Proposal review that checks every promise in the draft against your own records, before the customer reads it
Every promise in the draft checked against your records, before the customer reads it.
See one case, screen by screen ↓Proposal Review is proposal review software that checks a draft proposal for unsupported claims, missed requirements and inconsistencies before the customer reads it. Seven agents test every factual claim against your own records — SLA schedules, contract playbook, certificate and project registers, price book and past proposals — trace every RFP requirement to the paragraph that answers it, and draft each fix from the source. The proposal lead decides every fix, legal decides legal wording, and the VP signs gold team.
How to choose AI solutions for back-office and enterprise work →
Updated
Proposal teams now draft faster than they can check. The response library, generative tools and paragraphs copied from won bids fill a 64-page draft in days, and every reused paragraph brings its old numbers with it: 99.99% availability when the SLA schedule says 99.9%, 120 deployments when the project register counts 86, P1 incidents answered in 15 minutes when the schedule says 30, another customer's name left in from an earlier bid.
The check still happens by reading. Five red-team reviewers each read the whole draft in their own copy, the proposal manager merges their comments by hand, and the same finding arrives five times — but nobody tests each figure against the SLA schedule, the certificate register or the price book unless they happen to know it. Once the bid wins, what the proposal promises becomes a contract obligation, and a figure nobody checked is paid for in service credits.
Estimated split for a red-team review of a large proposal, by hand and with the solution.
Seven agents test every claim and requirement in the draft against your own records, re-check every red and draft the fix; people decide and sign.
Kestrel & Vane Industrial Services is bidding $14.6M over five years to maintain and remotely monitor warehouse automation at Halvergate Cold Chain's refrigerated distribution centres. Draft v0.8 is 64 pages, the red team meets at 14:00 and the bid is due Tue 20 Oct. Here is the day, screen by screen, in the working solution.
Noah's day starts with three numbers: 15 contradicted claims to decide, 7 of 146 requirements still open, and the red team at 14:00 with 5 reviewers. The evidence agent has already said where to start: §4.2, where the draft promises 99.99% availability and the SLA schedule says 99.9%. Noah asks whether the same promise is in any other bid.
Evidence agent: “Yes: Dunmere Foods v0.4, §5.1, the same sentence. It came from library answer LA-0412, last edited in 2024. I suggest Tobias retires that answer.”
Amara Quist uploaded draft v0.8 at 18:12. Overnight the intake agent read 64 pages, the RFP and compliance matrix CM-031 v3; the claim agent found 318 factual claims in 8 sections; the evidence agent tested each against 27 sources. The draft opens as the customer will read it: 272 claims supported, 31 unsupported and 15 contradicted, with the counts for every paragraph in the margin and every section in the outline.
Second check: 16 reds re-checked · 15 agree · 1 overturned.
Noah asks in the conversation beside the draft. The evidence agent answers from the Service Level Schedule v7, which commits to 99.9% a month. At 99.99% the company would allow 4.3 minutes of downtime a month instead of 43.8, and each breached month costs a credit of 10% of the monthly fee — about $24,333 on this contract. It opens claim C-041 with both source passages.
Draft says “99.99% platform availability, measured monthly” · Source says “The Monitoring Platform will be available 99.9% of each calendar month”.
The fix agent drafted the change from the schedule's own wording, and the checker agent, re-deriving it without the first working, agrees; confidence 0.99. Noah approves it with reason code FIX-SRC, “Matches the source”. The draft now offers “99.9% monthly platform availability, as defined in our Service Level Schedule §2.1, with service credits under §4” as a tracked change.
No agent changes the draft. Every fix is a person's decision, with a reason code.
The checker agent re-derived all 16 reds from the claim and the source without seeing the first working. It agreed with 15 and overturned one: C-024, ISO 50001 energy management at four US sites. The first check read a register extract with three sites; the certificate register added Dayton on 2 Oct 2026. The claim is supported.
The checker agent can hold a finding and show both workings. It can never delete one.
Against the compliance matrix, 139 requirements are answered, 4 in part and 3 not at all. A-07 is pass/fail: Section 3 runs to 17 pages against a 15-page limit, and pages beyond it are not evaluated. The agent proposes cutting the 1.4-page company history that repeats §1 and moving the Gantt chart to Appendix A, for 14.8 pages. B-33 needs a person: the WMS interface specification lives with Engineering.
“Two CVs must come from the people named. The agents cannot write them.”
The consistency agent compared every number, name, date and term across sections, appendices and past bids, and found 9 places where the draft disagrees with itself: 42 field technicians in §3.1 against a staffing table of 38 in §5.2, go-live in 12 weeks against week 16 in the timeline, and “Pellford Foods' Springfield office”, reused from a June 2025 proposal. Each pair shows the source that settles it, and one click applies the right value everywhere.
Consistency agent: “Evaluators read that as a copy.”
22 comments arrived from 5 reviewers. The agents merged the duplicates into 16 and linked 13 to findings they already had — Lena Moritz's “§4.2 says 99.99%. Is that really our SLA?” sits on C-041, already decided. Their pre-review, posted at 07:50, counts 15 contradicted claims with drafted fixes, 31 unsupported, 7 requirements needing work and 9 inconsistencies. Three comments are left for the room: win themes and the account director.
“So the room can spend its two hours on strategy, not on checking figures.”
The fix agent drafted 31 fixes — 15 for claims, 9 for consistency, 7 for requirements — and changed nothing in the draft. Noah steps through them with Approve, Edit or Reject: C-012 says 120 deployments since 2018, the project register counts 86, and the March bid to Halvergate said 85. Twelve claim fixes copy the source word for word and the checker agrees with each, so he reviews the list and approves them together. Unlimited liability (C-061) and data location (C-066) go to Ines Calloway.
Fix agent: “The evaluator has that bid.”
The agents list every promise the final wording makes: 27 commitments, each with its source and its owner. Ines approves the legal wording — liability capped at twice the annual fees under contract playbook CP-9, US hosting with encrypted backups in Toronto — and accepts the 6 legal commitments. Marcus Hale accepts the 9 service commitments: “We can deliver this”. Grace Oduya, VP Sales, types her name and signs the gate: ready to submit.
The signature is blocked while any contradicted claim, requirement or inconsistency is open. The 31 unsupported claims still open do not block it; they are listed on the signed page.
The same bid, seen by five of the people who carry it — what bid week looked like, and what it looks like now.
Seven agents test every claim and requirement in the draft against your own records, re-check every red and draft the fix; people decide and sign.
Reads the draft in Word or PDF, the RFP and the compliance matrix, splits them into sections and paragraphs, and finds the right version of each.
Finds every factual claim in the draft — figures, commitments, certifications, references, capabilities — and types it.
Tests each claim against the SLA schedules, contract playbook, certificate and project registers, price book, resourcing plan, hosting register and past proposals to the same buyer, and marks it supported, unsupported or contradicted with the passage.
Maps every requirement in the compliance matrix to the paragraph that answers it, checks page limits, forms and attachments, and drafts the missing text where a source allows.
Compares numbers, names, dates and terms across sections, appendices and past bids to the same buyer, and merges reviewers' comments.
Re-derives every contradicted claim from the claim and the source without seeing the first working, and holds the finding when it disagrees.
Drafts each fix as a tracked change built from the source wording, and routes liability, data and IP wording to the legal reviewer.
Approve, edit or reject with a reason; sign. The agents propose; a named person decides.
The review team sits beside the draft on every screen. Ask about a claim, a requirement or a deadline, or give an instruction — every answer cites its source, and the agents act on what you say.
Why is 99.99% red?
Our Service Level Schedule v7 commits to 99.9% a month. 99.99% allows 4.3 minutes of downtime a month; 99.9% allows 43.8. Each breached month costs a 10% credit: about $24,333 on this contract. Opened C-041.
What did the checker overturn?
One: C-024, “ISO 50001 at four US sites”. The first check read a register extract with three sites; the register was updated on 2 Oct and lists Dayton as the fourth. It is supported.
Send the legal ones to Ines
Sent C-061 (unlimited liability; CP-9 caps it at 2× annual fees) and C-066 (backups replicated to Toronto) to Ines Calloway with the drafted wording. Her answer is due Fri 16 Oct 09:30.
Is any other customer named?
Only Pellford Foods, once, in §3.2. I also checked for the 41 customers in our CRM and for the names in the references list: no other appears.
13 screens from the working solution, on its sample data. Pick one to see it large.
What counts as supported, what goes to legal and what blocks the signature are settings. A change is a proposal; it takes effect when the administrator approves it.
| Setting | Default | Choose from |
|---|---|---|
| Supported only at confidence | 0.85 | Below it the claim is amber, never green |
| Every contradicted claim re-checked | On | By the checker agent, without the first working |
| Legal decides | Liability, data location, IP, indemnities | The fix agent routes these to the legal reviewer |
| Marketing words | industry-leading, best-in-class, guarantee, world-class, unrivalled and 9 more | Marked unsupported, with softer wording drafted |
| Warn the proposal manager | More than 5 unsupported claims in a section | A warning, not a block |
| Block the gold-team signature | Any open contradicted claim, requirement or inconsistency | Unsupported claims are listed, not blocking |
| Open red before gold team | 24 hours before | Escalates to the VP Sales |
| Legal answer due | 1 business day | Escalates to the General Counsel's office after |
Every figure is labelled: a target the solution is built to, an estimate, a typical published result, or a proven one.
“demo” = seen in the working solution, on its sample proposals · “target” = the design goal, measured in the live solution · “estimated” = our estimate · “typical” = published figures (a 2026 RFP benchmark survey of 1,533 respondents, run by an RFP software vendor — 79% of teams use generative AI in the RFP process; the average team answers 166 RFPs a year at 33 hours each). Pink, red and gold team are the colour-team reviews commonly used by proposal teams. People, companies, bids and documents named on this page are fictional — characters and sample data in the working solution.
Proposal Review is proposal review software that checks a draft proposal for unsupported claims, missed requirements and inconsistencies before the customer reads it. Seven agents test every factual claim against your own records — SLA schedules, contract playbook, certificate and project registers, price book and past proposals — trace every RFP requirement to the paragraph that answers it, and draft each fix from the source. The proposal lead decides every fix, legal decides legal wording, and the VP signs gold team.
A compliance check asks whether the draft answers the RFP. Proposal Review does that too — every requirement traced, page limits counted — and also asks whether what the draft says is true for your company. Each figure, certification and commitment is tested against the records that bind you, and shown as supported, unsupported or contradicted, with the source passage.
No. It prepares them. The agents post a pre-review before the room meets, merge reviewers' duplicate comments and link each one to the claim or requirement it is about, so the red team spends its time on strategy. At gold team the VP signs against a list of every commitment, each with its source and owner.
No. The agents propose; every fix is a tracked change that the proposal lead approves, edits or rejects with a reason code. Liability, data location, IP and indemnity wording goes only to the legal reviewer.
A claim is marked supported only at confidence 0.85 or above; below that it stays amber. Every contradicted claim is re-checked by a second agent that cannot see the first working. A scanned page the intake agent cannot read with confidence is reported by page, with a request for the native file, never guessed.
Yes. Only the proposal lead decides a contradicted claim, only the legal reviewer approves legal wording, the delivery owner accepts the service commitments, and the VP signs the gold-team gate by name. Every step and decision is on an audit trail nobody can edit.
Yes. The confidence threshold, the claim types checked, the marketing words list, legal routing, warnings, review clocks and what blocks the signature are all settings. A change is a proposal that takes effect when your administrator approves it.
The Agentic Solution Engine builds and deploys it from your requirements and documents — your SLA schedules, registers, price book and a few past proposals — and it goes live once every quality gate has passed. Setup connects your sources, chooses what counts as a claim and calibrates on three past proposals before you approve the rules.
Questionnaires back in days, not weeks, and every answer matches what customers were told.
1-min demoRFP Response AutomationHigher win rate, no promise you can't keep, and the bid in on time.
1-min demoDue Diligence Questionnaire (DDQ) ResponseEvery DDQ answered from approved sources, and no investor told two different stories.
1-min demoWe'll run Proposal Review on one of your own past drafts, against your SLA schedules, registers and price book.