Loading skill
Install any skill in seconds. Free to start, no credit card required.
Get Started Free →Use when planning retail credit-card installment marketing, customer segmentation, and outreach prioritization; trigger for requests that need Chinese skill content with clear inputs/outputs, compliance-aware actions, and optional scripts for segmentation.
.claude/skills/aifinlab-bank-t158-retail-finance-installment-marketing-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-21 | ✗→✓ | ▲ Improved | 42% | 0% |
| case-02 | ✗→✓ | ▲ Improved | 33% | 0% |
| case-03 | ✗→✓ | ▲ Improved | 23% | 0% |
| case-04 | ✗→✓ | ▲ Improved | -19% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 25% | 0% |
用于信用卡分期营销的客群分层、触达策略与行动清单输出，确保动作合规、可执行且可复盘。
customer_id、卡片类型、额度区间批量分层使用 scripts/installment_marketing_segment.py。
bashpython scripts/installment_marketing_segment.py --input installment_candidates.json --output installment_segmented.json
输入 JSON 示例：
json{ "customers": [ { "customer_id": "C003", "largest_purchase_3m": 9000, "revolving_ratio": 0.6, "installment_history": 2, "on_time_rate_12m": 0.99, "risk_flag": 0 } ] }
输出关键字段：
segment：高潜/中潜/低潜score：分层评分（用于排序）offer：推荐优惠策略text客户：{customer_id} 分层：{segment} 建议策略：{offer} 理由：{reasons} 待补信息：{missing_fields}
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-15 | pass→pass | 12,573 | 12,141 | -3% | 1 | 1 | 0% | 1,174 | 2,469 | +110% | 0 | 0 | — |
case-21 | fail→pass | 12,505 | 9,877 | -21% | 1 | 1 | 0% | 1,620 | 2,296 | +42% | 0 | 0 | — |
case-09 | pass→pass | 12,283 | 23,797 | +94% | 1 | 1 | 0% | 2,029 | 2,404 | +18% | 0 | 0 | — |
case-01 | fail→fail | 30,516 | 31,070 | +2% | 1 | 1 | 0% | 3,902 | 4,240 | +9% | 0 | 0 | — |
case-02 | fail→pass | 23,598 | 25,992 | +10% | 1 | 1 | 0% | 4,689 | 6,222 | +33% | 0 | 0 | — |
case-03 | fail→pass | 20,825 | 15,766 | -24% | 1 | 1 | 0% | 2,569 | 3,155 | +23% | 0 | 0 | — |
case-04 | fail→pass | 8,671 | 2,484 | -71% | 1 | 1 | 0% | 1,237 | 1,002 | -19% | 0 | 0 | — |
case-05 | fail→pass | 19,565 | 23,797 | +22% | 1 | 1 | 0% | 3,028 | 3,789 | +25% | 0 | 0 | — |
case-06 | pass→pass | 19,613 | 17,102 | -13% | 1 | 1 | 0% | 2,338 | 2,829 | +21% | 0 | 0 | — |
case-07 | fail→pass | 19,501 | 16,270 | -17% | 1 | 1 | 0% | 2,844 | 2,731 | -4% | 0 | 0 | — |
case-08 | fail→pass | 13,330 | 6,589 | -51% | 1 | 1 | 0% | 2,060 | 1,569 | -24% | 0 | 0 | — |
case-10 | pass→pass | 19,345 | 15,855 | -18% | 1 | 1 | 0% | 2,414 | 2,765 | +15% | 0 | 0 | — |
case-11 | pass→pass | 18,712 | 17,907 | -4% | 1 | 1 | 0% | 2,616 | 3,065 | +17% | 0 | 0 | — |
case-12 | pass→pass | 18,134 | 15,817 | -13% | 1 | 1 | 0% | 2,632 | 2,838 | +8% | 0 | 0 | — |
case-13 | fail→fail | 27,348 | 35,170 | +29% | 1 | 1 | 0% | 4,956 | 5,882 | +19% | 0 | 0 | — |
case-14 | fail→fail | 20,515 | 18,438 | -10% | 1 | 1 | 0% | 2,856 | 2,932 | +3% | 0 | 0 | — |
case-16 | pass→pass | 17,520 | 20,540 | +17% | 1 | 1 | 0% | 2,586 | 3,229 | +25% | 0 | 0 | — |
case-17 | fail→pass | 20,282 | 6,215 | -69% | 1 | 1 | 0% | 1,956 | 1,429 | -27% | 0 | 0 | — |
case-18 | fail→pass | 11,232 | 3,279 | -71% | 1 | 1 | 0% | 1,774 | 1,101 | -38% | 0 | 0 | — |
case-19 | pass→pass | 24,223 | 20,578 | -15% | 1 | 1 | 0% | 3,005 | 3,568 | +19% | 0 | 0 | — |
case-20 | fail→pass | 14,275 | 2,705 | -81% | 1 | 1 | 0% | 1,922 | 1,085 | -44% | 0 | 0 | — |
case-22 | pass→pass | 16,011 | 15,364 | -4% | 1 | 1 | 0% | 2,282 | 2,867 | +26% | 0 | 0 | — |
case-23 | pass→pass | 20,179 | 12,416 | -38% | 1 | 1 | 0% | 2,506 | 2,283 | -9% | 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 +43 percentage points is the difference between those two pass rates over the 23 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.