How to measure AI readiness before you spend anything

You measure AI readiness by scoring six dimensions of the business that AI depends on, not the AI itself. Data, process, people, technology, strategy and governance. Score each one from 0 to 20 against a written standard, add them up, and you have a number out of 120 that you can put in front of a board and defend line by line. A weak answer in any dimension is one you cannot evidence. A strong answer is one somebody could check tomorrow without asking you.

That is the whole method. The rest of this article is the standard for each dimension, so your 14 out of 20 means the same thing in June as it did in March, and the same thing to your CFO as it does to you. You can do this on a spreadsheet in an afternoon. You can also take the free 30-question audit, which takes 7 minutes and gives you the score out of 120 and a band per pillar without you building anything.

Why you score readiness rather than AI

PwC surveyed 4,454 CEOs across 95 countries and found 56% reporting neither revenue gains nor cost reductions from their AI initiatives. Only 12% got both (PwC Global CEO Survey, via Mindfuel). That is not a survey about bad models. Almost none of those CEOs bought a model that failed to work. They bought working software and put it into businesses that could not absorb it.

RAND’s breakdown of enterprise AI project outcomes says the same thing from a different angle: 80.3% fail to deliver the promised business value, split into 33.8% abandoned before production, 28.4% that reach production and underdeliver, and 18.1% that run but never recover the investment (AI Assembly Lines). Read the split carefully. A third die before anyone can blame the technology. The failures are upstream.

So the thing worth measuring before you spend is the upstream. Six dimensions, because that is where the failures actually live, and because six is small enough that a board can hold all of it in their heads in one meeting.

The scoring rule

Each dimension gets five questions. Each question scores 0 to 4. That gives 20 per dimension and 120 overall.

The only rule that matters: score the evidence, not the intention. If you cannot point at a document, a person, a report or a system that proves the answer, the answer is a 1. Plans score zero. “We’re working on it” scores zero. This is the discipline that makes the number useful, and it is the discipline most internal assessments quietly abandon around question nine.

Bands, so the number means something:

• 0 to 40. Do not spend on AI yet. You will lose the money and the credibility.
• 41 to 70. Fix the two lowest pillars first. One narrow use case is possible in parallel if you pick it carefully.
• 71 to 95. You can run real deployments. Expect scaling friction rather than outright failure.
• 96 to 120. Scale. Your constraint is ambition, not foundation.

Dimension one: data, out of 20

The test question I use first, because it separates confidence from reality inside thirty seconds: if you needed a clean, accurate list of your best 500 customers right now, how confident are you in what you would get?

Weak answer (0 to 1): Somebody would export three systems, reconcile them by hand over two days, and you would still argue about the definition of “best”.

Strong answer (3 to 4): One system is the agreed source of truth, the query takes minutes, and a named person owns the accuracy of the result.

Score the other four on the same basis: how complete is the data behind your decisions, can key systems exchange data without someone manually exporting spreadsheets, how often does reliable data actually drive an important decision, and who owns data quality by name. That last one is the one that catches people. “Everyone” is a zero. “IT” is a one. A named individual with it in their objectives is a four.

This dimension is where most mid-sized businesses bleed points, and it is also the cheapest to fix. I have written separately on why dirty data is the cheapest AI mistake to fix first.

Dimension two: process, out of 20

AI automates a process. If the process is undocumented, you are automating whatever the last person who touched it happened to do.

Weak answer: Ask three people to complete the same core task and you get three different approaches, all defensible, none written down.

Strong answer: Core processes are documented well enough that a new starter could follow them without asking anyone, and someone reviews them on a schedule rather than when something breaks.

The five: documentation quality, consistency across people, frequency of deliberate process review, awareness of where the bottlenecks actually sit, and named ownership of process improvement.

A note on the bottleneck question. Most leaders score themselves a 3 here and are wrong. If you have not timed a process end to end in the last year, you do not know where the bottleneck is, you know where it was. Score a 1.

Dimension three: people, out of 20

Weak answer: The team is quietly using AI tools you have not approved, nobody has been trained, and you personally have not opened one of these tools in three months.

Strong answer: There has been formal investment in helping the team understand what these tools do, capability is honestly assessed rather than assumed, and the leadership is visibly using the things it is asking others to use.

The five cover team attitude, practical capability if you asked them to use a tool tomorrow, whether you as the leader are modelling adoption yourself, formal investment in understanding, and how that capability is being sustained.

The leader-modelling question is uncomfortable and it should be. If the CEO has never used the tool, the organisation reads that accurately as a signal about how much it matters. Score yourself honestly. Nobody else is going to.

Dimension four: technology, out of 20

This is not “do we have modern software”. It is whether your estate can support something new without a six-month integration project nobody costed.

Weak answer: Nobody has a current inventory of what you run. Two departments bought overlapping tools. Integration means a person and a CSV.

Strong answer: You know what you run, what it costs, what connects to what, and adding a system is a decision rather than an archaeology project.

If you have no internal technical capability at all, that is a specific position rather than a hopeless one. I have set out what a business without a tech team can realistically do first.

Dimension five: strategy, out of 20

Weak answer: “We’re looking at AI.” There is a budget line and no defined outcome attached to it. The success measure is “adoption”.

Strong answer: There is a named business problem, a number that would move if it were solved, a person accountable for that number, and a date.

BCG found CEOs estimate 35% of their performance evaluation depends on achieving AI ROI, against boards’ own estimate of 27% (BCG). An eight-point gap between how exposed the CEO feels and how exposed the board thinks they are. That gap closes fastest when the strategy dimension has an actual number in it. If you are preparing for that conversation, the mechanics of proving AI ROI to a board are worth reading alongside this.

Dimension six: governance, out of 20

One question settles most of this. If an AI tool your business bought created a compliance failure tomorrow, name the person who would be held responsible. One name, one job title.

Most people cannot. Schellman’s 2026 State of AI Governance Report found 42% of organisations place AI purchasing and adoption authority with the CIO or head of IT, 16% with a Chief Data Officer or AI Officer, and only 10% with the CEO (Schellman). In most of those organisations the executive who decides to bring the tool in is the same one carrying the blame if it goes wrong. That is not a governance structure. It is one person and a hope.

Grant Thornton’s 2026 AI Impact Survey found 78% of business executives lack strong confidence they could pass an independent AI governance audit within 90 days (Grant Thornton). Score this dimension on written policy, approval routes for new tools, monitoring of what is already in use, a defined position on data leaving the business, and named accountability. Not “we have a policy”. A policy somebody outside the company could read and understand who decides what.

What the number is for

A score out of 120 does three things a narrative cannot.

It makes the conversation comparable over time. A board that hears “we’re making good progress on AI readiness” twice in a year has learned nothing. A board that hears 61 then 78 knows what happened and can ask which pillars moved.

It makes the gaps addressable. A total of 64 is not actionable. A total of 64 made up of data 9, process 11, people 14, technology 13, strategy 8 and governance 9 tells you exactly where next quarter’s effort goes, and it is not where most people assume.

And it stops a working pilot being mistaken for readiness. A March 2026 survey of 650 enterprise technology leaders found 78% of enterprises running AI agent pilots, with under 15% reaching production (Digital Applied). Pilots run on hand-picked data, hand-picked people and no governance, which is precisely the three dimensions a score measures and a pilot conceals.

Do it yourself or do it in 7 minutes

Build the spreadsheet if you want to. Thirty questions, six tabs, a scoring rubric you agree with your exec team so the numbers survive contact with disagreement. It is a decent afternoon’s work and the arguing is half the value.

If you would rather have the number today, the free AI readiness audit asks the same 30 questions across those six pillars, takes 7 minutes, and returns a score out of 120 with a band for each pillar and a short synopsis. Every answer is self-reported, which means the honesty of the output depends entirely on the honesty of the input. Score the evidence, not the intention. That rule applies whichever way you run it.