Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Turn an at-risk or churned account into a save play — root-cause hypothesis, the offer ladder, the outreach sequence, and the honest call on when to let go. Use when asked to save a churning customer, build a win-back plan, re-engage a lost account, or stop a renewal from slipping. Produces the churn diagnosis, a ranked set of save levers, a timed outreach sequence, and the walk-away line so you don't over-invest in an account that's gone.
.claude/skills/mohitagw15856-winback-playbook/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 41 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-15 | ✗→✓ | ▲ Improved | 24% | 0% |
| case-19 | ✗→✓ | ▲ Improved | 26% | 0% |
| case-01 | ✓→✓ | = Same ✓ | -7% | 0% |
| case-02 | ✓→✓ | = Same ✓ | -10% | 0% |
| case-03 | ✓→✓ | = Same ✓ | -36% | 0% |
Most churn saves fail because they lead with a discount before anyone knows why the customer is leaving. This starts with the diagnosis — is it value, fit, budget, a champion who left, or a competitor? — then matches the save lever to the cause, sequences the outreach, and sets a walk-away point so a healthy team doesn't burn a month chasing an account that was never coming back.
Ask for these if not provided:
Diagnosis: most likely cause(s), with evidence · Save odds: honest read
| Lever | Fixes which cause | Cost/effort | Use when | |---|---|---|---|
| Day | Who | Channel | Message focus | |---|---|---|---|
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | pass→pass | 21,568 | 21,063 | -2% | 1 | 1 | 0% | 3,791 | 3,521 | -7% | 0 | 0 | — |
case-02 | pass→pass | 28,604 | 20,974 | -27% | 1 | 1 | 0% | 3,765 | 3,378 | -10% | 0 | 0 | — |
case-03 | pass→pass | 48,951 | 21,749 | -56% | 1 | 1 | 0% | 6,791 | 4,325 | -36% | 0 | 0 | — |
case-04 | pass→pass | 21,805 | 21,832 | +0% | 1 | 1 | 0% | 2,703 | 3,597 | +33% | 0 | 0 | — |
case-05 | pass→pass | 22,844 | 27,901 | +22% | 1 | 1 | 0% | 2,959 | 4,502 | +52% | 0 | 0 | — |
case-06 | pass→pass | 23,718 | 17,949 | -24% | 1 | 1 | 0% | 2,901 | 3,613 | +25% | 0 | 0 | — |
case-07 | pass→pass | 20,278 | 19,192 | -5% | 1 | 1 | 0% | 2,267 | 3,034 | +34% | 0 | 0 | — |
case-08 | pass→pass | 21,458 | 20,996 | -2% | 1 | 1 | 0% | 2,501 | 3,159 | +26% | 0 | 0 | — |
case-09 | pass→pass | 20,538 | 18,441 | -10% | 1 | 1 | 0% | 2,235 | 2,803 | +25% | 0 | 0 | — |
case-10 | pass→pass | 18,390 | 20,590 | +12% | 1 | 1 | 0% | 2,576 | 3,374 | +31% | 0 | 0 | — |
case-11 | pass→pass | 18,767 | 21,302 | +14% | 1 | 1 | 0% | 2,105 | 3,114 | +48% | 0 | 0 | — |
case-12 | pass→pass | 17,948 | 16,428 | -8% | 1 | 1 | 0% | 2,013 | 3,201 | +59% | 0 | 0 | — |
case-13 | pass→pass | 23,929 | 17,628 | -26% | 1 | 1 | 0% | 2,888 | 3,364 | +16% | 0 | 0 | — |
case-14 | pass→pass | 17,824 | 19,193 | +8% | 1 | 1 | 0% | 2,652 | 2,959 | +12% | 0 | 0 | — |
case-15 | fail→pass | 21,666 | 20,145 | -7% | 1 | 1 | 0% | 2,529 | 3,131 | +24% | 0 | 0 | — |
case-16 | pass→pass | 24,467 | 18,865 | -23% | 1 | 1 | 0% | 2,953 | 2,881 | -2% | 0 | 0 | — |
case-17 | pass→pass | 21,661 | 20,817 | -4% | 1 | 1 | 0% | 2,593 | 3,207 | +24% | 0 | 0 | — |
case-18 | pass→pass | 24,377 | 17,026 | -30% | 1 | 1 | 0% | 3,037 | 3,488 | +15% | 0 | 0 | — |
case-19 | fail→pass | 19,820 | 20,994 | +6% | 1 | 1 | 0% | 2,334 | 2,940 | +26% | 0 | 0 | — |
case-20 | pass→pass | 21,263 | 16,164 | -24% | 1 | 1 | 0% | 2,054 | 3,252 | +58% | 0 | 0 | — |
case-21 | pass→pass | 21,928 | 16,524 | -25% | 1 | 1 | 0% | 2,657 | 2,548 | -4% | 0 | 0 | — |
case-22 | pass→pass | 14,032 | 19,045 | +36% | 1 | 1 | 0% | 1,390 | 3,127 | +125% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +9 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.