Most AI audits ask your leadership team how AI is going, then write the answer in a deck. This one goes and looks: an anonymous survey across everyone, interviews, your actual billing and delivery data, and sessions where I watch real people do real work with AI.
You end up knowing where AI actually is in your business, what it is costing you including the part nobody has added up, and the five things worth doing in the next 90 days, in order, with owners.
78% of executives believe they have a clear picture of AI usage inside their organisation. The figure employee surveys actually support is closer to 23%. Roughly half of workers use AI in ways their employer has not approved, and about a third have put confidential company data into a public AI tool.
That gap is expensive, because everything downstream rests on it. Budget, training, hiring plans, the answer you give the board, the tools you renew. All decided on a picture that is too optimistic in some places and too pessimistic in others, and nobody knows which is which.
Meanwhile MIT looked at 300 public deployments and found 95% of enterprise AI pilots delivered no measurable P&L impact. Not because the models were bad: because the tools did not learn, integrate, or match how people actually work. And the 2025 DORA report found AI adoption raising delivery throughput and instability together. AI is an amplifier. Point it at a broken system and you get broken faster.
Each one scores 0 to 100 and lands on the same five-level ladder, so the language carries straight into the work that follows.
Is AI changing how your leadership team decides, or only how it drafts? Covers where AI touches real decisions, whether your execs can judge AI-assisted work from their own teams, and how much of what the business knows still lives in one person's head.
Is AI making your team faster, or faster at producing rework? Licences bought against daily active use, change failure rate and time to restore before and after adoption, review and provenance standards for AI-written code, and whether the platform underneath is good enough for AI to amplify.
Which functions have quietly gone AI-native and which have not. A function-by-function heatmap, the shadow AI register, and the pockets of excellence nobody knows about: in most companies somebody has built something genuinely good in a personal account and nobody else can use it.
What is in the way, and what is being wasted. Every licence and seat mapped to actual usage, whether there is a knowledge store your AI can reach, whether the policy survives a customer security questionnaire, and a regulatory read sized to your business rather than to a compliance programme you do not need.
Does the AI story in your deck survive a technical diligence conversation? Where AI sits in the product, cost per transaction, model dependency, whether you would know if a model change broke something, and whether the narrative overclaims. Included in the Deep Dive, £1,500 as an add-on.
Dabbling, Assisted, Installed, Compounding, Native. It is deliberately hard to climb: "Installed" means somebody other than the person who built a workflow can run it, and most companies cannot honestly claim that. Expect to score lower than you would like. That is the normal result.
Five sources. The mix is what makes the findings defensible when somebody in the room disagrees with them.
Including your exec team, who answer first, on behalf of the typical employee. Both numbers go in the report side by side. It is the most uncomfortable page in the document and the most useful.
CEO, each exec, the engineering lead, and individual contributors from different functions, including one the survey shows as low adoption. Fixed question spines, so answers are comparable.
Seat utilisation against licences paid, AI spend across the business including the subscriptions on personal cards, your SSO app list, and delivery metrics before and after adoption.
Your AI policy if one exists, vendor terms for the tools actually in use, engineering standards, and the last two board decks, to see how AI is described upward against what is happening.
The part nobody else does. Three to five people, 30 minutes each, doing a real task they were going to do anyway. What people say they do and what they do are different things, and the difference is usually the whole engagement.
Every score in the report cites at least one source that is not the survey. Where the evidence is not there, the report says the criterion was not evidenced rather than guessing. That rule is the difference between a diagnostic and a deck.
Ten outputs, one session, and a promise to come back in 90 days.
Plus a section called "What I would not do": the subscriptions to cancel, the pilots to stop, the tools not to buy, and where AI is the wrong answer. It is the section that gets the report forwarded, and it is what makes the recommendations in the rest of it credible.
Priced against a working CTO's day rate, not a consultancy partner's chargeout. Fees credit against whatever you do next.
Anonymous survey, three interviews, a 12-page read and a 60-minute readout. Leadership, engineering and business planes. Enough to know whether the full thing is worth doing.
Credited in full against either tier below within 60 days.
All four core planes. The full deliverable set, the exec readout, and a 90-day install plan with named owners. This is the one most companies want.
50% credited against a Sprint, a Tech Audit or your first retainer month.
Adds delivery telemetry analysis, AI code guardrails, the product-AI plane, a governance pack, board meeting attendance, and a free re-score at 90 days.
50% credited, same terms. Above 80 people, priced by the day.
Every plane has a remedy that already exists and already has a price. The report says which, and it says plainly where you can do it yourselves.
Leadership gap points at fluency: the AI Fluency course exists to close exactly that.
Business gap points at installation: the AI Brain Starter or Sprint, which starts on day one rather than week two because the backlog is already written in its format.
Engineering gap points at the 30-Day Tech Audit or an ongoing fractional retainer, depending on whether the problem is the platform or the leadership of it.
Foundations gap points at the policy and the context store, which is usually the cheapest work in the whole plan and almost never the work people expect.
Product gap points at raise readiness, if the AI story in your deck has to hold up under diligence this year.
Not every gap needs me. The 90-day plan says which items you can do yourselves in an afternoon, because a plan that recommends buying something at every step is not a plan, it is a quote.
The AI Reality Gap check is ten of the same statements. You answer as you believe your team would, they answer anonymously, and you see the gap. It costs nothing, takes about eight minutes each, and it will tell you fairly quickly whether the full report is worth your money.
Measure your gapA structured diagnostic of how AI is actually being used across a company, rather than how its leadership believes it is being used. This one covers five planes: the executive team, the engineering team, the rest of the business, the foundations underneath (spend, context, policy, shadow AI), and optionally AI in the product itself. It ends in scores with the evidence behind them, a prioritised opportunity backlog, and a 90-day plan with named owners.
Most AI readiness assessments ask the leadership team how AI is going and write the answer in a deck. This one collects evidence: an anonymous survey across everyone, six to twelve interviews, tool billing and seat-utilisation data, engineering delivery telemetry, and observed work sessions where I watch real people do real tasks with AI. Every score in the report cites at least one source that is not the survey.
Three fixed fees. The AI Pulse is £1,750 for one week and up to 25 people. The AI Discovery Report is £7,500 for three weeks and up to 80 people. Adding the Engineering Deep Dive takes it to £15,000 over four to five weeks. Fees are credited against follow-on work: the Pulse in full, the larger tiers at 50%.
One week for the Pulse, three weeks for the Report, four to five weeks with the Engineering Deep Dive. Your team gives up about ten minutes each for the survey, plus 45 minutes for anyone interviewed and 30 minutes for anyone in an observed session.
Yes, and the rules are agreed in writing with the CEO before it goes out. No cut of the data is reported where fewer than 5 people responded, nothing anyone types is quoted verbatim without their permission, and no finding is ever attributed to an individual. That holds even when the person paying asks me directly who said what.
Then that is what the report says. This is worth settling before the invoice rather than after the draft: the report follows the evidence, and a diagnostic that can only return good news is not a diagnostic. In practice most companies score lower than they expect, and saying so plainly with evidence behind it is the part that is worth the fee.
Thirty minutes, no deck. Tell me the size of the company and what prompted the question, and I will tell you which tier fits, or that you already know enough to act and do not need one.
Engagements are remote-first. On-site works too where it genuinely helps, particularly for the observed sessions.