TheGood RFP.

RFP is Dead.
Long live the RFP.

RFP / Request for ProposalDescribe the custom AI solution or software project you need to build. Walk away with a technical specification you can take to any vendor, written so the right builders win.

Build your specFree. No signup until you preview the doc.

01 / THE PROBLEM

Most RFPs lose the project before it starts.

You've seen this happen. Probably more than once.

The document goes out. Ten vendors reply. The quotes vary by a factor of six. The cheapest one wins. Nine months later, scope is blown, the product ships late, and nobody can say exactly where it went wrong.

The RFP wasn't wrong because it was too short. It was wrong because it asked the wrong questions.

Vague on purpose

Written so broadly that every vendor can say yes. Which means nothing can be compared.

Feature lists, not outcomes

Two hundred checkboxes. Zero sentences about what success looks like for the business.

No real technical criteria

Nothing in the document lets a reader tell a serious engineering team from a body shop.

A race to the bottom

Price per hour becomes the only axis of comparison. The rest is a handshake.

02 / WHAT YOU GET

A document that does the work for you.

§DeliverableWhat it containsLength

Business objective, clarified

The real problem behind the request, rewritten in one paragraph you can defend to a board.

~ 1 pg

Proposed architecture

A reference architecture with component boundaries, data flow, and the decisions a vendor has to explicitly agree or disagree with.

~ 3 pg

Recommended stack

Languages, frameworks, data and infrastructure choices, each justified against your constraints rather than vendor preference.

~ 2 pg

Realistic effort estimate

A range, not a point. Broken down by workstream, with the assumptions that make the number move if they change.

~ 2 pg

Risk & dependency map

The parts most likely to slip, the integrations outside your control, and what "done" has to look like for each.

~ 2 pg

Vendor evaluation criteria

A scoring rubric that rewards outcome thinking, engineering rigor and discovery discipline. Not the lowest hourly rate.

~ 2 pg

03 / HOW IT WORKS

Three steps. No sales call.

STEP 01

Describe the project.

Business objective, constraints, integrations, timeline. Plain language. As messy as it really is, right now.

input / "Custom AI agent to triage
support tickets. Trained on
5 years of Zendesk data."

STEP 02

Receive the specification.

A complete technical document, structured like it was written by an engineering team that has shipped this kind of thing before.

output / 12 – 16 page spec
architecture, stack, estimate,
risks, evaluation rubric

STEP 03

Use it with any vendor.

Send it to anyone. Compare responses against your own criteria, not against a sales pitch. You don't owe us the project.

result / fewer vendors,
better answers, no race
to the bottom on price

Used by engineering leaders building software and custom AI, at companies that take both seriously.

Case studies · Q3 2026

Write the Good one.

Takes about fifteen minutes. You leave with the document, whether you ever talk to us or not.

Build your spec