AI Readiness Assessment

AI projects tend to stall on the same three things: an unclear goal, data nobody has checked, and no one placed to judge the technical work. Bay Information Systems, an independent AI consultancy in the UK, asks these ten questions at the start of every engagement, in roughly this order.

How does the assessment work?

Ten questions, about five minutes. Choose an answer to see guidance on it and a link to further reading; you can change an answer at any time, and the result updates as you go. Seven questions are scored, and three shape the advice without affecting the score. We record which answers are chosen, without any name or contact details; those reach us only if you use one of the forms at the end.

The questions

(Answers are recorded without your name or email address, and may be used in aggregate statistics. Details are in the privacy notice.)

1. What do you want AI to do for your organisation?

A specific problem with a measurable outcome

This is where scoping starts. Put a value on the outcome next, so that scope, budget and the decision to continue can all be weighed against it.

Further reading: The cost of machine learning

We know the area, but not the exact problem

A common position. List the tasks and decisions that take the most time or money, and who repeats them each week; the first use case is usually on that list.

Further reading: Don't build models, build measurements

There is pressure to "do something with AI"

Pressure without a defined problem is how many AI projects stall. Find the problem first; the choice of technology follows from it.

Further reading: 5 mistakes to avoid when building your first AI product

2. Where does the relevant data live?

Mostly in one system

A strong position. The open question is whether that system's data is complete and consistent.

Further reading: What is data maturity?

Several systems that don't talk to each other

Each system holds part of the picture. A data model that maps one onto another is usually the first piece of work.

Further reading: What is data engineering?

Spreadsheets, documents and inboxes

Workable, though it needs structure before AI can use it reliably. The first job is agreeing which copy of each record is the true one.

Further reading: What is data maturity?

Mostly in people's heads

Common in small teams, and a risk whenever someone leaves. Writing it down pays off with or without AI.

Further reading: Don't build models, build measurements

3. Do you know what's in your data?

Yes; it's documented, and we know its gaps

Uncommon, and it means AI work can be scoped accurately.

Roughly, but nobody has written it down

A short audit turns this into a document the whole team can use.

Further reading: The data maturity checklist

No

Start here. What the data contains decides what AI can and cannot do for you.

Further reading: What data do I need?

4. Is any of the data sensitive?

No

You have the widest choice of tools, hosted AI services included.

Personal data about customers, members or staff

Check whether it can go to a third-party AI service under UK GDPR, and what the people it describes would expect.

Further reading: Sovereign AI for UK organisations

Confidential to clients, or commercially sensitive

Client contracts often restrict third-party processing. Read them before choosing a tool.

Further reading: Sovereign AI for UK organisations

Regulated (health, financial, legal) or security-related

You will probably need UK-hosted or self-hosted models.

Further reading: Sovereign AI for UK organisations

5. How do you use AI today?

Not at all

Start with a small, low-risk task and learn from it before committing to anything larger.

Staff use chat tools individually (ChatGPT, Copilot, Claude)

Set a short policy on what data can be pasted in. Anything pasted into a commercial service has left the organisation.

Further reading: Why private inference

Connected to our systems or data (connectors, agents)

Check whether the model is being asked to filter, count or calculate. Those jobs belong in ordinary code, which does them exactly, every time.

We've built something custom

The next question is how you measure whether it works, and who notices when it stops.

Further reading: Evaluating language models

6. Who can judge whether technical work is good?

An in-house lead with AI or data experience

You can direct outside help and check what it delivers.

A technical team that's new to AI

Good foundations. An experienced reviewer at the main decision points avoids expensive wrong turns.

Further reading: ML engineer vs software engineer

An outside vendor or agency

Make sure the documentation, access credentials and data sit in your own accounts, so the work survives a change of supplier.

No one

A common reason projects go wrong unnoticed. Get independent technical judgement before spending on a build.

Further reading: How to structure an ML team

7. Has an AI or data project stalled before?

No, this is our first

Then there is no earlier approach to unpick.

Yes; it stalled or was abandoned

Find out why before starting again. The cause is rarely the technology alone.

Further reading: How to structure an ML project

Yes; it's live, but disappointing

An independent review can weigh fixing it against replacing it, without the sunk cost affecting the answer.

Further reading: How to structure an ML project

8. How will you know if it works?

We have an agreed measure and a way to check it

That makes every later decision easier, including the decision to stop.

Staff will try it and tell us

Using the tool is the best early test. Ask people to note where it is right and where it is wrong; that list becomes your automated test set later.

Further reading: How to evaluate an AI pipeline

We haven't thought about it

Decide before building, or there will be no basis for deciding whether to continue.

Further reading: The ML measurement checklist

9. What should happen with the AI's output?

It triggers an action in a system or workflow

Data in, action out, and a change you can measure: the strongest shape for an AI project.

People read it to inform decisions

Useful, though reports are easy to ignore. Decide who acts on each output, and what they do with it.

Not sure

Work this out before anything else; it shapes the whole design.

10. What happens if you do nothing for a year?

We can put a number on it

That number is your budget ceiling and the core of your business case.

Further reading: The cost of machine learning

It gets worse, but we haven't costed it

A rough figure is enough to decide how much to invest.

Further reading: The cost of machine learning

Not much

Then it may not be the right time, which is a valid conclusion.

Your results

Scoring runs in your browser and needs JavaScript. The printable checklist below lists every question and answer.

What does an AI readiness assessment cover?

An AI readiness assessment checks whether an organisation has what an AI project needs before any money is spent on one. The questions run in a fixed order because each answer limits the next.

The goal comes first. A problem with a value on it sets the budget, the scope and the point at which to stop, and without one nothing later in the assessment can be weighed. The data comes second: where it lives, whether anyone knows what it contains, and whether it is sensitive. Sensitivity decides where a model can run, and so which tools are available at all (personal data under UK GDPR, client confidentiality and regulated records each narrow the choice). The people come third: who can judge whether technical work is good, and whether an earlier project stalled. The last questions cover measurement, what happens to the output, and what a year of waiting would cost.

Technology has no question of its own. It follows from the answers.

How do you run an AI readiness assessment internally?

The questions work as a meeting agenda. Answer them for one use case at a time, since a goal that is clear for one team is often vague for another, and bring three people: whoever owns the problem, whoever knows the data, and whoever would build or buy the system. Where they disagree on an answer, write both down.

The answers about data need checking against the data itself. "Mostly in one system" remains a claim until someone pulls a sample and counts the gaps, and that check usually takes longer than the meeting. Answers of "not sure" or "we haven't thought about it" are findings in their own right; they mark the work to do before any build. (The printable version below serves as a template for the meeting.)

How does it differ from an AI audit?

The two terms are used loosely, and some providers sell one under the other's name. On this site, an AI audit examines AI already in use (which tools staff use, what data reaches them, whether a deployed system does what it was bought to do), and a readiness assessment examines a decision that has not yet been made. They meet at the questions on current use and sensitivity: data already pasted into a chat tool is an audit finding and a readiness constraint at once.

What does a consultant-led assessment add?

This page gives a score and guidance on each answer. The paid version, an AI roadmap, covers the same ground in more depth: one week, interviews with three key people, a direct look at the data, and independent judgement on any technical claims. It produces a short written roadmap setting out the goal in business terms, what is at stake, a phased plan with a fixed price and a go, adjust or stop decision after each phase, and what not to do. The roadmap costs £8,000, fixed and agreed before work starts (the forms below reach us directly, and the first conversation is free).

Can I print the questions as a template?

Yes. The version below lists every question with the points for each answer. Questions marked "no points" change the advice without changing the score.

  1. What do you want AI to do for your organisation?
    • [ ] A specific problem with a measurable outcome (2)
    • [ ] We know the area, but not the exact problem (1)
    • [ ] There is pressure to "do something with AI" (0)
  2. Where does the relevant data live?
    • [ ] Mostly in one system (2)
    • [ ] Several systems that don't talk to each other (1)
    • [ ] Spreadsheets, documents and inboxes (1)
    • [ ] Mostly in people's heads (0)
  3. Do you know what's in your data?
    • [ ] Yes; it's documented, and we know its gaps (2)
    • [ ] Roughly, but nobody has written it down (1)
    • [ ] No (0)
  4. Is any of the data sensitive? (no points)
    • [ ] No
    • [ ] Personal data about customers, members or staff
    • [ ] Confidential to clients, or commercially sensitive
    • [ ] Regulated (health, financial, legal) or security-related
  5. How do you use AI today? (no points)
    • [ ] Not at all
    • [ ] Staff use chat tools individually (ChatGPT, Copilot, Claude)
    • [ ] Connected to our systems or data (connectors, agents)
    • [ ] We've built something custom
  6. Who can judge whether technical work is good?
    • [ ] An in-house lead with AI or data experience (2)
    • [ ] A technical team that's new to AI (1)
    • [ ] An outside vendor or agency (1)
    • [ ] No one (0)
  7. Has an AI or data project stalled before? (no points)
    • [ ] No, this is our first
    • [ ] Yes; it stalled or was abandoned
    • [ ] Yes; it's live, but disappointing
  8. How will you know if it works?
    • [ ] We have an agreed measure and a way to check it (2)
    • [ ] Staff will try it and tell us (1)
    • [ ] We haven't thought about it (0)
  9. What should happen with the AI's output?
    • [ ] It triggers an action in a system or workflow (2)
    • [ ] People read it to inform decisions (1)
    • [ ] Not sure (0)
  10. What happens if you do nothing for a year?
    • [ ] We can put a number on it (2)
    • [ ] It gets worse, but we haven't costed it (1)
    • [ ] Not much (0)

Score out of 14: 0-5, foundations first; 6-10, ready to test; 11-14, ready to build.

www.bayis.co.uk/ai-readiness-assessment.html

Frequently asked questions

What is AI readiness?

AI readiness is whether an organisation has what an AI project needs before it starts: a defined problem with a value on it, data that is known and reachable, someone able to judge the technical work, and a way to tell whether the result works. The choice of technology comes later and depends on those four.

How long does a full assessment take?

It depends on how many systems hold the relevant data and how many people we need to speak to. We agree the length and the fee in writing before any work starts.

What does a full assessment cost?

The first conversation is free. The assessment itself is fixed price, agreed before it starts.

What do we receive?

A data strategy: a short document setting out your situation, the goal in business terms, what is at stake if the work goes wrong, a phased plan with a fixed price and a go, adjust or stop decision after each phase, and what we recommend not doing. It is the same document our data strategy consulting work produces, scoped to the AI decision in front of you.

Are my answers sent anywhere?

We record which answers are chosen, without any name or contact details, to see how the assessment is used. Your contact details, and your answers linked to them, reach us only if you use one of the forms at the end.