Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Simulate the exact customer who will quietly cancel in month 4 — their internal monologue through the lifecycle and the honest exit interview they never gave you. Use when asked why do customers really churn, simulate a churning customer, roleplay the customer who cancels, or what does silent churn look like for my product. Produces the customer's lifecycle monologue, their never-given exit interview, and a debrief with the earliest detectable signals and interventions.
.claude/skills/mohitagw15856-the-churning-customer/SKILL.md| Model | Eval pass | Runs |
|---|---|---|
| gemini-3.6-flash | 100% | 4 |
| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-02 | ✗→✓ | ▲ Improved | 35% | 0% |
| case-07 | ✗→✓ | ▲ Improved | 95% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 220% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 31% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 71% | 0% |
The customers who churn loudest are the least dangerous — they tell you why. This skill simulates the dangerous one: the customer who onboards politely, disappoints quietly, ignores the renewal email, and never files a ticket. It reconstructs their inner monologue so you can meet them before month 4 does. (For the data-side view, use churn-analysis; this is the human inside the cohort.)
Ask for these if not provided:
Simulate the monologue at exactly these points — churn is decided at moments, not gradually:
| Moment | The question in the customer's head | |---|---| | 1. Purchase rationalization (day 0) | "What did I just tell my boss/spouse this would do?" | | 2. First value attempt (day 2–7) | "Is this doing the thing I bought it for, or am I doing extra work?" | | 3. The silent disappointment (week 2–6) | The moment expectations quietly reprice — usually never spoken | | 4. The workaround (month 2–3) | "It's easier to just spreadsheet/old tool/ignore it]" — churn is now decided | | 5. The renewal email (month 4+) | Reads the price with fresh eyes; cancellation is administration, not decision |
The monologue must be specific to THIS product's actual onboarding, in a believable human voice — mildly busy, non-technical unless the buyer is technical, never cartoonishly angry. Quiet disappointment, not rage.
> Simulation — a plausible composite, not a prediction. Validate against real churned-customer interviews.
First-person monologue at each moment, 3–6 sentences each, referencing the product's real onboarding steps]
What did you buy this to do? · When did you first doubt it? · What did you replace it with? · What would have kept you? (honest answer, which is usually smaller than a feature) · Why didn't you tell us?
| Signal | Detectable when | Where it shows | Intervention | |---|---|---|---| 3 signals, each earlier than the last section's moment 4]
One paragraph: the single change to onboarding that moves moment 2 from "extra work" to "did the thing."
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 33,300 | 27,876 | -16% | 1 | 1 | 0% | 3,990 | 4,235 | +6% | 0 | 0 | — |
case-02 | fail→pass | 22,809 | 27,897 | +22% | 1 | 1 | 0% | 3,252 | 4,397 | +35% | 0 | 0 | — |
case-03 | fail→fail | 35,770 | 28,145 | -21% | 1 | 1 | 0% | 4,521 | 4,554 | +1% | 0 | 0 | — |
case-04 | pass→pass | 15,652 | 22,997 | +47% | 1 | 1 | 0% | 3,061 | 4,437 | +45% | 0 | 0 | — |
case-05 | pass→pass | 22,322 | 39,342 | +76% | 1 | 1 | 0% | 2,603 | 4,240 | +63% | 0 | 0 | — |
case-06 | pass→fail | 20,177 | 34,583 | +71% | 1 | 1 | 0% | 2,344 | 5,708 | +144% | 0 | 0 | — |
case-07 | fail→pass | 21,808 | 25,027 | +15% | 1 | 1 | 0% | 2,061 | 4,021 | +95% | 0 | 0 | — |
case-08 | fail→pass | 10,105 | 28,014 | +177% | 1 | 1 | 0% | 1,431 | 4,574 | +220% | 0 | 0 | — |
case-09 | fail→fail | 28,535 | 17,395 | -39% | 1 | 1 | 0% | 2,017 | 3,532 | +75% | 0 | 0 | — |
case-10 | fail→pass | 28,588 | 23,714 | -17% | 1 | 1 | 0% | 3,016 | 3,960 | +31% | 0 | 0 | — |
case-11 | fail→fail | 21,782 | 31,649 | +45% | 1 | 1 | 0% | 2,573 | 4,267 | +66% | 0 | 0 | — |
case-12 | fail→fail | 28,440 | 30,859 | +9% | 1 | 1 | 0% | 2,702 | 4,307 | +59% | 0 | 0 | — |
case-13 | fail→fail | 41,857 | 37,584 | -10% | 1 | 1 | 0% | 5,935 | 5,245 | -12% | 0 | 0 | — |
case-14 | fail→pass | 19,980 | 21,834 | +9% | 1 | 1 | 0% | 2,123 | 3,625 | +71% | 0 | 0 | — |
case-15 | fail→pass | 17,562 | 25,307 | +44% | 1 | 1 | 0% | 1,854 | 3,513 | +89% | 0 | 0 | — |
case-16 | fail→pass | 19,785 | 35,021 | +77% | 1 | 1 | 0% | 2,078 | 5,553 | +167% | 0 | 0 | — |
case-17 | fail→fail | 15,330 | 23,421 | +53% | 1 | 1 | 0% | 2,098 | 3,282 | +56% | 0 | 0 | — |
case-18 | fail→fail | 15,633 | 11,881 | -24% | 1 | 1 | 0% | 1,261 | 1,886 | +50% | 0 | 0 | — |
case-19 | fail→pass | 13,409 | 22,273 | +66% | 1 | 1 | 0% | 1,644 | 2,263 | +38% | 0 | 0 | — |
case-20 | fail→pass | 28,275 | 21,341 | -25% | 1 | 1 | 0% | 3,112 | 3,771 | +21% | 0 | 0 | — |
case-21 | fail→pass | 28,928 | 26,665 | -8% | 1 | 1 | 0% | 3,467 | 3,957 | +14% | 0 | 0 | — |
case-22 | fail→pass | 22,268 | 34,569 | +55% | 1 | 1 | 0% | 2,180 | 5,181 | +138% | 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 +45 percentage points is the difference between those two pass rates over the 22 comparable cases. 1 case got worse with the skill loaded, and it is included in that figure.
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.