AI readiness means your business could put AI into production and keep it there: the data an idea needs exists and is usable, someone can say what the AI must achieve in numbers, and a person can check its output. An AI readiness assessment measures those things before money is spent, so the first project is chosen on evidence rather than enthusiasm.
Most frameworks you will find online end with "contact us to find out your score". This one includes the actual test. Twelve questions, fifteen minutes, scored bands at the end, and you do not have to tell us your answers.
The five pillars of AI readiness
Every serious framework converges on roughly the same territory. We assess five pillars, and the order is deliberate: the ones that kill projects most often come first.
| Pillar | The question it answers | Kills projects when |
|---|---|---|
| Data | Does the data each idea needs exist, and is it usable? | The data lives in three systems that disagree, or in nobody's field |
| Process | Is the process being automated actually written down? | The AI inherits confusion instead of removing it |
| People | Who checks the output, and who owns the system after launch? | Nobody is named, so quality drifts unwatched |
| Measurement | Can success be stated as a number before building? | "Feels better" is the only metric, so nothing can be defended |
| Economics and risk | Does the volume justify the cost, and what does a wrong answer cost? | A low-volume process gets a high-fixed-cost solution |
Notice what is not a pillar: model choice, vendor choice, and every other question that dominates the marketing. Those are decisions you make after readiness, not parts of it.
The 15-minute self-test
Score each question 0, 1 or 2. Be harsh; the test only works if the answers are true. Definitions: 0 means no or nobody knows, 1 means partially or informally, 2 means yes and it is written down or measurable.
Data
- For your best AI idea, does the data it needs exist in a system you control, rather than in email threads and spreadsheets?
- Would two people pulling that data independently get the same numbers?
- Is the data allowed to be used this way, considering customer consent, retention rules and anything a regulator or enterprise customer would ask?
Process
- Is the process you want to improve written down, step by step, as it actually runs today rather than as it was designed?
- Could you name the three most common exceptions and what people currently do about them?
People
- Is there a named person who could check the AI's output and say "right" or "wrong" case by case?
- Is there a named owner for the system after launch, someone whose job includes noticing when it degrades?
Measurement
- Can you state what success looks like as a number, minutes saved per case, percentage resolved, error rate, before anything is built?
- Do you know today's baseline for that number, or could you measure it within two weeks?
Economics and risk
- Does the process run at real volume, hundreds of cases a month or more, rather than a task someone does occasionally?
- Can you state what a wrong answer costs, and would that cost be survivable while the system is bedding in?
- If the AI vanished after six months, would the work carry on, meaning a person or fallback still exists?
What your score means
18 to 24: ready. Your constraint is choice, not readiness. Pick the highest-value idea, capture the baseline, and build. A formal assessment would add confidence but probably not change the answer.
10 to 17: ready in parts, and this is most businesses. Usually the pattern is strong on economics and people, weak on data and measurement. The correct next step is not a build; it is four to six weeks fixing the two or three low-scoring questions, which costs a fraction of a stalled project. Question 8 is the one to fix first, because every other investment is unmeasurable without it.
0 to 9: not yet, and knowing that just saved you the money. The honest recommendation at this score is ordinary process work: write down how the work runs, get the data into one system, name owners. None of that is AI, all of it raises the score, and most of it pays for itself even if you never build anything intelligent on top.
Why data is the pillar that decides most cases
Across every assessment we have run, and across our own products, the pattern repeats: ideas fail on question 1 and 2 far more than on anything technical. The model was never the problem. The data the idea needed lived in three systems that disagreed, or in a free-text field every person filled differently, or in WhatsApp.
This is fixable, and it is boring, and the fix is worth more than the AI. A business that consolidates its order history to answer question 2 gets better reporting, better forecasting and cleaner operations before any model touches anything. We have watched clients get most of the projected benefit from the cleanup alone.
A worked example: a distributor scoring 13
Numbers mean more with a face on them, so here is a composite of the mid-band business we meet most often: a regional distributor, forty staff, orders arriving by phone and WhatsApp, an accounting system, and a director who wants "AI in the sales process".
Their scores, honestly assessed: data 3 of 6, orders end up in the accounting system, but item-level details live in WhatsApp threads and two clerks' memories. Process 2 of 4, everyone knows how ordering works, nobody has written it down, and the two clerks handle exceptions differently. People 3 of 4, there is an obvious checker and a plausible owner. Measurement 1 of 4, nobody can say how long order entry takes today, so no improvement could be proven. Economics 4 of 6, volume is real, wrong orders are costly but survivable, the fallback is the existing clerks. Total: 13.
What the score buys them is a sequenced plan instead of a stalled project. First, six weeks of unglamorous work: a structured order-capture step so item data lands in a system rather than a chat thread, and a one-week measurement of order-entry time to set the baseline. That work needs no AI budget and pays for itself in fewer mis-keyed orders regardless of what happens next. Then, with questions 1, 2 and 9 fixed, the original idea, extracting orders from WhatsApp messages automatically, becomes a well-posed project with a measurable target: minutes per order against a known baseline, checked by a named clerk.
The alternative timeline, the one where they build first, is easy to describe because we have watched other businesses live it: the extraction works in the demo, fails on the real message history the demo never saw, and the project is remembered as "AI does not work for us". Nothing about the technology differs between the two timelines. Only the sequence does.
Running the assessment internally
The self-test above is genuinely enough to start, and here is how to make it rigorous without hiring anyone.
Have three people score it independently: someone senior, someone who does the actual work, and someone technical. The disagreements are the findings. When the founder scores question 4 a 2 and the operator scores it a 0, you have learned the process exists on paper and not in the building, which is precisely the kind of thing a paid assessor is hired to discover.
Then, for your single best idea, write one page: the data it needs and where that lives, the number it must move, the baseline today, who checks the output, and what a wrong answer costs. If the page is easy to write, build. If it is hard, the hard parts are your project plan.
When to pay for an assessment instead
Doing it yourself breaks down in three situations: when the ideas span departments that do not share data or incentives, when the scores are contested and a decision needs an outside referee, and when a budget holder needs a costed plan with numbers they can take to a board. That is what a formal two-week engagement produces, and it is described plainly on our AI readiness audit page, including who should not buy one. The short version applies here too: if you scored 18 or above and everyone agrees, do not pay us to confirm it.
Questions people ask
What are the five pillars of AI readiness?
Data, process, people, measurement, and economics with risk, as laid out above. Frameworks differ in labels, some split infrastructure or governance into their own pillars, but every credible one covers this territory. What matters is scoring them honestly, not which taxonomy you pick.
How do you measure AI readiness?
With scored, specific questions rather than sentiment, answered by more than one person. The self-test above is a working instrument: twelve questions, three scorers, and the disagreements treated as findings. Measurement of the AI itself comes later, and question 8 is its foundation.
What is the difference between AI readiness and digital maturity?
Digital maturity asks whether your operations run on software at all. AI readiness assumes they do and asks whether the data those systems produce can support automated decisions. A business can be digitally mature and score poorly here, usually on questions 2 and 8.
How long does an AI readiness assessment take?
The self-test takes fifteen minutes. A formal assessment takes about two weeks, most of it interviews and data review rather than technology. Anyone quoting months is assessing something else, or padding.
The one-line summary
Readiness is not about the model, the vendor or the year. It is whether the data exists, the process is real, someone can check the output, and success has a number. Score yourself honestly on those twelve questions, fix the lowest scores first, and the AI decision largely makes itself.