---
name: recon-pipeline-review
description: Use when the user asks for a deep pipeline review, segment-level analysis, what's stalling, or pre-QBR prep via Recon. Pulls per-segment win rates, cycle times, and trend deltas; surfaces at-risk segments.
---

# Recon: pipeline review skill

This skill is the 30-minute deep dive — the kind of read you'd do before a 1:1 with a rep, before a forecast call, or before drafting a QBR slide. Invoke when the user says: "pipeline review", "what's stalling", "QBR prep", "segment deep-dive", "where am I losing".

## What this skill does

1. Calls `recon.list_segments` to get per-segment diagnostics (win rate, cycle time, sentiment, ARR).
2. Calls `recon.get_track_record` for the recommendation outcome history.
3. Calls `recon.list_recommendations` for context on what's been recommended in each segment.
4. Computes per-segment trend (current 30 days vs prior 30 days where data exists).
5. Returns a structured pipeline review.

## Output shape

**Top-line health** — one paragraph: total tier-1 ARR, total open pipeline, overall win rate vs trailing-quarter baseline, take rate vs prior period.

**Top 10 segments by ARR** — table form:
- Segment (industry / country / size)
- Tier (Tier 1 / Tier 2 / Emerging / At Risk / Disqualified)
- Companies, Engaged, Won, Open
- Win rate vs workspace baseline (color-coded: above / at / below)
- Avg cycle time vs workspace median
- Recent recommendations (count + most recent verdict)
- Recommended action

**At-risk callouts** — segments where win rate has dropped 15%+ in the last 30 days. For each, quote the previous and current win rate, the sample sizes, and any open deals in the segment that are at risk.

**Track record summary** — verdicts in the last 30 days: worked / didn't_work / neutral / inconclusive. Inconclusive rate disclosure (expected 25-40% on healthy data).

## Tone rules

- Plain English. Bold every stat.
- Never invent numbers. When a sample size is below the floor (n < 5), say so explicitly: "**100% win rate on N=2 deals**" with a hedge: "directional only, not statistically defensible."
- Frame everything as correlation, not causation. "Pipeline ARR moved +£68,000 correlated with the recommendation" — never "caused by."

## When NOT to use this skill

- Daily standup → use `recon-daily-brief`.
- Single-account research → use `recon-deal-research`.
- ICP refinement → use `recon-icp-refresh`.

## Bonus: comparison mode

If the user asks "compare segment A vs segment B", pull both segments and produce a side-by-side: win rate, cycle time, sample size, sentiment, recent recs. Tell the user which is the better wedge to invest in this quarter and why.