---
name: recon-icp-refresh
description: Use when the user asks for ICP analysis, validated vs stated ICP comparison, drift detection, persona analysis, or "who actually buys" via Recon. Pulls multi-dimensional ICP profile + drift signals.
---

# Recon: ICP refresh skill

This skill compares what the customer SAID they target (during onboarding) vs what's ACTUALLY closing in their HubSpot. Invoke when the user says: "ICP review", "who actually buys", "validated ICP", "stated vs validated", "where am I drifting", "should I update my ICP doc".

## What this skill does

1. Calls `recon.list_segments` to get the firmographic shape of validated wins.
2. Reads workspace_profile (via internal admin API or returns surfaced via Recon MCP) for the stated ICP — target_industries, target_geos, target_employee_bands.
3. Computes drift between stated and validated.
4. Surfaces emerging segments (high engagement, no closed-won yet).
5. Produces an "update your ICP doc" recommendation.

## Output shape

**Stated ICP** (from onboarding):
- Industries you said you target
- Geos you said you target
- Sizes you said you target
- Who-we-sell-to summary

**Validated ICP** (from closed-won data):
- Top 5 industries closing, by ARR + deal count
- Top 5 countries closing
- Top 3 employee bands closing
- Sample size disclosure: "**confidence requires ≥5 closed-won**; below that, treat as directional"

**Drift detected**:
- Industries you stated but haven't closed in (count + ARR-at-risk if any open pipeline exists)
- Industries closing that weren't stated (count + ARR — opportunity to formalise)
- Same for geos and sizes

**Emerging wedges**:
- Segments with ≥10 engaged companies, positive sentiment, no closed-won yet
- For each: company count, engagement rate, sentiment mix
- Recommendation: run as Scout experiment vs hold

**Recommended ICP doc update**:
- Concrete diff: add X / drop Y / add Z size band
- Citation list (deal IDs grounding the recommendation)

## Tone rules

- Frame the gap honestly. If stated ICP doesn't match validated, say so plainly. "You said US enterprise; you've closed UK mid-market." This is the job.
- Always cite. Every claim has deal IDs or company IDs.
- Sample-size honesty. With <5 closed-won, every claim is directional.

## Persona analysis (bonus)

When the user asks "who buys?" or "what titles drive my closed-won?", switch to persona mode:

1. Pull contacts grouped by job_title.
2. Cross-reference with each contact's company tier.
3. Surface top 10 titles by tier-1 company count.
4. Add the engagement signal (positive vs negative replies in last 90 days).

## When NOT to use this skill

- Daily standup → `recon-daily-brief`.
- Single-deal research → `recon-deal-research`.
- Pipeline operations → `recon-pipeline-review`.