L2product

RFP Engine

Compounding answer library — answer once, auto-fill forever, win rate climbs.

720

Priority Score

Pain × Demand × Edge × Trend × Conversion

1Principles

The compounding tenth of RFP.

Why does this matter?

Struggling moment: RFP lands. 3 weeks to respond. Half the questions answered before. Team copy-pastes from old Word docs. 20-40 hours. 70% duplicate effort.

What we build: Answer library that compounds. Write once, approve, auto-fill next time. Each bid makes the next one faster.

What's out: Document generation (PDF/PPTX), team assignment workflows, AI classification. Those need a different appetite.

The drawings that prove the shape

Picture

Fat marker: Answer library with 3 entries. Arrow to new venture showing 2 of 5 auto-filled

2Performance

Auto-fill rate over 50%.

How will I know it's working?

North star: auto_filled_answers / total_questions > 0.50

Appetite: 6 weeks. The Dot Grid principle — not a full RFP platform. Just the compounding part.

Known (thin tail)	Unknown (check)
Venture kanban — live	Auto-fill matching quality
Q&A CRUD — live	Library seeding — chicken-and-egg
Answer editor — built	SME adoption — will they use the queue?

Kill signal: Library has 0 queries after 14 days.

Thin tail assessment

Picture

Fat marker: Graph — library growth over time. X: ventures. Y: fill rate climbing 0 to 70%

3Platform

Four places. One flywheel.

What do I already have?

Place	Affordances	Connection
Venture List	Kanban by stage, create new	→ Venture Detail
Venture Detail	Q&A list, trigger AI, fill rate %	→ Answer Editor, ← Library
Answer Editor	AI draft, manual override, approve	→ Library (on approve)
Library	Search approved, filter by venture	→ Venture Detail (auto-fill)

The flywheel: each approved answer increases the next venture's fill rate.

Build contract

Picture

Fat marker breadboard: Venture List → Venture Detail → Answer Editor → Library → auto-fill back

4Process

One rabbit hole: matching quality.

What could go wrong?

Hole	Severity	Resolution
Auto-fill returns wrong answers	Significant	Patch: start with keyword matching
Empty library — 0 entries	Significant	Patch: manual seed from 1 real RFP
Fill rate shows 70% with 0 data	Minor	Patch: fix calculation
Go/No-Go not built	Minor	Cut: out of this appetite

Dependency evidence

Picture

Fat marker: Two question cards with '78% keyword overlap' label and question mark

5Players

Bid manager drowning in copy-paste.

Who needs this and why now?

ICP: Bid manager at construction/solar EPC firm. 5+ bids per quarter. Rebuilding proposals from scratch each time.

Hidden objection: "AI answers will be generic." The unlock: SME approval gate — every auto-filled answer was approved by a human first.

What's explicitly out:

CRM contacts and deals (separate PRD)
Document generation (PDF/PPTX)
Team assignment and reviewer routing
AI document classification
Analytics dashboard (currently 404)

Story contracts

Picture

Fat marker: Person at desk surrounded by Word docs. Bubble: 'Didn't we answer this?'

1 / 5

The pitch is the shape. The pictures prove the thinking. The spec is the test contract.

Pictures Spec

Problem

Situation

RFP lands. 3 weeks to respond. Half the questions answered before in previous bids. Team copy-pastes from old Word docs. 20-40 hours per response. 70% is duplicate effort.

Intention

Answer library that compounds. Write an answer once, approve it, auto-fill it next time. Each bid makes the next one faster. The Dot Grid of bids — not a full RFP platform, just the compounding part.

Obstacle

Chicken-and-egg: auto-fill needs library entries. Library needs someone to write and approve answers. Nobody writes answers without seeing the auto-fill value first.

Hardest Thing

Hidden objection: 'AI-generated answers will be generic and need so much editing they're not worth it.' The unlock: SME approval gate — every auto-filled answer was approved by a human first.

Scorecard

Priority (5P)

5/5

Pain

3/5

Demand

4/5

Edge

4/5

Trend

3/5

Convert

Readiness (5R)

Principles4 / 5

Performance3 / 5

Platform5 / 5

Process3 / 5

Players3 / 5

What Exists

Component	State	Gap
Venture kanban	Working	Cards show stage, linked deal, deadline.
Venture detail + Q&A	Working	Questions list, answer editing, approval status.
AI answer generation	Partial	Endpoint exists. Returns nothing (empty library).
Auto-fill from library	Partial	Algorithm works. 0 approved answers to match.
Answer library	Broken	Built but 0 entries. Fill rate displays 70% (misleading).
SME review queue	Partial	Queue exists. No entries to review.
Deck generation	Partial	Endpoint exists. Not tested with real data.
Document upload	Working	Files persist, linked to venture.
Go/No-Go scoring	Missing	Not built. 8-criteria checklist specified.
RFP analytics	Broken	/crm/analytics returns 404.

Relationships

PRD	Contributes
Sales CRM	Peer PRD. RFP depends on CRM contacts and deals.
Identity & Access	Auth dependency
Construction Industry	Domain context

Kill Signal

Ventures but no answers = project board. Library has 0 queries after 14 days = adoption failed.

Questions

What seeds the auto-fill flywheel — and what happens if the library stays empty?

If the hidden objection is 'AI answers are generic,' does answer-once-auto-fill-forever actually overcome it?
At what library size does the compounding become visible enough to change behavior?
Should Go/No-Go scoring live here or in the CRM — where does the decision actually happen?