Most AI investment conversations in private equity still over-weight the demo. Founders and management teams show you agents, copilots, and slick workflows. The better question is simpler and more brutal: can this company actually absorb AI as labor, govern it, and convert it into operating value on the time horizon the fund cares about.
When I sit on the PE side of the table, these are the five questions I would ask before underwriting the next AI program or thesis.
1. What Operating Metric Must Move?
If management cannot identify the operating metric that needs to move, the investment case is weak. "Modernization" is not a metric. Neither is "staying competitive." Those are stories you tell when you haven't decided what you're willing to be judged on.
You are not funding an AI program. You are underwriting a change in margin, cycle time, revenue yield, or risk exposure over a defined period. Force the conversation into one lane: which specific metric, in which part of the P&L or risk stack, is this AI labor meant to move - and by how much, on what clock.
2. Who Owns the Outcome?
A shared transformation office cannot be the final answer. Program offices can coordinate work and track milestones, but they cannot own business results. Somebody has to carry the accountability for the number after AI is in production, not just the project plan on the way there.
In practice, that means naming an executive who:
- Owns the operating metric tied to the AI initiative.
- Has the authority to change workflows, staffing, and policy to make it real.
- Will still be in the firing line when the next buyer asks, "What did this actually do."
If you can't put one name next to that responsibility, you are underwriting diffusion of accountability. The technology might work. The value will not show up where you need it.
3. Can the Organization Absorb the Change?
This is where deals get overconfident. Most portfolios can buy the software, hire the integrator, and run a pilot. That says almost nothing about whether they can change decision rights, process steps, or frontline behavior fast enough to create value inside your hold period.
You want a clear-eyed view of:
- Which workflows will actually change and who owns their redesign.
- How decision rights will shift when non-human systems do more of the work.
- What will break in span of control, incentives, or skills as that happens.
An organization that treats AI as an add-on - same meetings, same approvals, same controls, plus a new tool - will not absorb it. AI will remain optional, and optional systems rarely move the numbers that matter.
4. What Governance Posture Is Required?
Not every AI use case deserves the same control structure. But every use case needs a defined governance posture. If you don't decide that up front, regulators, auditors, or the next buyer will decide it for you - usually in ways that shrink your thesis.
At minimum, you should hear a clear answer to:
- What trust tier this initiative sits in (from low-risk assistance to high-stakes automation).
- Where human review or override is mandatory, and how often it really happens.
- How incidents, near misses, and model changes are logged and reviewed.
You are not looking for a 40-page policy deck. You are looking for evidence that management understands AI as delegated labor with risk attached, and has a simple, enforceable way to keep that risk bounded.
5. What Will Due Diligence Ask About Later?
Exit readiness starts much earlier than most teams think. By the time bankers are drafting the CIM, the operating system is already built. If the company cannot show policy, controls, ownership, and documented operating impact, the next diligence cycle will treat the AI story as noise - or worse, as an unpriced risk.
Ask explicitly:
- What will a sophisticated buyer want to see to believe this AI story.
- What artifacts exist today: before/after metrics, governance decisions, exception logs, model and policy changes.
- How long it would take to produce an exit-ready AI narrative backed by evidence, not just screenshots.
If the honest answer is "we'd have to assemble it from scratch," assume a discount at exit. You're not just underwriting whether the AI works. You're underwriting whether the story will survive someone else's diligence.
What To Do Next
Before approving another AI investment, turn these five questions into a one-page diligence screen. Use it both pre-close for theses that lean on AI, and post-close before you release significant capital into AI-heavy value creation plans. If management cannot answer cleanly - in operating language, not marketing language - they are not ready for the spend they are requesting.
The point is not to slow everything down. The point is to ensure that, when you do move, you are backing AI as governed labor inside a real operating model - not chasing the best demo in the deck.