The AI Readiness Question Most Executive Teams Are Not Asking

Every board I talk to is asking some version of the same question: what is our AI strategy? It is a reasonable question. It is also the wrong place to start. Strategy decks are easy to produce. What separates the companies that get value from AI from the ones that stall out is a different question entirely: what happens operationally when this works?

That question sounds like a luxury problem. It is not. If your pilot succeeds and you have no answer for how to run it at scale, on call, under audit, at a defensible cost, the pilot did not succeed. It just created a new liability with a good demo.

I spent 25 years running infrastructure and reliability at Apple, AWS, PayPal, AppDynamics, and Fannie Mae. I scaled AppDynamics through its acquisition by Cisco and led platform and reliability work across more than 600 applications at Fannie Mae. In all of that time, I never saw a production incident caused by a strategy document. I saw plenty caused by systems that were promoted to production before anyone asked who would carry the pager.

AI is following the same pattern. Executive teams debate which model to use, which vendor to trust, which use case to pilot. Those are real decisions, but they are the small ones. The models are increasingly interchangeable. The organization that has to operate the result is not.

A pilot proves that something can work once, in a controlled setting, with the most motivated people in the company watching it closely. Production is a different contract. Production means the system works when nobody is watching, degrades gracefully when a dependency fails, and produces an answer you can defend when a regulator or an auditor asks how it made a decision.

When an executive team asks me whether they are ready for AI, I do not ask about their model choices. I ask a handful of operational questions, and the answers tell me almost everything.

First: when this system gives a wrong answer at 2 a.m., who gets paged, and what runbook do they open? If the answer is a shrug, you have a demo, not a product. Second: can you explain, to an auditor, why the system produced a specific output for a specific customer? In a regulated business, that answer cannot be improvised after the fact. I spent enough years in financial services to know that the audit question arrives whether or not you prepared for it.

Third: what does this cost per transaction at ten times today's volume, and does the unit economics still work? Inference costs that look trivial in a pilot have a way of becoming a line item the CFO wants to discuss. Fourth: if the two people who built this leave, who maintains it? A system only one team can operate is a resignation letter away from being an orphan.

Fifth, and the one that gets the longest silence: what is your rollback plan? Every mature operations organization can undo a bad deploy. Very few AI initiatives can undo a bad model, a bad prompt change, or a bad data pipeline update with the same confidence.

Teams that are genuinely ready treat AI systems like any other production dependency. They have service level objectives for quality, not just uptime. They have monitoring that catches drift before customers do. They budget for evaluation and retraining the way they budget for capacity. They put AI changes through the same change management discipline as any other release.

None of this is exotic. It is the same operational maturity that reliable companies already apply to payments, to trading systems, to anything where failure is expensive. The mistake is assuming AI is exempt because it is new. It is not exempt. It is harder, because the failure modes are quieter.

My advice to executive teams is simple. Before you approve the next AI initiative, ask what happens if it works. Write down who operates it, who audits it, what it costs at scale, and how you turn it off. If those answers come easily, fund the pilot. If they do not, the readiness gap is not in your models. It is in your organization, and no vendor can close it for you.

The companies that win with AI over the next five years will not be the ones with the best strategy decks. They will be the ones that could run what they built.