What is an AI readiness assessment, and what does it cost?

An AI readiness assessment is a structured diagnosis of whether your organisation can absorb AI, not a review of the AI tools you have already bought. It examines the conditions AI depends on: whether your data is clean and joined up, whether your processes are documented, whether your people will actually use the thing, whether your technology can carry it, whether anyone has decided what AI is for, and whether anyone owns the outcome. It is a question about your business, not about the model.

The cost runs from nothing to five figures. At one end, a free self-assessment: our own AI readiness audit is 30 questions, six pillars, five questions each, and takes 7 minutes. At the other, RSM sells its AI readiness assessment as a four-week engagement delivering a comprehensive roadmap and implementation plan. Accenture and Deloitte run comparable exercises at enterprise scale. Almost none of them publish a price. That silence is the reason this page exists.

What the assessment is actually diagnosing

Most people arrive at the term expecting a technology review. They want someone to look at their stack and say yes, you can run this. That is not what a readiness assessment does, and if that is what you are sold, you have been sold a procurement exercise with a nicer name.

The diagnosis is organisational. Deloitte frames its version around six pillars: strategy, data, technology, talent, governance and research capability. Accenture examines strategy, workforce, workflows, risk and governance, data, engineering, operations and ecosystem. Our own six are Data, Process, People, Technology, Strategy, Governance. Different labels, same underlying question. Can this organisation take a working AI capability and actually keep it running once the pilot budget stops?

The evidence says most cannot. RAND found that 80.3% of enterprise AI projects fail to deliver promised business value, and the same research notes 78% of enterprises have AI agent pilots while only 14% reach production scale. PwC’s 29th Global CEO Survey, covering 4,454 leaders across 95 countries, found 56% of CEOs reported neither higher revenues nor lower costs from AI over the past 12 months. Only 12% got both.

Those failures are not model failures. A readiness assessment exists because the things that kill AI programmes are boring and internal: nobody can produce a clean customer list, three people do the same task three different ways, and nobody’s name is on the outcome.

The ownership question, and why it dominates the diagnosis

If an assessment only has time for one question, this is it. VentureBeat survey data found 85% of enterprises run contested AI platforms with no single owner, and only 10% can automatically detect a failing model. Asked to name the biggest obstacle to governing AI across platforms, respondents put “no single owner or accountable team” second at 29%.

Try it now. Say out loud the name of the person accountable for AI outcomes in your business. If that takes more than three seconds, or if the answer is a committee, you have your result and you did not need a consultant to get it. We have written separately on who owns AI outcomes in a company, because it is the single most reliable predictor of whether anything reaches production.

The board layer is no better informed. Diligent’s 2026 survey found only 8% of directors report strong AI expertise on their board, the lowest of every area surveyed, while 44% want AI elevated to the top of the agenda. BCG’s survey of 625 CEOs and board members found 60% of CEOs think their boards are too impatient with the pace of AI transformation. Impatience from people without expertise is not pressure. It is noise, and a readiness assessment is partly a device for converting it into something a board can act on.

What the free tier can and cannot tell you

A free self-assessment is a diagnosis, not a plan. Ours gives you a readiness score out of 120, a band for each of the six pillars, and a short synopsis. That is the complete output. There is no action plan at the free tier and no sales call unless you ask for one.

What it can tell you: where you are weakest, in ranked order, in 7 minutes. That is genuinely useful, because most leadership teams disagree about where the weakness sits. Finance thinks it is data. Operations thinks it is process. The CEO thinks it is that nobody is trying hard enough. A scored band per pillar settles the argument fast enough to be worth the coffee break it costs.

What it cannot tell you: anything it did not ask. Every answer is self-reported. It does not connect to your systems, read your data or inspect your code. It does not review the tools you have already paid for or check whether anyone still logs into them. If you overestimate your data quality, the score will overestimate your readiness. That is a real limitation and worth stating plainly rather than burying.

The questions themselves are deliberately unflattering. The first Data question asks: if you needed a clean accurate list of your best 500 customers right now, how confident are you in what you would get? The second Process question asks how similar three different people’s approaches would be to the same task. The third People question asks how actively you, as the leader, are modelling AI adoption yourself. Nobody enjoys answering those. That is the point. Our longer piece on what an AI readiness audit actually measures goes pillar by pillar.

What the four-week paid engagement buys you

RSM’s four weeks delivers a roadmap and implementation plan. That is real work and it is worth real money. What you are paying for is not a better diagnosis. It is three things a self-assessment structurally cannot provide.

First, independent verification. A consultant can ask for the 500-customer list rather than asking whether you could produce it. The gap between those two answers is often the whole story.

Second, interviews across the organisation. A single person filling in a form gives you one person’s view. Twenty interviews give you the variance, and the variance is diagnostic: when the CIO and the finance director describe the same data estate in incompatible terms, that disagreement is the finding.

Third, a plan with sequencing, owners and cost. A score tells you data is your weakest pillar. A plan tells you which three data problems to fix first, who fixes them, and what it costs.

What four weeks does not buy you is certainty about the numbers above. RSM, Accenture and Deloitte all run these engagements, and the 80.3% failure rate persists in a market where they are widely bought. A good assessment raises your odds. It does not remove the part where you have to fix the process afterwards.

The middle of the market, and what sits between free and four weeks

The gap between a scored self-assessment and a five-figure engagement is where most mid-sized businesses actually live. Fifty to ten thousand employees, no dedicated AI function, a board asking questions, and no appetite to spend the price of a junior hire finding out what you half-suspect already.

Our own middle tiers sit there deliberately. A Full Report at £97 takes the same 30 answers and returns a much more robust report with action plans. A Deep Dive at £497 returns a 30/60/90-day plan covering how to fix the gaps before spending anything on AI itself. Both sit on the same diagnosis. The difference is that one tells you where you stand and the other tells you what to do about it in the next quarter.

Publishing those numbers is not a flex. It is the minimum this audience should expect. The people this is written for distrust price-on-application for good reason: it usually means the price is a function of what the seller thinks you can pay.

How to choose a tier without wasting the money

Start free. Always. Thirty questions costs you 7 minutes and it tells you whether the problem you think you have is the problem you actually have. If the score comes back weak on Data and Process, which it usually does, you have learned something specific enough to act on, and you have learned it before signing anything.

Escalate to paid analysis when the free result is either contested internally or unanimously bad. Contested means two directors read the same score differently, in which case you need an outsider. Unanimously bad means you need sequencing, because fixing six pillars at once is how programmes stall.

Escalate to a four-week engagement when the spend at stake is large enough that four weeks of consultancy fees round to nothing against it, or when a board or regulator needs an independent document with somebody else’s name at the bottom. Those are legitimate reasons. “We should probably get consultants in” is not one.

And if you are early enough that none of this applies yet, the honest answer is that you do not need AI consultants to fix dirty data. You need someone to own the customer list.

One test to run this week

Before you buy any assessment at any price, run this. Ask whoever owns your customer data for a clean, accurate list of your best 500 customers by end of day. Then time it, and count how many people it passes through.

If it arrives in an hour from one system, your data pillar is in better shape than most. If it takes three days, four people and a reconciliation argument about what “best” means, you have the finding. No assessment at any price is going to tell you something more useful than that, and it cost you one email.