AI Readiness

Start with a business task

Begin with the work that needs to improve. Describe the current process, the people affected and the outcome worth achieving. Identify an accountable business owner and decide how the result will be observed. Access to a model or an attractive demonstration is not, on its own, evidence that the organisation can operate the use case.

Keep the first evaluation bounded. State what the system may do, what it must not do and which decisions remain with people. Define the baseline against which the proposed improvement will be assessed. If the problem cannot be described clearly, clarify it before choosing an AI platform.

Check the data foundations

Identify the information required for the task, where it comes from and who owns it. Assess whether the information is accessible, sufficiently accurate and appropriate for the proposed use. Check permissions and restrictions before placing information into an external service. Refer unresolved legal or contractual questions to the appropriate reviewers.

Use representative samples to test quality. Record omissions, inconsistency, outdated material and the work needed to make the data usable. Separate a known defect from an unassessed condition. Assign an owner to each significant gap and decide whether it prevents evaluation or can be managed within a carefully bounded test.

Validate infrastructure and economics

Translate the use case into operating requirements: expected demand, response needs, integration points, data movement, compute, storage and network capacity. Compare feasible public, private and hybrid arrangements against those requirements and the controls needed. The appropriate venue follows the use case, not a general preference for a particular deployment model.

Build a cost model that can be tested against observed usage. Include the service, data preparation, integration, evaluation, monitoring, support and human review required. Document the assumptions behind demand and unit cost. A low trial cost does not demonstrate that production use will be economical at a different scale or review burden.

Put accountability and review in place

Name the business owner, technical owner and the people who approve changes and handle exceptions. Define permitted use, sensitive-information handling, escalation and the required human review. Consider what happens when an output is wrong, incomplete or inappropriate for the business decision being made.

Create an evaluation approach using examples that reflect the actual task. Decide which failures are unacceptable and what evidence is required before release. Keep version and change records for the relevant model, prompts, data and integrations so an unexpected result can be investigated and a change can be reversed when needed.

Prepare the operating model

Specify who monitors availability, usage, cost and output quality, and who responds when a limit is exceeded. Identify the skills required in the business and technical teams. Provide training on the system's intended use, limitations and escalation route. The operating model should cover the workflow around the AI service as well as the service itself.

Agree pilot exit criteria before the evaluation starts. A decision to expand should consider the measured business result, data quality, control effectiveness, support readiness and known limitations. Retain a practical way to pause the service or return to the previous workflow while issues are investigated.

Use readiness evidence to decide the next step

Review the five foundations together: data, infrastructure, governance, capability and operations. A strong position in one area does not compensate automatically for an unassessed dependency in another. An indicative questionnaire can expose the questions worth asking; it cannot certify a production service or select the right product without validation.

Produce a decision record with the baseline, the evidence reviewed, gaps, owners, proposed scope and acceptance conditions. Choose whether to investigate further, run a bounded evaluation, remediate a dependency or defer the use case. The next commercial discussion should be about that specific outcome and evidence, rather than a generic AI purchase.

Discuss your next decision

The assessment provides an initial, self-reported view. A follow-up should validate the relevant answers, agree the outcome and define the scope before any commitment.