Twenty questions on the six things that decide whether AI reaches production in your organisation. Your stage and profile appear on screen when you finish. If you want the full report, with a recommendation for each dimension and a 90-day plan, I'll send it.
Most AI readiness checklists ask whether you have data and a strategy. Those matter, but the questions that decide whether a pilot reaches production are narrower: who owns it, what it has to achieve, and whether anyone can prove it works.
Whether AI has one accountable owner, a budget that covers running what succeeds as well as the experiments, and a place in the business plan. A committee can advise. It can't be accountable.
Whether use cases were chosen on the measured cost and volume of the work they replace, and whether you will be able to show the board a number rather than an anecdote when the programme needs money to scale.
Whether the information a use case needs is reachable, trustworthy and owned, and whether an AI system could act on your core systems without a fresh integration project every time.
Whether anyone can say, with evidence, that the system is good enough. A written set of real questions with expected answers is the cheapest risk control in AI and the step most often skipped. It gets a dimension of its own here because it decides whether a pilot is ever signed off.
Whether risk, security and legal set the constraints at the start or veto the result at the end, whether staff know which tools they may use with which data, and what the system does when it's wrong.
Whether you have the people and engineering practices to build, run and hand over AI systems, and whether the teams whose work changes are part of the design rather than the audience for a launch email.
Your stage starts from your average across the six dimensions. A weak score on ownership, value or evaluation caps it, because those are the gaps that stop work shipping however strong everything else is.
People are experimenting, often on personal accounts. Nobody owns the outcome and no problem has been chosen on its merits.
Someone has been asked to make AI happen. Nothing has been chosen on measured value, so there is no baseline to prove a result against.
You can build, and may have a pilot that demos well. You can't yet prove it's good enough, so nobody signs it off.
Ownership, measured value, evaluation and governance are in place. The questions now are cost, consistency and which use case comes next.
Short answers, so you know what you're getting into.
No. Your stage, your six-dimension profile and the diagnosis appear on screen as soon as you finish. If you want the full report, with a recommendation for each dimension, a 90-day plan and a view on whether you need outside help, I'll email it to you.
Your answers stay in your browser unless you ask for the full report. If you do, they are stored by Kiseki Labs with your email address and I read them myself. They are never sold, and only shared with the services that run this site.
For your organisation as it is today, not as the strategy deck describes it. If parts of the business would answer differently, answer for the part you're responsible for. Where a question mentions your leading use case, that means the AI initiative furthest along, or the one you'd start first.
Partly, and I'd rather say so. If your answers suggest you don't need outside help, the report says that and points you to what to read instead. If they suggest you do, it names the engagement that fits and explains why.
Eddie Forson, founder of Kiseki Labs. I've built production AI systems since 2017, much of it inside regulated financial services. The questions reflect what blocks deployments in practice, which is why evaluation is a dimension of its own.
Thirty minutes on your situation: what you are trying to do, what is in the way, and whether the work is worth doing at all. You will get a straight answer either way, including when the answer is that you don't need me.