Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →基于资产变化、交易活跃度、渠道行为、产品持有变化等多维信号，识别客户流失风险等级，输出预警结论和建议跟进动作。
.claude/skills/aifinlab-customer-churn-alert/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 76% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 23% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 28% | 0% |
| case-09 | ✗→✓ | ▲ Improved | 65% | 0% |
| case-11 | ✗→✓ | ▲ Improved | 30% | 0% |
银行零售客户流失往往有先兆信号，如AUM持续下降、交易频次骤减、产品逐步赎回等。但客户经理管户数量多，难以逐一跟踪每位客户的行为变化。需要系统化监测多维度流失信号，及时识别高风险客户并推送预警，为客户挽留争取时间窗口。
| 输入项 | 说明 | 是否必填 | |--------|------|----------| | 客户编号/客户列表 | 单个或批量客户标识 | 必填 | | AUM变化数据 | 近1/3/6个月AUM及变化趋势 | 必填 | | 交易活跃度 | 交易频次、金额变化趋势 | 必填 | | 渠道行为 | 手机银行/网银登录频次变化 | 选填 | | 产品持有变化 | 产品赎回、到期未续、持有数量变化 | 选填 | | 客户关系信号 | 客户经理互动频次、服务工单、投诉记录 | 选填 | | 服务体验 | 满意度评分变化、近期投诉情况 | 选填 | | 客户价值层级 | 当前分层结果（如有） | 选填 |
python# 调用 skill result = run_skill({ "param1": "value1", "param2": "value2" })
bashpython scripts/run_skill.py --input data.json
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-07 | pass→pass | 17,658 | 16,579 | -6% | 1 | 1 | 0% | 2,543 | 3,742 | +47% | 0 | 0 | — |
case-01 | fail→fail | 23,424 | 21,323 | -9% | 1 | 1 | 0% | 3,376 | 4,100 | +21% | 0 | 0 | — |
case-02 | fail→fail | 27,247 | 29,637 | +9% | 1 | 1 | 0% | 4,153 | 6,496 | +56% | 0 | 0 | — |
case-03 | fail→fail | 22,476 | 24,339 | +8% | 1 | 1 | 0% | 3,459 | 4,841 | +40% | 0 | 0 | — |
case-04 | fail→pass | 15,781 | 17,832 | +13% | 1 | 1 | 0% | 2,320 | 4,077 | +76% | 0 | 0 | — |
case-05 | pass→fail | 16,660 | 12,205 | -27% | 1 | 1 | 0% | 2,216 | 2,922 | +32% | 0 | 0 | — |
case-06 | fail→pass | 13,030 | 7,795 | -40% | 1 | 1 | 0% | 1,875 | 2,300 | +23% | 0 | 0 | — |
case-08 | fail→pass | 17,623 | 14,425 | -18% | 1 | 1 | 0% | 2,468 | 3,157 | +28% | 0 | 0 | — |
case-09 | fail→pass | 13,603 | 14,399 | +6% | 1 | 1 | 0% | 2,095 | 3,460 | +65% | 0 | 0 | — |
case-10 | pass→pass | 14,265 | 12,907 | -10% | 1 | 1 | 0% | 1,985 | 3,124 | +57% | 0 | 0 | — |
case-11 | fail→pass | 17,192 | 13,587 | -21% | 1 | 1 | 0% | 2,376 | 3,081 | +30% | 0 | 0 | — |
case-12 | fail→pass | 20,736 | 15,879 | -23% | 1 | 1 | 0% | 3,013 | 2,981 | -1% | 0 | 0 | — |
case-13 | fail→pass | 18,660 | 9,502 | -49% | 1 | 1 | 0% | 2,591 | 2,467 | -5% | 0 | 0 | — |
case-14 | fail→pass | 17,475 | 18,160 | +4% | 1 | 1 | 0% | 2,458 | 3,766 | +53% | 0 | 0 | — |
case-15 | fail→pass | 19,730 | 16,494 | -16% | 1 | 1 | 0% | 2,757 | 3,783 | +37% | 0 | 0 | — |
case-16 | pass→pass | 18,289 | 16,856 | -8% | 1 | 1 | 0% | 2,485 | 3,440 | +38% | 0 | 0 | — |
case-17 | pass→pass | 15,273 | 15,231 | -0% | 1 | 1 | 0% | 2,269 | 3,326 | +47% | 0 | 0 | — |
case-18 | fail→pass | 19,976 | 19,689 | -1% | 1 | 1 | 0% | 2,879 | 4,118 | +43% | 0 | 0 | — |
case-19 | fail→pass | 17,613 | 16,381 | -7% | 1 | 1 | 0% | 2,487 | 3,593 | +44% | 0 | 0 | — |
case-20 | fail→pass | 12,726 | 19,440 | +53% | 1 | 1 | 0% | 1,989 | 3,800 | +91% | 0 | 0 | — |
case-21 | fail→fail | 20,464 | 14,409 | -30% | 1 | 1 | 0% | 2,917 | 3,192 | +9% | 0 | 0 | — |
case-22 | fail→pass | 21,704 | 22,093 | +2% | 1 | 1 | 0% | 3,237 | 4,860 | +50% | 0 | 0 | — |
case-23 | fail→pass | 14,853 | 12,636 | -15% | 1 | 1 | 0% | 2,459 | 3,027 | +23% | 0 | 0 | — |
case-24 | fail→pass | 14,057 | 11,571 | -18% | 1 | 1 | 0% | 2,134 | 2,783 | +30% | 0 | 0 | — |
case-25 | fail→pass | 17,845 | 17,058 | -4% | 1 | 1 | 0% | 2,716 | 3,542 | +30% | 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. 25 cases were attempted. The headline lift of +60 percentage points is the difference between those two pass rates over the 25 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.