Install any skill in seconds. Free to start, no credit card required.
Get Started Free →零售客户流失预警与经营挽留助手,识别高流失风险客户并提供挽留策略。当用户需要客户流失预警、流失原因分析、挽留方案时使用。
.claude/skills/aifinlab-customer-churn-early-warning-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-07 | ✗→✓ | ▲ Improved | 57% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 114% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 41% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 100% | 0% |
| case-14 | ✗→✓ | ▲ Improved | 107% | 0% |
本技能用于银行零售金融场景下的客户流失预警分析与经营挽留建议生成。技能目标不是简单判断“客户是否会流失”,而是基于客户资产、交易、产品持有、渠道活跃、触达响应、服务体验、竞品迁移迹象以及行业对标数据,形成可解释、可执行、可跟踪的预警结论。
本技能特别适用于以下任务:
在出现以下情形时,优先使用本技能:
以下情形不建议直接使用本技能得出强结论:
优先提供以下信息:
输出内容至少应包括:
本技能采用“多维信号识别 + 行业对标校准 + 经营动作生成”的框架。
本技能对行业数据依赖较强,原因如下:
因此,在进行以下判断时必须提示行业数据支持程度:
references/churn_signal_catalog.mdreferences/industry_data_requirements.mdreferences/retention_playbook.mdreferences/compliance_notes.mdreferences/output_schema.mdscripts/customer_churn_signal_engine.py:识别客户流失预警信号。scripts/industry_benchmark_adjuster.py:结合行业基线对风险进行校准。scripts/render_churn_report.py:将结果渲染为 markdown 报告。| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-01 | fail→fail | 28,414 | 32,642 | +15% | 1 | 1 | 0% | 4,072 | 6,119 | +50% | 0 | 0 | — |
case-02 | pass→pass | 15,629 | 24,582 | +57% | 1 | 1 | 0% | 2,177 | 5,340 | +145% | 0 | 0 | — |
case-03 | pass→pass | 18,440 | 20,891 | +13% | 1 | 1 | 0% | 2,776 | 4,752 | +71% | 0 | 0 | — |
case-04 | pass→pass | 17,917 | 29,068 | +62% | 1 | 1 | 0% | 2,440 | 5,968 | +145% | 0 | 0 | — |
case-05 | pass→pass | 21,204 | 26,214 | +24% | 1 | 1 | 0% | 2,876 | 5,327 | +85% | 0 | 0 | — |
case-06 | pass→pass | 21,744 | 25,817 | +19% | 1 | 1 | 0% | 3,012 | 5,202 | +73% | 0 | 0 | — |
case-07 | fail→pass | 19,631 | 19,824 | +1% | 1 | 1 | 0% | 2,856 | 4,493 | +57% | 0 | 0 | — |
case-08 | fail→pass | 10,504 | 12,479 | +19% | 1 | 1 | 0% | 1,574 | 3,362 | +114% | 0 | 0 | — |
case-09 | pass→pass | 18,708 | 23,723 | +27% | 1 | 1 | 0% | 2,656 | 4,890 | +84% | 0 | 0 | — |
case-10 | pass→pass | 18,031 | 17,895 | -1% | 1 | 1 | 0% | 2,490 | 4,084 | +64% | 0 | 0 | — |
case-11 | fail→pass | 17,228 | 14,152 | -18% | 1 | 1 | 0% | 2,581 | 3,640 | +41% | 0 | 0 | — |
case-12 | pass→pass | 21,772 | 26,384 | +21% | 1 | 1 | 0% | 3,002 | 5,260 | +75% | 0 | 0 | — |
case-13 | fail→pass | 15,580 | 18,784 | +21% | 1 | 1 | 0% | 2,176 | 4,352 | +100% | 0 | 0 | — |
case-14 | fail→pass | 24,708 | 37,600 | +52% | 1 | 1 | 0% | 3,367 | 6,986 | +107% | 0 | 0 | — |
case-15 | pass→pass | 16,577 | 22,877 | +38% | 1 | 1 | 0% | 2,404 | 4,928 | +105% | 0 | 0 | — |
case-16 | pass→pass | 16,995 | 23,419 | +38% | 1 | 1 | 0% | 2,385 | 4,451 | +87% | 0 | 0 | — |
case-17 | pass→pass | 14,210 | 21,233 | +49% | 1 | 1 | 0% | 1,960 | 4,748 | +142% | 0 | 0 | — |
case-18 | pass→pass | 23,105 | 24,164 | +5% | 1 | 1 | 0% | 3,175 | 4,924 | +55% | 0 | 0 | — |
case-19 | fail→pass | 18,970 | 19,742 | +4% | 1 | 1 | 0% | 2,650 | 4,577 | +73% | 0 | 0 | — |
case-20 | fail→pass | 13,853 | 18,118 | +31% | 1 | 1 | 0% | 2,145 | 4,078 | +90% | 0 | 0 | — |
case-21 | fail→pass | 24,207 | 11,714 | -52% | 1 | 1 | 0% | 3,372 | 3,314 | -2% | 0 | 0 | — |
case-22 | pass→pass | 14,660 | 16,320 | +11% | 1 | 1 | 0% | 2,204 | 4,077 | +85% | 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 +36 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.