Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →用于银行零售理财客户流失预警，当需要监测AUM下滑、赎回集中与到期流失风险并输出处置建议时触发。
.claude/skills/aifinlab-bank-t165-retail-finance-wealth-churn-warning-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-05 | ✗→✓ | ▲ Improved | 73% | 0% |
| case-11 | ✗→✓ | ▲ Improved | -48% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 24% | 0% |
| case-10 | ✗→✓ | ▲ Improved | 13% | 0% |
| case-12 | ✗→✓ | ▲ Improved | -44% | 0% |
对理财客户的AUM、赎回与到期结构进行预警监测，识别异常赎回、资产下滑与产品到期集中导致的潜在流失风险，输出风险分层、驱动因素与挽留建议。
scripts/wealth_churn_warning.py：计算AUM下滑、赎回比例与到期集中度，输出风险等级。bashpython scripts/wealth_churn_warning.py --input wealth.json --rules rules.json --output out.json
json{ "customer_id": "W2001", "aum": [ {"date": "2026-01-01", "value": 800000}, {"date": "2026-02-01", "value": 720000}, {"date": "2026-03-01", "value": 600000} ], "redemptions_30d": 150000, "maturity_60d": 200000 }
json{ "thresholds": { "aum_drop_pct": 0.2, "redemption_ratio": 0.3, "maturity_ratio": 0.4 } }
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-05 | fail→pass | 11,905 | 14,608 | +23% | 1 | 1 | 0% | 1,488 | 2,580 | +73% | 0 | 0 | — |
case-11 | fail→pass | 19,353 | 2,182 | -89% | 1 | 1 | 0% | 1,924 | 1,002 | -48% | 0 | 0 | — |
case-01 | pass→fail | 13,345 | 42,581 | +219% | 1 | 1 | 0% | 2,142 | 2,645 | +23% | 0 | 0 | — |
case-02 | pass→pass | 14,550 | 11,115 | -24% | 1 | 1 | 0% | 1,820 | 2,283 | +25% | 0 | 0 | — |
case-03 | fail→pass | 16,426 | 14,410 | -12% | 1 | 1 | 0% | 2,071 | 2,561 | +24% | 0 | 0 | — |
case-04 | pass→pass | 12,527 | 8,896 | -29% | 1 | 1 | 0% | 1,747 | 2,145 | +23% | 0 | 0 | — |
case-06 | pass→pass | 9,169 | 6,288 | -31% | 1 | 1 | 0% | 1,367 | 1,776 | +30% | 0 | 0 | — |
case-07 | pass→pass | 6,059 | 2,367 | -61% | 1 | 1 | 0% | 1,090 | 1,004 | -8% | 0 | 0 | — |
case-08 | pass→pass | 10,745 | 6,141 | -43% | 1 | 1 | 0% | 2,116 | 1,932 | -9% | 0 | 0 | — |
case-09 | pass→pass | 5,044 | 3,444 | -32% | 1 | 1 | 0% | 729 | 1,220 | +67% | 0 | 0 | — |
case-10 | fail→pass | 6,364 | 2,323 | -63% | 1 | 1 | 0% | 880 | 993 | +13% | 0 | 0 | — |
case-12 | fail→pass | 13,371 | 2,632 | -80% | 1 | 1 | 0% | 1,783 | 992 | -44% | 0 | 0 | — |
case-13 | pass→pass | 9,472 | 15,983 | +69% | 1 | 1 | 0% | 1,472 | 2,824 | +92% | 0 | 0 | — |
case-14 | pass→pass | 10,008 | 7,822 | -22% | 1 | 1 | 0% | 1,183 | 1,986 | +68% | 0 | 0 | — |
case-15 | pass→pass | 10,209 | 8,551 | -16% | 1 | 1 | 0% | 1,415 | 1,807 | +28% | 0 | 0 | — |
case-16 | fail→pass | 5,986 | 4,766 | -20% | 1 | 1 | 0% | 814 | 1,340 | +65% | 0 | 0 | — |
case-17 | pass→pass | 12,085 | 10,073 | -17% | 1 | 1 | 0% | 1,604 | 2,008 | +25% | 0 | 0 | — |
case-18 | fail→pass | 10,055 | 8,316 | -17% | 1 | 1 | 0% | 1,595 | 1,797 | +13% | 0 | 0 | — |
case-19 | fail→pass | 7,306 | 4,090 | -44% | 1 | 1 | 0% | 989 | 1,135 | +15% | 0 | 0 | — |
case-20 | pass→pass | 19,637 | 15,300 | -22% | 1 | 1 | 0% | 2,480 | 2,965 | +20% | 0 | 0 | — |
case-21 | fail→pass | 7,166 | 3,024 | -58% | 1 | 1 | 0% | 1,117 | 1,081 | -3% | 0 | 0 | — |
case-22 | fail→pass | 13,904 | 4,917 | -65% | 1 | 1 | 0% | 1,863 | 1,476 | -21% | 0 | 0 | — |
case-23 | pass→pass | 18,640 | 8,604 | -54% | 1 | 1 | 0% | 1,375 | 1,852 | +35% | 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. 23 cases were attempted. The headline lift of +39 percentage points is the difference between those two pass rates over the 23 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.