Use case

AI Workflows for RFP Responses

Name: Fleet
Author: Fleet

RFPs are won and lost on two failure modes: missed deadlines (the response takes a week of an SE's time it didn't have) and wrong answers (a stale security claim copy-pasted from last year's response becomes a contractual problem). The knowledge to answer correctly exists — scattered across product docs, security pages, and past responses — but assembling it under deadline reliably produces both failures.

A chatbot makes the second failure worse: confident answers, no provenance, nobody structurally assigned to catch the one that's wrong.

How it works with an agent fleet

A Fleet workflow drafts responses from a curated facts corpus — your product docs, security documentation, and approved past answers — then a review step flags every claim it can't trace to a source, and a sales engineer approves before anything leaves.

genflows:
  - name: rfp-response
    steps:
      - {name: draft, prompt: "Answer the RFP questions in the trigger ticket. Use ONLY the corpus; mark any question the corpus can't answer as NEEDS-INPUT.", corpus: ["docs/product/**/*.md", "docs/security/*.md", "rfp/approved-answers/*.md"], kind: report, out: response.md}
      - {name: claims, prompt: "Trace every claim to a corpus source. Flag unsupported or stale claims, especially security and compliance answers.", depends_on: [draft], kind: review, out: flags.md}
      - {name: se-approval, depends_on: [draft, claims], kind: approval, out: decision.md}

The NEEDS-INPUT convention matters: questions the corpus can't answer surface explicitly instead of being improvised. Approved responses feed back into rfp/approved-answers/, so the corpus — and the next response — gets better with each RFP.

The fleet pattern

Ticket-triggered draft from a facts corpus → claim-tracing review → SE approval. The compounding loop is the real payoff: every approved answer becomes reusable, reviewed source material.

Guardrails that matter here

The claims review flags anything not traceable to the corpus — the unsupported-security-claim failure mode is the exact thing the review step exists for
An SE approves the full response; the recorded decision covers exactly what was sent
Runs execute on your own worker — RFP contents and your internal docs stay on your infrastructure

Who this is for

Sales engineering teams answering recurring RFPs and security questionnaires, where answer accuracy is contractual and the same fifty questions recur in different clothes.

Frequently asked questions

How do we keep the answer corpus current?

It's files in a repo — product docs you already maintain, plus an approved-answers directory that grows with each completed RFP. Stale answers get flagged by the claims review when they contradict current docs.

Can it fill in the customer's spreadsheet format?

The output is a document artifact. Teams paste or transform into the customer's portal/format — the win is that the content arrived reviewed, traced, and approved, not that the formatting is automated.

Related use cases

AI Workflows for Sales Battlecards →AI Workflows for Customer Onboarding Documents →AI Workflows for Compliance Evidence →

Run your first agent fleet

One binary. Five minutes. See every agent, coordinate every handoff, and keep a full audit trail of what your fleet did.

See how it works Install Fleet