How to Analyse an RFP with AI — Without Expensive Bid Software

Set the source and the boundaries

AI can help organise, extract and compare candidate information, but it cannot guarantee completeness, settle an ambiguous clause or decide whether to bid. Start with the issued invitation, instructions, schedules, specifications, pricing forms and any official clarifications or amendments. Record document names, versions and issue dates so the team knows which set is current. Check client restrictions and your organisation’s rules before putting any material into an AI tool.

If an approved environment is unavailable, carry out the first pass manually. A short, controlled extract can sometimes be enough for a permitted AI task; do not treat an incomplete extract as the whole tender.

Map the instructions before drafting

Read the submission instructions and build a plain-language map of the structure: scope, response format, portal steps, mandatory attachments, questions and deadlines. Include time zone, clarification cut-off and any separate pricing deadline. Identify requested deliverables and obvious commercial or contract flags for the responsible specialists to review. AI can suggest a map, but a person must check every date and instruction against the issued source.

In practice: For example, a schedule may set a different deadline for questions than the main invitation. Record both with their document and section references rather than accepting one AI-generated date.

Extract candidate requirements and evaluation criteria

Ask an approved tool for candidate requirements with document, section, page or quoted text. Distinguish mandatory conditions from scored criteria and general background. Do not infer a requirement where the document is silent. Capture uncertain wording as a question for a human reviewer.

Review the published evaluation criteria separately: what evidence is requested, which questions are scored, and whether any pass/fail conditions apply? An AI summary can omit a condition in an annex; someone must trace each finding back to the original.

Compare amendments and resolve uncertainty

When an addendum arrives, compare it to the previous issued version and list possible additions, deletions and changed wording with both source references. Verify the comparison line by line against the complete pack, schedules and addenda; then update the team’s requirements, dates and response plan. Record unclear points and submit formal clarifications through the buyer’s stated process when appropriate.

The result is a structured first-pass review list, not a final compliance decision. Finish with a human review of eligibility, feasibility, technical interpretation and commercial exposure. AI does not make the final interpretation or the bid/no-bid decision.

In practice: Sample prompt for an approved extract: ‘List candidate requirements, deadlines and evaluation criteria. Separate quoted source facts from assumptions or unknowns; give the document and section for each point. Do not decide whether we comply.’