The Enterprise AI Readiness Self-Check

Most enterprise AI never ships. It demos well, gets a budget line, and then stalls somewhere between the pilot and the first real user. The pattern is consistent enough to predict: the organisations that cross into production share a named executive sponsor, a success metric tied to a business outcome, and at least one internal champion. The ones that don't, don't.

This is a self-check, not a maturity certificate. Answer each question honestly with yes, no, or not sure — a "not sure" counts as a no. Twelve questions, five dimensions. Score yourself at the end.

1. Is there a named executive sponsor who owns the outcome — not just the budget? A sponsor who can unblock a data-access fight or a procurement freeze is worth more than the model.
2. Is the first use case tied to a metric someone already reports on? "Reduce handle time," "recover abandoned carts," "cut Rx leakage" beat "explore AI." If it doesn't move a number a director already watches, it won't survive the next planning cycle.
3. Can you name the internal champion who will use the thing every day? Pilots without a daily user are demos.

1. Do you know where the data lives, who owns it, and whether you're allowed to use it for this? Data readiness is the single most-cited cause of AI failure — usually access and lineage, not volume.
2. Is the data good enough to be wrong about in production? You don't need perfect data. You need to know its failure modes before a customer finds them.

1. Is there someone on the team who can tell a good output from a plausible-but-wrong one? Evaluation is a skill, not a dashboard.
2. Do the people around the system — support, ops, compliance — understand what it can and can't do? Enterprise-wide AI literacy is what keeps a model from being quietly switched off after the first bad week.

1. Can the system reach the systems it needs to — read and write, by role? Read access is a demo. Writing records, triggering workflows, and surfacing data by permission is production.
2. Is there a path to monitor it in production — latency, cost, drift, and a way to roll back? If you can't see it degrade, you can't keep it alive.

1. Have you classified this system under the EU AI Act? The four risk tiers decide your obligations. High-risk provider duties bind from 2 August 2026 — this is not a 2027 problem.
2. Is there human oversight built into the architecture, not bolted on after? Someone must be able to understand, override, and detect anomalies — by design.
3. Do you have one accountable owner for AI governance? Not a committee that meets quarterly — a name.

Count your yes answers.

• 0 to 4 — Not yet. You have an idea, not a plan. Start by naming a sponsor and a metric; everything else is downstream of those two.
• 5 to 8 — Pilot-ready, not production-ready. You can run a credible pilot, but the gaps — usually data lineage, integration depth, or governance — are exactly where pilots die. Close them before you scale.
• 9 to 12 — Ship it. You have the conditions for production. The remaining risk is execution discipline, not readiness.

Where you scored low is where an embedded product lead earns their keep — turning a stalled pilot into a measured outcome is most of what I do. If three of these questions made you wince, that's usually the conversation worth having.