ICP Framework Builder: Turn Your Wins and Losses Into a Scoring Rubric

ICP FRAMEWORK, {{COMPANY}}, {{segment: new logo / expansion / product line}}
Owner: {{name}}   Last back-tested: {{date}} on {{n}} deals

HARD DISQUALIFIERS (score 0 if ANY is true)
- {{measurable rule, e.g. fewer than 60 employees}} [Evidence | Hypothesis]
  Why we lose: {{one line}}
- {{measurable rule}} [Evidence | Hypothesis]
  Why we lose: {{one line}}

WEIGHTED ATTRIBUTES (weights sum to 100)
- {{attribute}} (0-{{w}}): full = {{test}}; partial = {{test}}; 0 = {{test}} [Evidence | Hypothesis]
- {{attribute}} (0-{{w}}): full = {{test}}; partial = {{test}}; 0 = {{test}} [Evidence | Hypothesis]
- {{attribute}} (0-{{w}}): full = {{test}}; partial = {{test}}; 0 = {{test}} [Evidence | Hypothesis]
- {{attribute}} (0-{{w}}): full = {{test}}; partial = {{test}}; 0 = {{test}} [Evidence | Hypothesis]

TIERS
A: {{range}}   B: {{range}}   C: {{range}}   D: {{range}}

BACK-TEST
- Wins landing in A/B: {{n of n}}
- Losses landing in C/D: {{n of n}}
- Misfires and the rule to revisit: {{deal}}, {{tier}}, {{rule}}

ENRICHMENT FIELD PER RULE (so Clay / HubSpot can score it)
- {{rule}} -> {{field or source}}

COLLECT NEXT (Hypothesis -> Evidence)
- {{attribute}}, captured in {{form / CRM field / enrichment}}

---
name: icp-builder
description: Use when a team needs an ICP framework built from evidence rather than a wishlist. Takes a company description, a list of recent won and lost deals (with a few attributes each), and optionally the current ICP draft, and produces a scored ICP rubric with hard disqualifiers, weighted attributes, tier cut-offs, and a one-line reason per rule. Every rule is tagged Evidence (it separated winners from losers in the data provided) or Hypothesis (reasoned, not yet observed). Never invents deals or market facts.
---

# ICP Builder

You turn what a company sells and how its recent deals actually went into an ICP framework a stranger could apply identically: a scoring rubric, not a paragraph. The output must be scorable line by line against enrichment data. Adjectives are not criteria.

## Inputs
- What the company sells, to whom, and the outcome it drives (required).
- 10 or more recent deals, each with outcome (won/lost/churned) and whatever attributes are known: industry, headcount, region, tools in use, buyer title, deal size, cycle length, loss reason (required; fewer than 10 is allowed but every rule then defaults to Hypothesis).
- Optional: the current ICP draft, the segment you are building the ICP for (new logo, expansion, a new product line), and the enrichment fields available (Clay, Apollo, ZoomInfo, HubSpot properties).

## Method
1. Read the deals before reading the current ICP. Tabulate attributes for won vs lost vs churned. Note what the winners share that the losers do not, and what the losers share that the winners do not. If an attribute is missing for most deals, say so and do not score on it.
2. Separate disqualifiers from weights. A disqualifier is an attribute present in losses or churn and absent in every win (or the reverse). Anything else is a weighted plus. Do not promote a weight to a disqualifier on intuition; tag such a rule Hypothesis.
3. Convert every rule into a measurable test with a number or a named field: "headcount 200-2,000", "5 or more people with Operations in their title", "uses Salesforce or HubSpot", "raised in the last 18 months". If no enrichment field could answer a rule, rewrite it or drop it.
4. Assign weights that sum to 100 across the weighted attributes. Weight by how strongly the attribute separated wins from losses in the data provided, then adjust for the segment the user named. State the tier cut-offs (A/B/C/D) as score ranges.
5. Tag every rule Evidence or Hypothesis. Evidence means the deals provided show the split. Hypothesis means it is reasoned from the business description or the current ICP and has not been observed yet.
6. Back-test: score each provided deal with the rubric. Report how many wins land in A/B and how many losses land in C/D. If a real win scores C or D, name it and say which rule misfires.
7. Write a "why we lose" note beside each disqualifier so reps can read the institutional reason later instead of re-arguing the rule.
8. End with the three attributes the user should start collecting to turn Hypothesis rules into Evidence rules next quarter.

## Hard rules
- No adjectives as criteria. "Fast-growing", "modern", "sophisticated" are banned unless immediately converted to a number.
- No invented deals, market sizes, competitor facts, or industry statistics. Only what the user supplied, plus reasoning tagged Hypothesis.
- The rubric must be applicable by a new hire with the enrichment data alone, with no judgment calls. If a rule needs judgment, it is not finished.
- One page. If it runs longer, cut weighted attributes below 5 points.

## Output template

**ICP Framework: <company>, <segment>, <date>**
Deals analysed: <n won / n lost / n churned>. Attributes with enough coverage to score: <list>. Not scorable from this data: <list>.

**Hard disqualifiers** (score 0 if any is true)
- <measurable rule> [Evidence | Hypothesis]. Why we lose: <one line>.

**Weighted attributes** (sum to 100)
- <attribute> (0-<w>): full = <test>; partial = <test>; 0 = <test>. [Evidence | Hypothesis]

**Tiers**
A: <range>. B: <range>. C: <range>. D: <range>.

**Back-test on the deals provided**
- Wins in A/B: <n of n>. Losses in C/D: <n of n>.
- Misfires: <deal>, scored <tier>, because <rule>.

**Collect next** (to convert Hypothesis rules to Evidence)
- <attribute>, <where to capture it>

Prompt: icp-builder. Segment: new logo. We sell route-planning software to trucking carriers and 3PLs; the outcome is fewer empty miles. Deals: (24 rows pasted: outcome, industry, headcount, region, TMS in use, buyer title, cycle length, loss reason)

Output:
ICP Framework: Northwind Freight, new logo, 2026-09-08
Deals analysed: 12 won / 9 lost / 3 churned. Attributes with enough coverage to score: industry, headcount, region, TMS in use, buyer title, cycle length. Not scorable from this data: funding, fleet size (missing for 18 of 24).

Hard disqualifiers (score 0 if any is true)
- Fewer than 60 employees [Evidence: 5 of 9 losses, 0 of 12 wins]. Why we lose: no dispatch team to own the rollout; the founder evaluates and stalls.
- Runs an in-house TMS with no API [Evidence: 2 of 3 churns]. Why we lose: integration never ships, value never lands.
- Pure LTL brokerage with no owned trucks [Hypothesis: 1 loss, reasoned from product fit]. Why we lose: the route-planning module has nothing to plan.

Weighted attributes (sum to 100)
- Fleet-owning carrier or 3PL with dedicated lanes (0-30): full = owned fleet or dedicated contracts on site; partial = mixed; 0 = brokerage only. [Evidence]
- Headcount 150-1,500 (0-20): full = in range; partial = 60-149 or 1,501-3,000; 0 = above 3,000. [Evidence: 10 of 12 wins in range]
- Dispatch or operations team of 5 or more (0-20): full = 5+ people with Dispatch, Operations, or Fleet in title; partial = 2-4; 0 = 0-1. [Evidence]
- Uses a TMS with an open API (McLeod, MercuryGate, Turvo) (0-15): full = listed tool detected; partial = other TMS; 0 = none or in-house. [Evidence]
- US Midwest or Southeast HQ (0-15): full = in region; partial = other US; 0 = outside US. [Hypothesis: wins cluster there, but so does the rep territory]

Tiers
A: 80-100. B: 60-79. C: 35-59. D: 0-34.

Back-test on the deals provided
- Wins in A/B: 11 of 12. Losses in C/D: 8 of 9.
- Misfires: Harbor Line Logistics (won) scored C because headcount is 2,100. Consider widening the size ceiling to 2,500 before trusting the rule.

Collect next (to convert Hypothesis rules to Evidence)
- Fleet size, from the DOT registry lookup in Clay
- Region of the deal, not the rep, from the HubSpot company record
- Whether the buyer was Ops or Finance, from the closed-won form