An AI readiness assessment from someone with nothing to sell you

An AI readiness assessment answers one question: if you spend money on AI, will it land on something that can hold it?

That is a different question from whether AI is useful, or whether your competitors are doing it. It is a question about your own building — whether the data is clean enough to feed a model, whether the systems holding it can be integrated without a rebuild, whether anyone on your side can own the result once the implementer leaves, and whether the job is worth doing at the price it costs to run each month.

Most of that is knowable before the money moves. It just isn’t usually looked for: MIT’s NANDA initiative reviewed 300 public deployments for The GenAI Divide: State of AI in Business 2025 and found 95 percent of enterprise generative-AI pilots produced no measurable P&L impact — which it attributed to approach rather than model quality.

Elsewhere on this site the same work is called a Technical Assessment — the second of three engagements, 3–4 WEEKS. Same read, same deliverables; people type “AI readiness assessment” when the worry is AI specifically.

Not a questionnaire

The free tools are real and some are good. Microsoft’s AI Readiness Assessment and Cisco’s AI Readiness Index each take ten minutes and will orient you. But the input is your self-report and the output is a score. A questionnaire asks whether your data is clean; it cannot open the database and find out.

So, the honest version: if a score is all you need, take a free one and keep your money. Come back when the answer has to survive a contract.

Not a maturity-model workshop

Maturity models — Gartner’s AI Maturity Model is the one most of them descend from — sort you into a stage. That is a description, not a plan. It cannot tell you which of your six problems is the one that will actually stop the project.

Not a vendor’s pre-sales assessment

An assessment offered by a company that also sells the implementation has a structural tendency to conclude that you should implement. That isn’t dishonesty; it is what a funnel does. The fix is a reader with nothing downstream: I work for the company buying the technology, not the vendor selling it. No commission, no referral fee, no revenue share — which is why I can write “don’t buy this yet” and lose nothing by it.

The order matters: each one can end the project on its own, and they get more expensive to fix in roughly this sequence.

1. 1

   Your data. Where it lives, who owns it, and how much of “we have years of data” turns out to be spreadsheets holding three spellings of every customer name. Most readiness problems are data problems wearing a different hat.

2. 2

   Your stack. Whether the systems holding that data can be integrated at all — APIs, export paths, licence terms, the line-of-business system nobody has upgraded in years. This is where timelines go wrong, and it is knowable in week one.

3. 3

   Your vendors. What is already under contract, what those contracts let you walk away with, and what the proposals on your desk are actually quoting for. I read the contracts, not the sales deck; where a quote is defensible I say so.

4. 4

   Your team. Who owns the thing after the pilot ends and the implementer goes home — who retrains it, who notices when it starts being wrong. An AI capability with no internal owner is not a capability. It is a subscription.

5. 5

   The unit economics of the use case. The one that gets skipped. What the job costs per run at the volume you actually have, and what it replaces. A use case that works in a demo and loses money at your volume isn’t a readiness problem — it is the wrong use case.

Your stack, your vendors and your roadmap, read end to end. You get a written assessment, a ranked list of what to fix first, and the cost of leaving each one alone.

The ranking is the part that matters. Any competent reader can produce twenty findings; the work is saying which three are worth money this quarter, and putting a number on the seventeen you choose to live with. The document lands in your inbox first, so you can read it before we talk. Then a working session: what I found, what I’d do, and what I’d refuse to do.

It runs 3–4 WEEKS. The fee is fixed before anything starts and invoiced 50/50. If you began with a single-decision review, everything you paid for it is credited in full against this within 90 days. Capacity is ONE NEW ENGAGEMENT PER MONTH and the next opening is SEPTEMBER; the full shape of the work is on how engagements work.

Not for teams that already have a CTO they trust. If you have a senior technical person with no stake in the outcome, you already have this — ask them.

Skip it too if only one decision is blocking you. A single proposal or quote is a two-week Decision Review rather than a whole-stack read, and it costs less; the ladder explains which rung fits.

If the read is for a transaction — you are acquiring a company, or an investor wants the codebase judged before the money moves — that is technical due diligence, a different scope with a different audience. If the real question is whether AI has any place in your business at all, start with an AI consultant for small business rather than a full assessment of a stack you already understand.

And if the decision is already made and you want a document that blesses it, I’m the wrong person. Assessments that reach a predetermined conclusion are already free.