Why readiness decides outcomes
When an AI project stalls, the post-mortem rarely blames the model. It blames the data that could not be accessed, the approval that took four months, the team that had never shipped a model-backed feature, or the sponsor who lost interest. Every one of those was knowable in advance.
Readiness is a map of where the first project will get stuck rather than a gate to pass before you are allowed to start. Use it to choose a project that avoids the worst gaps and fix the rest in parallel. Organisations that score themselves honestly pick better first projects and ship them faster. That is the whole value.
The six dimensions
1. Strategy and sponsorship
Ready looks like: a named executive sponsor, a short list of business problems AI is expected to help with, and a budget line. Not ready looks like enthusiasm without an owner, or a mandate to "do something with AI".
Ask: Who loses sleep if the first project fails? Which three problems would we fix first, and what do they cost today? Our AI strategy work exists to answer the second question properly.
2. Data
Ready looks like: the documents or records the first project needs are identifiable, accessible through an API or export, reasonably clean, and owned by someone who will say yes. Not ready looks like data in scanned PDFs, spread across systems with no common key, or guarded by a team with no incentive to share.
Ask: Can an engineer get a sample of the data this week? Who decides what may be used? How stale is it? This dimension is so often the bottleneck that we wrote a separate guide: data readiness for AI.
3. People and skills
Ready looks like: engineers who have shipped at least one model-backed feature, product people who can write an evaluation case, and business users willing to test something imperfect. Not ready looks like a reliance on one enthusiast, or a plan to hire a team before any project exists.
Ask: Who has built something with a model API? Who would review outputs for quality? The gap closes fastest by pairing your people with experienced practitioners on a real project, which is how our AI onboarding and training is structured.
4. Technology and integration
Ready looks like: a sanctioned way to call a capable model, a cloud or server environment where a small service can run, and APIs into the systems the project touches. Not ready looks like every integration being a bespoke project, or security having no answer to "can we use a model API".
Ask: How long would it take to deploy a small service that calls a model and reads from our CRM? If the answer is months, the first project needs to include fixing that. Our AI data engineering team often starts here.
5. Governance and risk
Ready looks like: a one-page acceptable-use policy, someone who owns AI risk, a basic review step for customer-facing or people-affecting uses, and an understanding of which regulations apply. Not ready looks like no policy and quiet use of public tools with customer data.
Ask: What is our answer when a customer asks how we use AI with their data? Who would sign off a customer-facing assistant? See AI governance for what the minimum looks like and our governance service for setting it up.
6. Process and change
Ready looks like: the team whose work will change is involved from the start, there is a baseline measurement of how the work is done now, and someone owns adoption after launch. Not ready looks like a tool built for a team that was never asked.
Ask: Do we know how long the task takes today? Who will change their working day, and have they agreed?
How to score honestly
Score each dimension from 0 to 4 using the rubric below. Have four or five people score independently, then meet to reconcile. Where scores differ by two or more, someone knows something the others do not; that conversation is the real assessment.
| Score | Meaning | Evidence you should be able to show |
|---|---|---|
| 0 | Absent | Nothing exists; nobody owns it |
| 1 | Ad hoc | Individual efforts; no shared artefact or decision |
| 2 | Defined for one project | Enough to start a bounded pilot: a named owner, a document, a working path |
| 3 | Repeatable | Used by more than one project; written down; someone maintains it |
| 4 | Embedded | Part of how the organisation works; measured and improved |
Resist averaging. A profile of 4, 1, 3, 3, 1, 3 does not reduce to a useful "2.5 out of 4". It describes an organisation with keen leadership that will hit a data wall and a governance wall in month two. The lowest scores set the plan.
Readiness level and the right first move
| Lowest scores | What it means | Recommended first move |
|---|---|---|
| Strategy below 2 | No owner or problem statement | A short strategy sprint: pick three problems, cost them, name a sponsor. Do not start building. |
| Data below 2 | The first project will stall on access or quality | A data readiness sprint on one source, then a project scoped to that source only. |
| Governance below 2 | Shadow use and no answer for customers | A one-page policy and a sanctioned tool this month; governance set-up alongside the pilot. |
| People below 2, others fine | A skills gap rather than a readiness gap | Start the pilot with experienced partners embedded, and make skills transfer a deliverable. |
| Everything 2 or above | Ready for a bounded production project | Follow the AI project roadmap; pick the project with the clearest baseline. |
| Everything 3 or above | Ready to run AI as a programme | Portfolio planning with the build vs buy matrix; platform investment. |
What to fix before a pilot
Fix three things, in this order, and leave the rest.
- A sanctioned path to a model. An approved provider account or gateway with logging and a data rule. Without it, the pilot is either blocked or unofficial.
- Access to one clean data source. One source, exported or queryable, with an owner's blessing; a data platform can wait.
- A baseline. How long does the target task take now, how often is it wrong, what does it cost? Measure it before the pilot, or you cannot show the result.
Everything else, including skills, broader governance and the platform, is best built while a real project is running and demanding it.
Pitfalls
Scoring by aspiration. "We are working on a data platform" is a 1 rather than a 3. Score what exists today.
Treating readiness as a gate. Waiting until every bar is green means never starting. Pick the project that avoids the short bars.
Fixing technology first. It is the most comfortable dimension to spend money on and the least often the real blocker.
Leaving out the people whose work changes. Their absence from the scoring predicts their absence from adoption.
What matters most
- Readiness is a map of where the first project will get stuck rather than a gate.
- Score six dimensions from 0 to 4, independently, then reconcile the disagreements.
- The lowest scores set the plan; never average them.
- Data and governance are the usual short bars; technology rarely is.
- Before a pilot, fix three things: a sanctioned model path, one clean data source and a baseline.
- Build skills and platform while a real project demands them.
Related services
AI readiness assessment
Score the six dimensions interactively in ten minutes and get a recommended first move.
Run the assessmentAI strategy & consulting
An assessment run with your leadership team, a costed shortlist of problems and a sponsor-ready plan.
ExploreAI data engineering
Turn the data bar from 1 to 3: access, pipelines, quality and ownership for the sources that matter.
Explore