An AI Readiness Score Nobody Can Game

Almost every AI readiness assessment has the same defect: it asks you how ready you feel.

“How mature is your data strategy?” “How open is your culture to change?” “Rate your automation capability.” Every one of those can be answered anywhere on the scale by the same person on the same day, depending on mood and on what they hope the answer is. The output looks like a measurement and is a mirror.

We built a ten-question version and the design constraint was narrow: every question must ask about something that either exists or does not.

The structure

Ten questions, three minutes. Each answer scores 0, 1 or 2, for a total out of 20. Four areas: process, team, tools and risk.

Three bands:

Score	Band
0–6	Foundations first
7–13	Ready to pilot
14–20	Ready to scale

Nothing unusual there. The difference is entirely in how the questions are written.

Facts, not feelings

Here is the pattern, using the first three.

Can you name a repetitive task that eats several hours of your team’s week? 0 — Not yet, nothing stands out 1 — Roughly, but I could not describe it step by step 2 — Yes, I can describe the task and who does it

If two of your people did that task, would they do it the same way? 0 — No, it depends on who does it 1 — Mostly, with some personal habits 2 — Yes, it follows a written or well-understood process

Is the information that task relies on digital and reasonably accurate? 0 — Much of it is on paper, in heads or scattered across inboxes 1 — Mostly digital, but messy or duplicated 2 — Yes, it sits in systems we trust

Notice what the top answer requires in each case. A task you can name and describe. A process that is written down or genuinely shared. Systems you actually trust. Each of those is a thing that exists or does not, and you know which within about two seconds of reading the option.

The rest follow the same rule. Is there a named owner for decisions about new tools, even part-time? Can your core systems connect to other tools, or do you not know? Do customer and job records live in one main place? Have you tried automation on real work and measured what it saved?

That last one is the sharpest. “We’ve tried some AI” scores 1. “We tried it and measured the result” scores 2. The difference between those two answers is a number you either have or do not.

★ Insight ───────────────────────────────────── The design principle generalises to any diagnostic: make the top answer require an artefact. Not a judgement that things are good, but a thing that exists — a document, a name, a measurement, a single system. Artefacts are binary and self-evident to the respondent, which removes the entire space where optimism lives. It also makes the result actionable in a way a rating never is: a missing artefact tells you exactly what to go and make. ─────────────────────────────────────────────────

Why “nobody can game it” is a claim about incentives, not security

You can obviously tick every top box. There is no verification step and there could not be.

The reason that does not matter is that the score has no gatekeeping function. Nothing is granted or withheld on the basis of it. Inflating it buys you a nicer number on your own screen and a recommendation aimed at a business you do not run.

Contrast that with an assessment attached to a certification, a discount or a procurement decision. The moment a score controls access to something, the incentive to optimise it appears, and the questions stop measuring the thing and start measuring willingness to answer strategically. That is not a flaw in any particular assessment. It is what happens to every measure that becomes a target.

So the honest framing is not that the instrument is tamper-proof. It is that the questions are concrete enough that lying to yourself requires noticing you are doing it, and there is nothing to gain by continuing.

What the bands are actually for

The score is the hook. The bands are the product, and they say different things rather than more of the same thing.

Foundations first (0–6). The recommendation is not to start smaller with AI. It is to stop and fix the inputs — write down one process, consolidate one set of records, name one owner — and rescore in a month. Honest progress moves this band quickly, because the artefacts are cheap to create when you know which one is missing.

Ready to pilot (7–13). Enough is in place for one narrow, well-scoped pilot. The advice is to keep it small, measure it, and fix the weakest of the four areas while it runs.

Ready to scale (14–20). The constraint has moved off capability and onto sequencing and ownership.

The per-area breakdown matters more than the total, and this is the part most maturity models get backwards. A business scoring 12 with everything at a flat 3 out of 5 is in a very different position from one scoring 12 with strong process and nothing in tools. The total tells you which band. The areas tell you what to do on Monday.

What the four areas are doing

The choice of process, team, tools and risk is not arbitrary, and each is measuring a different kind of blocker.

Process asks whether the work is legible. Can you name the task, would two people do it the same way, is the information it relies on digital and trustworthy. This is the area that most often scores lowest, and it is the one people least expect, because it has nothing to do with technology.

Team asks whether anyone owns the decision, and whether the people doing the work are open to changing it. A business can be technically ready and organisationally stuck: good systems, clean data, and no named person with the authority to approve a change.

Tools asks whether your systems can connect and whether records live in one place. This is the closest thing to a conventional technology question, and it is deliberately only a quarter of the score.

Risk asks what happens when the automation is wrong — whether anyone would notice, and whether there is a point where a person checks. It is the area most likely to be scored optimistically, because it is the one people have thought about least.

The pattern we see most often is a business scoring well on tools and poorly on process. Modern systems, clean integrations, and no written description of the work anybody would automate. That combination produces the most expensive kind of pilot: technically successful, operationally useless, because it automated a process that three people were each doing differently.

Why a self-assessment is worth anything at all

The reasonable objection to all of this is that a business assessing itself will get the answer it wants, and a consultant’s scorecard is a lead-generation instrument wearing a diagnostic costume.

Both of those are true of most of them. What makes this one defensible is narrow and worth stating: it produces the same answer whoever runs it, because every question has a fact behind it. Give the ten questions to the owner, the operations manager and the person actually doing the work, and where they disagree, the disagreement is itself the finding. Three different answers to “is there a clear owner for decisions about new tools” tells you more than any score.

That is the version worth running internally. Not one person filling it in. Three, separately, then comparing.

Building your own

If you want to run this exercise internally rather than use ours, the rules that made it work:

1. Every option must be checkable inside your own business in seconds. If answering requires a survey, a workshop or an opinion, rewrite the question.
2. The top answer names an artefact. A written process, a named person, a single system, a recorded measurement.
3. Score the areas separately and report them separately. A single number hides the shape, and the shape is the advice.
4. Give the lowest band something to do this month. A diagnostic that tells a business it is not ready and stops there has told it nothing it did not already suspect.
5. Keep it under ten questions. Completion rate is a real constraint, and question eleven never changes the band.

The uncomfortable insight that keeps showing up when businesses run this: the questions that score lowest are almost never about AI. They are about whether a process is written down, whether records live in one place, and whether anyone owns the decision. Those were problems before AI was on the agenda. AI just makes them expensive.

---

The scorecard is on ashganda.com — ten questions, three minutes, no obligation. AI consulting is delivered through Ganda Tech Services.