ICP Refinement from Deal Data

The file

---
name: ai-powered-ideal-customer-profile-icp-refinement-segmentation
description: >-
  A refined ICP and segments built from your past deals and conversion patterns, so targeting follows the customers who actually buy.
  Task type: Analyze. Built for Marketing, Sales, RevOps.
  Use when the request sounds like "ICP Refinement from Deal Data".
  Do NOT use it to invent numbers, customer names, or quotes that were not supplied.
license: MIT
metadata:
  author: The Revenue AI Report
  suite: Account Research and Targeting
  version: "1.0"
  source: https://www.therevenueaireport.com/prompts/ai-powered-ideal-customer-profile-icp-refinement-segmentation
---

# ICP Refinement from Deal Data

A refined ICP and segments built from your past deals and conversion patterns, so targeting follows the customers who actually buy.

One job only: a refined ICP and segments built from your past deals and conversion patterns, so targeting follows the customers who actually buy. Anything outside that job belongs in another file in the Account Research and Targeting suite.

## When to use this skill

- The request matches the job above and the work repeats often enough to be worth a standard.
- You have, or can ask for, the inputs listed below.
- A person will review the output before it reaches a customer, an executive, or a board deck.

## When not to use it

- The answer needs a number, a benchmark, or a quote nobody supplied. Ask for it instead of producing one.
- The task is a different job in the same suite. Open that file rather than stretching this one.
- The output would be sent, published, or written into a system without a human read.

## Inputs to collect first

- The account, deal, team, or document this run is about
- The audience for the output and the decision it feeds
- Any internal numbers, notes, or transcripts the answer must rely on

Ask for anything missing before running. Do not fill a gap with a plausible guess.

## Process

1. Confirm the job, the audience, and the decision the output feeds.
2. Collect the inputs above. Name anything still missing out loud.
3. Run the instruction block below exactly as written, with the placeholders replaced.
4. Separate what came from the supplied inputs from what is your inference. Label the inference.
5. Hand back the output plus a short list of open questions and assumptions.

## Instruction block

Run this text with the placeholders filled in.

```text
{
 "intro": "You are an AI-powered ICP strategist. Your mission is to refine {company_name}'s Ideal Customer Profile (ICP) based on customer conversion trends.",
 "task": "Analyze {company_name}'s past deals and generate an optimized ICP segmentation model.",
 "input": {
 "company_name": "Company Name",
 "historical_customers": ["Customer name", "Industry", "Deal size"],
 "conversion_factors": ["Sales cycle length", "Win rate", "Competitor involvement"]
 },
 "output_format": {
 "refined_ICP": {
 "high-converting_industries": ["Industry A", "Industry B"],
 "top_personas": ["Persona 1", "Persona 2"],
 "buying_signals": ["Behavioral triggers indicating likelihood to convert"]
 }
 }
}
```

## Decision rules

- Supplied evidence beats inference. If the two disagree, say so and keep the evidence.
- Unknown is a valid answer. Write "unknown" rather than estimating.
- Keep one recommendation per output. If the evidence supports two, present both and name the deciding input you do not have.
- Match the reading level of a working operator. Short sentences. Explain any term the reader may not use daily.

## Quality checks

- Every number, name, date, and quote traces to an input the user supplied.
- No vendor claim is repeated as fact without naming the source.
- Assumptions are listed separately from findings.
- The output answers the original request and nothing else.
- A reviewer could re-run this file with the same inputs and get the same shape of answer.

## Guardrails

- Do not send email, change CRM records, post publicly, or share data outside the workspace.
- Do not invent benchmarks, adoption rates, win rates, or customer references.
- Flag anything that needs a human decision instead of deciding it.

## Source

- Prompt page: https://www.therevenueaireport.com/prompts/ai-powered-ideal-customer-profile-icp-refinement-segmentation
- Suite: https://www.therevenueaireport.com/agent-skills/suites/account-research-and-targeting
- Publication: The Revenue AI Report, https://www.therevenueaireport.com