AI readiness checklist

74% of companies have yet to show tangible value from AI, according to BCG's 2024 survey of 1,000 senior executives across 59 countries. The same research found that the companies getting results put 70% of their resources into people and processes, and only 10% into the algorithms themselves.

That ratio is the whole point. AI does not fix a mess. It runs on top of one and produces confident output from bad inputs.

This checklist has three parts. Part one scores whether your business is ready. Part two finds which of your processes, if any, are genuinely AI-shaped. Part three tells you what to do with both answers together.

Part one: are you ready?

Fifteen statements across three areas. Give yourself a point for each one that is true, and be strict. A qualified yes is a no.

Your data

• You can list every system holding operational data, including the spreadsheets
• The information that matters sits in fields, not in email threads and attachments
• Where the same information exists in two systems, you know which one is authoritative
• Nothing business-critical lives only in one person's head or inbox
• You could pull the full history of a client or project without asking someone to compile it

Governance

• Every significant dataset has a named owner
• Someone decides what data may leave your systems, and people know who that is
• You know what your client contracts permit you to process this way
• You know what personal data you hold and what GDPR allows you to do with it
• If AI produced something wrong and a client acted on it, you know who is accountable

Your people

• Someone internally will own this after it is built
• Your team can describe their own processes accurately when asked
• There is a stated position on what staff may and may not put into AI tools
• Leadership uses the systems they ask everyone else to use
• A previous change to how people work actually stuck

Your score is out of 15. Twelve or above means you are ready. Below twelve means there is preparation to do first, and how much depends on how far below. A ten is a few weeks of tidying. A five is a project in its own right.

Part two: which processes are AI-shaped?

This part is not scored. It produces a shortlist, and for most businesses that shortlist is one or two processes rather than a program of work.

Step one: list the candidates

Answer these and write down the processes that come up.

• Which step do things wait at most often?
• Which tasks can only one person do?
• Where does work stop and wait for a human to move it between systems?
• What happens at month end that everyone dreads?
• Which recurring task consumes the most hours across the whole team, rather than per person?

Then check your list against reality:

• Ask the people doing the work what wastes their time. Their answer is usually different from leadership's.
• For each candidate, is the problem volume, accuracy, speed, or consistency?
• What is it costing you now, in hours or money?
• Would solving it change a business outcome, or would it just feel better?
• Has it survived a previous attempt to fix it? Find out why before trying again.

Anything that costs you nothing measurable, or that only feels bad, comes off the list.

Step two: test each survivor

Most of what people want AI for is not an AI problem. Rules-based automation is cheaper, faster, predictable, and it does not invent things. Run each remaining process through these:

What survives is your shortlist.

Part three: what to do with your results

Put your score next to your shortlist.

Twelve or above, with a shortlist. Take one process, not three. Write down what it costs today and what better looks like as a number. Decide who checks the output and what they are checking for. Decide what happens when it is wrong. Then run it alongside the existing process for one cycle before replacing anything, and set a date to review whether it earned its place.

Twelve or above, no shortlist. You do not need AI right now. That is a legitimate result rather than a failure of the exercise. Your operation is in order and nothing in it is AI-shaped. Revisit when volumes change or when a process appears that involves reading a lot of unstructured text.

Below twelve, but part two found real pain. The pain is real and AI is not yet the answer to it. Fix your lowest-scoring area first. There is a good chance the bottleneck resolves on the way, because a lot of what looks like an AI problem turns out to be a data or process problem wearing a costume.

Below twelve, nothing specific in part two. Do not start. At this point you are responding to pressure to have an AI story rather than to a problem you can name. Come back when you can name the process and what it costs.

Where to get help

If your score says the foundations come first, that is the work we do: getting data, ownership and processes into a state where automation and AI have something solid to run on. If you have a shortlist and want it built properly, with the checks and fallbacks in place, we do that too.