RFP & questionnaire autofill
Upload an RFP or security questionnaire, extract its questions, and draft each answer grounded strictly in your own win library — with confidence scores and verifiable citations.
RFPs and security questionnaires are where SE time goes to die. Pact's autofill drafts each answer for you, but with a hard constraint that keeps it trustworthy: it drafts only from your tenant's own win library — the proof points, past answers, and evidence you have already captured. It does not reach for arbitrary web content or general product knowledge, so a draft can never invent a capability you don't have.
How it works
- 1
Upload the document
Upload an RFP or security questionnaire. Pact stores it against a draft document with its own status, so you can track progress as answers fill in.
- 2
Extract the questions
Pact splits the document into individual questions using a deterministic heuristic — numbered lines, question marks, and known question keywords. This step is rule-based, not AI, so extraction is predictable and repeatable.
- 3
Draft each answer
For every question, Pact retrieves the most relevant evidence from your win library and drafts an answer grounded strictly in that evidence, using your own BYOK AI key. Each answer carries a self-reported confidence score and citations that point back to real records.
- 4
Review and edit
You review every draft. Confidence scores tell you where to look first; citations let you verify each claim against its source record before you trust it.
- 5
Export
Export the finished document to a Pact-branded PDF to send back to the prospect.
Grounded, cited, and verifiable
The draft model is only ever shown evidence retrieved from your win library — never free-form knowledge. It self-reports a confidence score per answer, and every citation it produces is validated against the real record IDs that were retrieved. A citation that doesn't map to a retrieved record is rejected, so the model cannot fabricate a source.
Retrieval blends semantic similarity (embeddings) with lexical overlap to find the best evidence for each question, and drafting runs on a background worker so a large questionnaire fills in progressively rather than blocking.
Drafts come from your win library — not the open web
Autofill is grounded exclusively in your tenant's win library. It does not draft from arbitrary web or product knowledge. If the library has no evidence for a question, the answer will say so rather than invent one.
It still works without an AI key
No BYOK AI key configured? Autofill falls back to a deterministic draft built from evidence excerpts — you still get a per-question starting point pulled from your win library, just without the model's phrasing. The whole flow works keyless.
Scoring and prep
Beyond drafting, Pact can score a completed RFP and generate a prep sheet to help you get ready for the follow-up conversation. Both build on the same evidence-grounded foundation.