Skip to content
ZYRA
繁中
Menu

PRACTICAL GUIDE

What should an AI feasibility audit include?

A useful feasibility audit should answer three questions: is there a worthwhile problem, can a controlled solution work in the actual environment, and what evidence would justify the investment?

1 / Describe the work, not just the desired tool

Write down what triggers the process, who handles it, which systems are used, and what counts as completion. Include routine cases and exceptions. A request such as “add a chatbot” is not yet a measurable business problem.

For example, enquiry preparation may involve locating product information, checking terms, drafting a reply, and obtaining approval. Measure those steps separately before deciding which should change.

2 / Establish a representative baseline

  • Case volume and complexity, including the proportion of unusual cases.
  • Actual human handling time, waiting, review, corrections, and follow-up.
  • Costs that are genuinely incurred, rather than an assumed saving from every minute released.
  • Quality criteria and the errors the business cannot accept.

3 / Check whether implementation is possible

Confirm data ownership, permitted use, user roles, available interfaces, and responsibility for approval. Test assumptions with authorised samples. If a system has no suitable connection or its permissions are unclear, record that limitation rather than treating it as solved.

4 / Set the pilot decision before running it

Specify scope, comparison cases, human fallback, acceptance criteria, total cost, and who decides the next step. Include failure tests: missing information, rejected access, timeouts, changed content, and duplicate submissions.

The assessment should end with a recommendation: proceed with a bounded pilot, collect more evidence, simplify the process, or defer. None of those outcomes requires a guaranteed saving or a preselected tool.