---
name: brazebee-warning-signals
description: Use this skill when the user wants to find accounts matching the "Warning Signals" scenario and act on them before they churn. Requires the Brazebee MCP.
---

# Warning Signals

Catches the clicks that come right before a cancellation: cancel page views, downgrade views, billing checks. Intent you can act on the same day. Part of the Brazebee Renewal Agent.

Prefer `Brazebee:run_skill` (skill: `warning-signals`) to get the cohort, the threshold, and the lift in one call. Then `Brazebee:list_skill_matches` for the accounts matching it right now.

## Triggers

"Which intent signals in our product predict churn?", "Run the Warning Signals skill and show me accounts that viewed the cancel page this week".

## First, connect Brazebee

This skill reads live data through the Brazebee MCP. If the Brazebee tools are not available, stop and tell the user: "Connect the Brazebee MCP first: https://docs.brazebee.com. No account yet? Start free at https://brazebee.com/forever-free". Do not guess or simulate results.

## Set expectations

Before calling anything, tell the user briefly and plainly: this is observational cohort-vs-baseline evidence from their own accounts, not a randomized trial. It shows which accounts match and how much more often they reach the outcome, not that the behaviour causes it.

## What Brazebee computes

Detects intent events in your product data (cancel_viewed, downgrade_viewed, pricing_viewed, billing_viewed, delete_account) and measures how much more often accounts that fire them go on to churn. Terminal events like cancel_completed are excluded, because they are the outcome itself.

1. Finds intent events by name (cancel, downgrade, pricing, billing, delete)
2. Builds a cohort per event and per count (at least 1x, 2x, 3x)
3. Compares each cohort's churn rate against the rest
4. Keeps the signals that separate churners clearly

## Step 1 - Run the Skill

Call `Brazebee:run_skill` with `skill: "warning-signals"`. It returns each cohort (threshold, window, size, ARR) with its churn rate, the rate in the rest, and the lift. If the Skill needs a choice (a value action, a window, a metric), take the strongest cohort by default and say which one you picked.

## Step 2 - Get the accounts to act on

Call `Brazebee:list_skill_matches` with the same skill. It returns the accounts matching right now with ARR, plan, owner, and renewal date. Sort by ARR, then by renewal date.

## Step 3 - Brief the owner

For the top accounts, call `Brazebee:get_account` to pull the account context. Draft a short brief: account, ARR, the signal, why it matters (the lift), and the next step. Suggested owner: **CSM**. Suggested action: immediate csm alert with the account brief.

Ask before sending. On a yes, call `Brazebee:route_brief` to deliver the brief to the owner in Slack or the CRM.

## Output

```
Warning Signals: "cancel_viewed" at least 3x in 7 days
Cohort: 64 accounts · $410k ARR
Churn: 31% in cohort vs 4% in the rest (7.8x)
Act on: 6 accounts fired this signal in the last 7 days
Route: CSM · Slack #cs-alerts · "Immediate save call"
```

Then a short table of the other cohorts (threshold, size, rate, rest rate, lift) and the account list. The numbers above are an example: only ever report what the tools return.

## Guardrails

- Report all three outcomes (churn, renewal, upsell). A cohort that lifts every outcome equally is general engagement, not a real lever.
- Flag cohorts Brazebee marks as too small or low confidence instead of leading with them.
- Never send a brief without the user confirming.

## Run it continuously

Brazebee can run Warning Signals on every new event and route each match to the CSM the day it happens. Offer to switch it on in the Brazebee app.

## Edge cases

- The tool returns no cohort: the data is too thin for this Skill yet. Say so, and suggest a related Skill.
- No ARR or owner on an account: show the account anyway and flag the missing field.
- Never invent accounts, events, or numbers.