AI Readiness Assessment

An AI readiness assessment answers one question before you commit engineering budget: can this organisation actually support the system you have in mind? Not in principle, and not after an eighteen month data programme. Now, with the data, permissions, skills and systems you have today.

Why AI projects fail before anyone writes code

The post-mortems are remarkably consistent. A customer service assistant is scoped, funded and staffed, and four months later it stalls because the knowledge base it depends on was last reviewed in 2021 and contradicts itself in about a fifth of articles. A document extraction pilot clears accuracy testing and then waits eleven weeks for a security exception because nobody checked what the data classification policy said about sending scanned contracts to a third party.

None of these are modelling problems. Every one of them was visible in advance to anyone who went looking. That is the entire purpose of an AI readiness assessment: to find the constraints while they still cost a conversation rather than a quarter.

What an AI readiness assessment measures

We score five dimensions. Each is graded against the specific use cases you are considering rather than against a generic industry ladder, because readiness is not a single number. An organisation can be entirely ready for internal document classification and nowhere near ready for an autonomous agent with write access to a billing system. Scoring them together produces a figure that describes neither.

Dimension	What we examine	Most common blocker
Data	Coverage, quality, lineage, labelling, refresh frequency, and whether the fields your use case needs exist at the granularity it needs them.	Required field exists but is free text, inconsistently filled, or only accurate for recent records.
Systems	Integration surface, API availability, authentication model, rate limits, and how much of your stack can be reached programmatically at all.	A core system with no API, where the only integration path is a nightly export.
Governance	Decision rights, risk appetite, existing policy, EU AI Act and sector exposure, and who signs off on a model going live.	No named approver, so the system is built and then cannot be launched.
People	In-house skills, who owns the system after launch, and whether the affected teams have been told this is coming.	No post-launch owner identified, which means no one maintains it in month four.
Economics	Realistic build and run cost including inference, monitoring, retraining, and the internal review time nobody counts.	Human review of exceptions consumes most of the projected saving.

Data readiness

We examine the data itself, not the data dictionary. Those two disagree more often than not, and the disagreement is usually where the project dies. For each candidate use case we check whether the required fields exist, whether they are populated consistently across the period the model would learn from, how quickly they refresh, and whether there is a documented basis for using them for this purpose. Personal data, data acquired under restrictive licence terms and data held under customer contracts with narrow use clauses all get flagged here rather than during legal review six weeks before launch.

Systems and integration readiness

An AI system is only as deployable as the systems it has to reach. We map every touchpoint your use case requires and grade each on how reachable it actually is: modern API, legacy API, database access, file export, or screen only. We also look at authentication, rate limits and what a service account can and cannot do, because a permissions model that blocks automated access will surface late and expensively.

Governance and regulatory readiness

We establish who decides that a model is safe to deploy, what evidence they will require, and how long that takes. In parallel we classify each candidate use case for regulatory exposure. Under the EU AI Act, classification determines documentation, testing and human oversight obligations, and those obligations materially change both cost and timeline. Discovering a high risk classification after the build is one of the more expensive ways to learn this.

People and ownership readiness

Two questions matter more than headcount. Who owns this system in month six, and have the people whose work it changes been told? A system without a named owner degrades quietly. A system announced to its users at launch meets resistance that no amount of accuracy will overcome.

Economic readiness

We build a rough cost envelope covering build, inference at projected volume, platform and monitoring, retraining cycles, and human review of the cases the system cannot handle. That last line is the one optimistic business cases omit, and it is frequently the largest. A process automated to ninety percent still routes ten percent to a person, and if that reviewer did not previously exist the saving is smaller than the slide claimed.

How readiness is scored

Each dimension receives one of four grades against each use case. We publish the scoring rubric with the report so the grades can be challenged and, more usefully, re-run by your own team in six months.

Grade	Meaning	What it implies
Ready	No material gap. Work can start.	Proceed to pilot.
Ready with conditions	Workable, with a known constraint to manage.	Proceed, with the constraint written into scope.
Not ready	A real gap that remediation can close.	Fix first. The plan says what and in what order.
Blocked	A legal, contractual or architectural obstacle that will not clear soon.	Do not build this. Pick a different use case.

A use case is only as ready as its weakest dimension. Strong data does not compensate for a blocked integration, and we do not average the scores to produce a more comfortable number.

How this differs from an AI maturity model

Maturity models benchmark you against an industry percentile and place you on a five stage curve. That is useful for a board narrative and almost useless for a build decision, because knowing you are at stage two tells you nothing about whether the specific project in front of you will work.

An AI readiness assessment is narrower on purpose. It answers a build question about named use cases with a named verdict and a costed remediation path. If you need both, run this first and let the maturity narrative follow from the evidence rather than the other way round.

What it costs and what we need from you

Fixed fee, quoted before the engagement starts and unaffected by what we find. Ten working days elapsed. Your total time commitment is usually six to ten hours: an initial scoping call, four to six interviews of forty five minutes each, and the closing readout.

• Read access to representative data samples for the candidate use cases.
• Named contacts who own the relevant data and the relevant process.
• Whatever documentation exists, including the outdated documentation.
• Any existing AI policy, data classification policy and security requirements.
• One hour with whoever signs off on production deployment.

We work under NDA and can operate entirely inside your environment where data cannot leave it. No production write access is required at any point.