Is Your Business Data Ready for AI?

Start with the output you need

Data readiness is specific to the job. An internal policy assistant needs reliable, current documents. A sales forecast needs consistent historical records and a clearly defined outcome. Collecting more data without a use case can increase complexity without improving the result.

Write down the decision or output the system must support. Trace each required fact to its source. If a person cannot establish which record is authoritative, an AI system should not be expected to resolve the disagreement automatically.

Check quality in practical terms

The Government Data Quality Framework describes completeness, uniqueness, consistency, timeliness, validity and accuracy. These dimensions are useful prompts for a business review; this UK framework is a reference, not a Singapore compliance standard.

For each source, ask whether required fields are present, duplicates are identifiable, definitions agree, updates arrive in time, formats are valid and records reflect reality. Set priorities according to the consequences of an error. A missing delivery address may matter more than an optional customer preference.

Assign ownership and control access

Give each source an owner who can resolve errors and approve changes. Document where records originate, how they are updated and which version the pilot will use. Decide what to retain and what to exclude from the working dataset.

Check that access follows the user’s role and that the selected tool is approved for the information involved. Do not copy sensitive material into a test environment simply because it is convenient. Obtain the appropriate privacy or security advice for your use case.

Illustrative review: a policy knowledge base

Before testing an internal assistant, a company could remove superseded policies, identify document owners and record effective dates. The test setData reserved for assessing a model after development decisi should include a current question, an outdated policy reference, conflicting instructions and a question the documents cannot answer.

Review whether the output points to the correct source and avoids inventing missing details. Keep an explicit route for employees to ask the policy owner. This example shows a preparation approach rather than a verified implementation outcome.

Make a readiness decision

Classify each issue as blocking the pilot, acceptable within a documented limit or suitable for later improvement. A small, well-understood dataset can support a contained experiment; unresolved ownership or unauthorised access should trigger further preparation.

Record the dataset version and test cases so the next review is comparable. Then return to the SME AI adoption guide or use the wider AI readiness checklist to assess people, process and governance alongside data.

Sources and further reading

• Government Data Quality Framework