Tender and RFP response desk

The problem

Bid teams answer the same few hundred questions across every tender, with wording that drifts between submissions. Building one response means chasing security questionnaires, project references, staff CVs, and pricing assumptions from people who are busy on delivery. The compliance matrix gets checked by hand near the deadline, which is exactly when requirements get missed. Past winning proposals sit in a shared drive nobody searches, because nobody knows what is in it.

The solution

A response desk built from three agents: one reads the tender and turns it into a requirement list, one drafts each response from the approved answer library and past submissions, and one checks each draft against what those sources actually support. The bid manager receives a full draft with the unsupported requirements already flagged.

How Chocolate Factory does it

- Knowledge bases hold the approved answer library, past submissions, certifications, and CVs, each with its own retrieval instructions.

- Three agents with their own models and tools: extraction, drafting, review.

- One workflow: extract the requirements, then a Branch separating what the library answers from what needs a subject-matter owner.

- The bid intake posts the tender to /execute and gets one JSON response back, token usage attached.

- Nothing submits itself. The bid manager approves each answer, with the source visible behind every claim and cost per bid readable by project.

What changes

A full first draft exists on day one of the bid, so review time goes to pricing, win themes, and the questions that need judgement. Unsupported claims surface early, because the review agent holds any answer whose source cannot be resolved. The answer library gets used and gets better, since every bid shows which sources were retrieved and which were missing. Compliance matrix completion stops being a deadline-night task, and a no-bid call can be made against a complete requirement list instead of a skim.