“AI readiness assessment” can sound like consultancy packaging for a document nobody reads. Done properly it is something narrower and more useful: a structured check of whether your business can support the thing you are about to buy.
The point is not to score the organisation. It is to surface the handful of constraints that will determine whether an initiative succeeds, while changing course is still cheap.
The six areas worth examining
Most assessments over-invest in the first three and skip the last three. The last three are usually where initiatives fail.
- Processes: how work is actually done, including the undocumented steps
- Information: what data and content exists, where it lives, and how accurate it is
- Systems: which tools are in use and how well they connect
- Capability: who could own a new workflow, and how much capacity they have
- Governance: what rules exist around data, privacy and approval
- Appetite: how the team feels about the change, honestly assessed
Processes: the documented version is rarely the real one
Every organisation has a gap between the process on paper and the process in practice. People build workarounds for genuine reasons, and those workarounds carry the knowledge that makes the process work.
Mapping the real version takes a few conversations with the people doing the work. It routinely reveals that the bottleneck leadership wanted to solve is downstream of a different problem entirely.
Information: accuracy matters more than volume
There is a persistent assumption that more data produces better results. In practice, a small set of current, well-owned information outperforms a large archive of contradictory versions.
The assessment should establish which sources are authoritative, who maintains them, and how stale they are. Where the answer is “nobody” and “very,” that becomes work item one rather than a reason to abandon the idea.
Capability and capacity: naming an owner
Automated workflows and internal assistants need an owner: someone who fields questions, approves changes and notices when output quality drifts. This is not a full-time role, but it cannot be nobody.
If no name can be attached to a proposed solution, that is a finding. It usually means either the scope is too large for the current team or the initiative is not actually a priority.
Governance: the rules that already exist
Before writing new AI policy, it is worth establishing what already governs your data. Client confidentiality terms, sector regulation and existing retention rules often answer the difficult questions without a new framework.
Where genuine gaps exist, they are usually narrow and specific: which systems approved tools may access, what must never be pasted into an external service, and who signs off on customer-facing output.
Appetite: the finding people avoid recording
Teams that have been through a poorly handled system change carry that experience into the next one. Scepticism is not obstruction, and treating it as such guarantees a difficult rollout.
Asking directly what people expect to go wrong is one of the more valuable questions in the process. The answers tend to be specific, accurate, and cheap to address early.
What the output should look like
A readiness assessment should produce something short and decision-ready: a picture of the current state, a prioritised list of opportunities with the constraints attached to each, and a recommended first step with a named owner and a measure of success.
If the honest recommendation is to fix an information or process problem before introducing any AI, the assessment has still paid for itself. That outcome is common, and it is far cheaper to reach before a purchase than after one.