What an AI Readiness Review Actually Covers

An AI readiness review is a structured look at how your business actually works, so you can decide where AI will help and where it will not. It covers processes, data, people and risk. It does not start with a tool, and it should not end with a sales pitch for one.

Owners sometimes expect a review to be a technical audit or a demonstration of what is possible. It is closer to a survey done before building work. You want to know what the ground is like before you decide what to put on it.

Why not start with the tool?

Tools change every few months. Your processes change slowly. If I start with a product, I end up fitting your business to it, and the evidence for the choice is whatever the vendor's demo showed. If I start with your processes, the tool becomes an easy decision at the end, and it is often a simpler one than you expected.

What do I look at?

Processes. Which tasks repeat, who does them, how long they take and where they go wrong. I ask to watch the work being done, not just hear it described, because the real process often differs from the official one. The spreadsheet that only one person uses is usually where the interesting detail lives.

Data and documents. Where information lives, in what format, and who can see it. Scattered folders, duplicate versions, inconsistent naming and files that exist only on one person's laptop are common. They limit what any AI system can do, and fixing them is sometimes the most valuable outcome of the whole exercise.

People. Who owns each process, who will use a new system and who is worried about it. A technically sound project can still fail if the people affected were never asked. I want to know who will be pleased by a change and who will quietly resist it, and why.

Risk and rules. What data is sensitive, which regulations apply, and where a human must always make the call. For UK businesses that includes data protection, and for some it includes the EU AI Act. This part is not legal advice, but it shapes which opportunities are safe to pursue.

Existing tools. What you already pay for. Many businesses have AI features sitting unused inside software they already own, and the right answer is sometimes to switch one on.

What does the process involve?

A good review does not need a long engagement. It needs four things.

1. A conversation with the owner or director about goals, frustrations and anything that has already been tried.
2. Short sessions with the people who do the work, ideally while they show the task on screen.
3. A look at sample material. A handful of real emails, invoices or documents, with anything sensitive removed.
4. A written summary that pulls it together.

What do you get at the end?

A short written summary. It should rank the opportunities, say which ones to leave alone for now and why, and describe a realistic first project with a rough scope and the checks it needs. It should also list what must be tidied up first, such as consolidating a folder structure, if that would make the project cheaper or safer.

You should be able to act on it yourself or hand it to another supplier. If it only makes sense when the same person builds the result, it is not a good review.

What should you prepare?

Not much. Make the people who do the work available for a conversation, and be ready to share a few real examples with sensitive details removed. You do not need to clean your data in advance, because seeing it messy is part of the point.

What are the warning signs of a poor review?

• It recommends a specific product before it has described your processes.
• It promises a figure for savings without having timed anything.
• It never mentions what could go wrong or which tasks to leave alone.
• It is mostly a list of everything AI can do, with little about your business.

Is it always worth doing?

Not always. If you already know your first project and your data is in reasonable shape, you may be able to go straight to scoping. A review earns its cost when you have many possible ideas and no clear way to choose between them, when you suspect your data is a mess, or when you want an independent view before spending real money.