new vocabulary. AI platforms increasingly embed themselves into workflows, data pipelines, and contracts in ways that make exit expensive and switching slow. Ask management to describe, in plain terms, the cost and timeline of unwinding each proposed commitment. If the answer is a shrug, the risk has not been priced.
Discussion guide: Request a one-page exit analysis for every AI contract above the materiality threshold, including data portability terms and renewal escalation clauses.
2. Is Our Data Actually Ready?
Most AI disappointments are data disappointments wearing a different label. Models inherit the fragmentation, inconsistency, and gaps of the systems beneath them. Boards should ask what percentage of the data feeding a proposed initiative has been assessed for quality and completeness—and what happens to the business case if that number is lower than assumed.
Discussion guide: Require a data readiness assessment as a precondition for any AI capital request, with findings reported alongside the proposal rather than after deployment.
3. Who Owns Model Risk?
When an AI-assisted decision produces a bad outcome—a discriminatory hiring result, a flawed credit decision, a hallucinated client report—who inside the enterprise is accountable? In many organizations, the honest answer is nobody, which in a regulatory environment as active as the United States is a governance failure waiting for a headline.
Discussion guide: Ask for a named executive owner of model risk, a defined escalation path, and confirmation that board reporting on AI incidents will occur at a set cadence.
4. What Is a Realistic Return Timeline?
AI business cases frequently promise transformation in twelve months. The evidence suggests value typically arrives in years, not quarters, and only where workflows are redesigned around the technology. Boards that accept the optimistic timeline set up management to fail—or to massage the numbers.
Discussion guide: Ask each sponsor to present a phased ROI curve with explicit checkpoints at which the investment would be expanded, held, or halted.
5. Are We Funding Adoption or Just Acquisition?
The pattern that defines wasted technology spend is well established: heavy investment in acquisition, minimal investment in adoption. AI is no exception. The budget line for change management, training, and workflow redesign should be proportional to the license cost. When it is not, the purchase is a down payment on shelfware.
Discussion guide: Require every AI proposal to allocate a stated percentage of total cost to adoption and measure utilization at six months against a board-approved target.
The Common Thread
None of these questions require technical expertise. All of them require a board willing to slow a fast-moving conversation long enough to ask who is accountable, what has been assumed, and how anyone will know whether it worked. That is not resistance to innovation. That is the job—and the enterprises that do it well are discovering that disciplined AI governance is a competitive advantage, not a brake.
