Use AI to gather requirements before building anything

Collect a description of the current process, a few anonymised examples, existing forms and known failure cases. Include the person responsible for the workflow and the users affected by it. Avoid uploading sensitive records to an unapproved tool simply because they are convenient examples.

Ask the assistant to produce an interview guide based on that pack. Useful questions concern triggers, handoffs, delays, exceptions and what people do when a system fails. Questions about preferred colours are premature if the workflow itself is unclear.

Create three lists: confirmed facts, proposed requirements and unanswered questions. Attach a source or owner to each important fact. If an interviewee says approvals “normally take a day,” distinguish that observation from a contractual deadline or a required service level.

Ask AI to identify contradictions, then take them back to the responsible people. Do not let the model silently choose between conflicting policies. The resulting decision should be recorded with its owner and rationale.

For an approval application, test a normal request, a missing document, an absent approver, a rejected request and a change after approval. Describe what each user sees and which records change.

These scenarios expose hidden requirements more effectively than a long feature list. Ask the assistant to translate them into acceptance criteria, then have the business owner check that the criteria reflect the intended process.

Identify the smallest complete workflow that delivers value. Separate essential controls from conveniences that can follow later. Keep the reason for each priority visible: user need, operational risk, contractual requirement or learning goal.

An AI-generated roadmap is a draft. Check dependencies and operating effort with the people who will build and maintain the system. A feature that looks small on a list may require a difficult integration or data-cleaning project.

• Problem, audience and expected outcome.
• Current workflow and proposed changes.
• Data sources, permissions and retention needs.
• User scenarios and acceptance criteria.
• Scope boundaries, dependencies and unresolved questions.
• Owner for decisions, operation and post-launch feedback.

Use AI to keep this brief consistent as decisions change. The value is not a longer document; it is fewer unspoken assumptions and a clearer basis for checking the delivered product.