Answering RFPs and questionnaires with AI

How the bot works

Nine steps. The extraction, the matrix, the citation checks and the staleness checks are ordinary code and cost almost nothing to run; the model is only used to read the question and draft from your own material.

1. It shreds the document into questions. Every question, requirement and instruction, numbered, with the page and section it came from — including the ones buried in an attachment, a pricing sheet or a terms exhibit that nobody opened until the last afternoon.

2. It builds the compliance matrix. One row per requirement: what is being asked, where it is asked, what format the answer has to take, what the word limit is, who owns it, and whether it is a knockout. The thing a bid desk builds by hand before anybody writes a sentence.

3. It runs your go or no-go screen first. Your knockout criteria against the matrix: the registrations, the insurances, the certifications, the delivery terms. If something disqualifies you, the best possible outcome is finding out on the first morning rather than the last.

4. It drafts only from material you have approved. Your answer library, your policies, your previously submitted responses, the documents you have put in scope. Nothing from the open web, nothing from another client, and nothing from the model's general knowledge of your industry.

5. Every answer cites what it came from. The source document, the section, and the date that source was last reviewed — attached to the answer, not in a separate report. A reviewer checks the citation instead of re-researching the answer.

6. Anything it cannot cite is a gap, not a guess. Left blank, named, and assigned to a person with the question and the context. An answer nobody can trace is worse than an empty box, because the empty box gets filled in and the untraceable answer gets submitted.

7. It flags the answers that have gone stale. An answer whose evidence is older than the limit you set comes through as needs-review even when it is word for word what you said last time. Last year's true answer is this year's are not always the same sentence.

8. It routes what a person must own. Security, pricing, legal terms and any commitment about a date, a person or a volume go to the named owner for that category. They are not drafted and left in the document hoping somebody notices; the row is theirs until they sign it.

9. A person reviews, and a person submits. The bot assembles the response in the format the buyer demanded, with the matrix, the citations and the gaps visible. Approved answers go back into the library so the next one starts from what you actually said. Submission is always somebody's click.

What you configure

It can be configured to run in any business. This is what that actually means — the list of things that are yours, not ours.

Your answer library

Which documents, policies and past responses are in scope, which are explicitly out, and which are the preferred source when two of them disagree.

Your citation rule

What counts as a source, how old an answer's evidence may be before it needs review, and whether an answer may ever cite another answer rather than a document.

Your owners

Who owns security, who owns pricing, who owns legal terms, who owns delivery and who owns the overall response. Every matrix row points at one of them.

Your sign-off rule

Which categories can never leave the building without a named person, and what that person has to see before they sign.

Your knockout criteria

What makes a bid not worth writing, checked on the first pass: registrations, insurance limits, certifications, bonding, delivery terms, geography.

Your handling rule for the buyer's documents

What may be uploaded, where it may be processed, where it may be stored, for how long, and which solicitations are excluded from the tooling entirely. We follow your rule; we do not write it.

Your format and voice

Answer length, structure, reading level, the house voice, the format the buyer insists on, and whether an answer may be reused verbatim or must be rewritten to the question.

Your library maintenance

Whether an approved answer goes back into the library automatically or on review, who reviews it, and how an answer gets retired.

Your systems

Where the library lives, where drafts are assembled, where the submitted response is filed, and whether we read and write by API or by export.

Transparent, and configurable by you

The dark band below is what the system will not do. This is what you can see and change — and it is how we build everything, not something invented for this page.

You see what it did, and why

Every answer carries its source document, its section and the date that source was last reviewed; every gap carries the question, the reason it is a gap and the person it belongs to. You can audit a response by reading the citations rather than the prose.

Versioned prompts you can tune yourself

The prompts behind reading a question and drafting from your library are yours to read and yours to tune. A change is a reviewable diff, replayed against responses you have already submitted so you can see exactly what it would have written, and rolled back the same way.

A human in the loop wherever judgment actually matters

Extraction, the matrix, citing and flagging staleness are automatic. Pricing, every security and legal answer, every commitment and the submission itself are a person's — that boundary is a property of the system, not a promise about how carefully we'll use it.

Every run logged, along with what it cost

One row per question and one per draft: the source cited, the staleness check, the owner, every edit a reviewer made, the model and prompt version that drafted it, and who approved it — plus token usage and cost for that run.

It can be configured to run in any business. The library and the rules are yours: your approved material, your citation rule, your owners, your sign-off, your handling rule for a buyer's documents. The solicitation wiring on this page is one configuration of it, not the limit of it.

Works with your tools

This runs on top of the document store and the proposal tooling you already have. We're not replacing your bid process.

• A proposal or RFP tool
• A GRC or compliance tool
• Google Drive, SharePoint or Dropbox
• Microsoft 365 or Outlook
• Google Workspace
• Salesforce
• A project or task tracker
• A folder of past responses
• Your own policy documents
• A spreadsheet you control

Works well for

• IT services and managed service providers — Every mid-sized prospect sends a security questionnaire, and the answers live in four people's heads.
• SaaS and software companies — Enterprise deals stall on a long questionnaire that has to agree with the policies you published.
• Government contractors — Public solicitations are long, strictly formatted, and unforgiving about a missed requirement.
• Marketing and creative agencies — Agency reviews arrive as a template to fill in, and the same questions come back every year.
• Commercial cleaning and facility services — Contracts are won on bids with fixed forms, insurance limits and a compliance checklist.

How a rollout works

No response is drafted for a live bid on day one. The first milestone is the bot answering a solicitation you have already submitted, so you can read its citations against what you actually said.

1. 01Discovery

   The quote call, then a working session. We go through where your approved material actually lives, which documents are in scope, who owns security, pricing and legal, your sign-off rule, your knockout criteria and your handling rule for a buyer's documents. You get the library design, the matrix template, the exception classes and a written scope.

2. 02Dry run

   The bot shreds and answers solicitations you have already submitted, from the library as it stood at the time. Nothing goes to a buyer. We read its citations next to your real answers, and every disagreement is either a gap in the library or an answer that was always somebody's judgment. This is where your own numbers come from.

3. 03Assisted mode

   Live on one kind of document first, usually the security questionnaire, because it repeats. The bot drafts and cites, your owners sign their categories, and a person submits. Answering stops being research and becomes review. The control doesn't change.

4. 04Optional wider rollout

   Optional. Once questionnaires have run clean for a while, full solicitations come on, with every hard stop in force, no submission rights ever, and the kill switch. Plenty of teams run questionnaires only, and that's a good place to stop.

What we don't claim

• No numbers we haven't earned. No share of questions answered, no turnaround multiple, no win-rate claim. We haven't answered an RFP for you. Every figure on pages like this one belongs to somebody else's marketing; yours come out of the dry run over responses you have already submitted.
• No customer claims on this page. No names, no case studies, no logos.
• It will not help you win a bid you should not be in. The go or no-go screen exists to tell you when to walk away. We are not claiming a model makes your proposal more persuasive, and nothing here improves your pricing, your references or your delivery record.
• It makes no legal or compliance judgment. Whether you comply with a clause, whether a certification satisfies a requirement, whether a term is acceptable: those are for your own people and your own counsel. The bot shows the requirement, cites what you have already said, and stops.
• Fully autonomous responses aren't on offer. Nothing is submitted by the bot, no uncited answer is ever written, and security, pricing and legal always need their owner. That's deliberate, and it's why the boundary is written down.

Questions people ask

Can AI answer our RFPs and questionnaires?

Yes, for the drafting - and the discipline is what makes it usable. An agent can shred a solicitation into every question and requirement with its section reference, build the compliance matrix, run your knockout criteria, and draft each answer from material you have approved, with the source document and its last-reviewed date cited on the answer. Anything it cannot trace to your own material is left blank as a named gap with an owner. A person reviews, the category owners sign theirs, and a person submits.

Where do the answers come from?

Only from material you have put in scope: your answer library, your policies, your previously submitted responses, the documents you named. Not the open web, not another client's answers, and not the model's general sense of how a company in your industry usually replies. If two approved sources disagree, your rule says which one wins, and the answer cites the one it used.

What happens to a question it cannot answer?

It stays empty and becomes a row with a name on it. That is the design rather than a limitation: an answer nobody can trace is more dangerous than a blank, because the blank gets noticed and filled in while the untraceable answer gets submitted. The gap carries the question, the section it came from and why nothing in your library covered it, which is also the shortest route to a better library.

Can it fill out security questionnaires?

That is usually the first thing it runs, because questionnaires repeat and the answers are already written down somewhere. The same rules apply with one extra: no answer in the security category leaves the building without the person who owns security, however well-sourced the draft looks. It also flags any answer whose evidence is older than the limit you set, so a control you changed in the spring does not get described the way it worked last year.

Does it submit the response?

No. It assembles the response in the format the buyer demanded, with the matrix, the citations and the gaps visible, and hands it over. There is no portal upload, no email to a procurement address and no e-signature in any configuration, because a submitted answer is a promise your company has made and a person should be the one making it.

What about pricing and legal terms?

It does not produce them. No rate, no discount, no total, no delivery date, no named person and no service level comes out of a model here. It can carry a figure a person has already signed off and show where that figure came from, and it can extract exactly what the buyer is asking for so the right person answers the right question. Deciding the number is human work.

Where do the buyer's documents go?

Wherever your handling rule says, and nowhere else. Some solicitations carry handling obligations and some buyers' material should not go into a general-purpose tool at all - that question gets settled in discovery, before anything runs. Processing can be confined to infrastructure you name, documents outside your rule are refused rather than quietly handled, and no client's material is ever visible to another.

Does it work with the tools we already use?

Yes, for most of them. The library is built from material you already hold - policy documents on a drive, past responses in a folder, evidence in a compliance tool, a content library in a proposal tool - and drafts are assembled where your team already writes, with the matrix and the gaps landing in the tracker you already watch. A folder of past responses works the same way. Naming a system means we can read or write it, not that we are partnered with it.

Is there a record of what we told each buyer?

Yes, and it is the quietly valuable part. Every answer records the source it cited, the staleness check, the owner, every edit a reviewer made, the model and prompt version that drafted it and who approved it. The next time a buyer asks the same question, or asks why the answer changed, you can show exactly what you said and what you were looking at when you said it.