Every digital leader we have spoken to this year has been asked the same question by their board: what are we doing about AI? The honest answer, in most organisations, is “buying licences and hoping”. A readiness health check is the alternative. It is a short, independent look at whether AI would actually help, and what has to be true before it can. These are the twelve questions we work through.
Services and users
1. Where do users wait, repeat themselves or give up? AI is a tool for removing friction. If you cannot point to the friction, you are not ready to choose a tool. Journey maps and failure-demand data answer this quickly.
2. Which of those moments are judgement and which are process? Process can often be automated. Judgement usually needs a person, with AI as support. Confusing the two is how services end up with an assistant that confidently gives the wrong answer.
3. What would “better” look like, in numbers? Fewer calls, faster decisions, higher completion. If nobody can say what success is, a proof of concept cannot prove anything.
Data
4. Does the data the feature would need actually exist? Not in a strategy document. In a system, with a field name, updated regularly.
5. Is it accurate enough to act on? Most organisations discover the state of their data when they first try to use it for something new. Better to find out in a two-week health check than a two-year programme.
6. Are you allowed to use it for this? Data collected for one purpose cannot always be used for another. For public bodies this is a data protection impact assessment question, and it needs answering early.
People
7. Who would own the service once it is live? Models drift, prompts need tuning and quality has to be reviewed. An AI feature without an owner is an incident waiting to happen.
8. What can your teams do today? You may already have people who can prompt, evaluate and prototype. You probably do not have people who can monitor bias or write a transparency record. Knowing the gap tells you what to hire, buy or train.
Risk and governance
9. What happens when it is wrong? Every AI feature is wrong sometimes. The design question is whether the user can tell, recover and reach a human. The governance question is whether anyone will notice.
10. Could an assessor or regulator follow your reasoning? For government teams this means the Service Standard and the Algorithmic Transparency Recording Standard. For everyone else it means being able to explain a decision to the person it affected.
Technology
11. What do you already pay for that has AI capability you are not using? Many organisations have assistants, search and summarisation sitting unused inside platforms they already licence. Switching those on is cheaper than a new procurement.
12. What is the smallest proof of concept that would settle the argument? The answer is usually a prototype in front of ten users, not a platform. If you can describe that experiment, you are ready to start.
What a health check gives you
We score each of the five areas red, amber or green, list the three to five opportunities most likely to pay back, name the risks to resolve first and write a 90-day plan with owners. It takes two to three weeks and it usually saves a great deal more than it costs, mostly in projects that never start. If you would like one, get in touch or read about the AI readiness health check.
