Install any skill in seconds. Free to start, no credit card required.
Get Started Free →当用户需要对银行零售金融场景中的客户投诉进行分类、归因、责任环节识别、 问题复盘、改进建议设计、闭环跟踪与管理汇报时,使用本技能。 适用于客服分析、服务体验分析、流程改进、客诉高发问题识别、渠道体验优化、 产品服务缺陷排查、客服坐席问题复盘等任务。 输出时应严格区分“已确认事实”“高概率归因”“待核验事项”,避免在证据不足时直接下结论。
.claude/skills/aifinlab-customer-complaint-attribution-and-improvement-assistant/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-04 | ✗→✓ | ▲ Improved | 94% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 71% | 0% |
| case-06 | ✗→✓ | ▲ Improved | 68% | 0% |
| case-13 | ✗→✓ | ▲ Improved | 168% | 0% |
| case-20 | ✗→✓ | ▲ Improved | 119% | 0% |
本技能用于支持银行零售金融场景下的客户投诉分析与改进工作。它不是简单对投诉文本做情绪总结,而是围绕“投诉发生了什么、为什么发生、责任链条在哪里、对客户与银行产生了什么影响、后续如何改进和追踪”展开结构化分析。
本技能适用于以下任务:
适用行业:银行 业务条线/能力域:零售金融 场景/能力:客服分析 层级:task
常见适用场景包括但不限于:
以下场景不建议直接使用本技能给出确定性结论:
在上述情况下,本技能应输出:事实梳理、疑点提示、待核验事项、建议的进一步核查路径,而不是越权定责。
建议尽量提供以下资料,资料越完整,归因和改进建议越可靠:
适合在以下情况下使用:
推荐步骤:
适合在以下情况下使用:
推荐步骤:
可从以下维度进行分类:
建议至少从以下层面归因:
责任判断时,不宜直接写“某部门有责”,应按以下方式表达:
建议至少覆盖四类措施:
输出时建议包括以下部分:
需要更详细的分类规则时,阅读:
references/complaint_taxonomy.mdreferences/root_cause_framework.md需要更详细的改进措施模板时,阅读:
references/improvement_playbook.mdreferences/follow_up_kpi_guide.md需要固定输出格式时,阅读:
references/output_schema.mdassets/templates/complaint_analysis_report_template.mdscripts/complaint_taxonomy_classifier.py:对投诉记录进行规则化分类;scripts/root_cause_scorer.py:根据投诉特征打分并输出潜在根因;scripts/render_complaint_report.py:将分析结果渲染为 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 | 29,400 | 56,928 | +94% | 1 | 1 | 0% | 4,403 | 7,084 | +61% | 0 | 0 | — |
case-02 | fail→fail | 29,040 | 36,297 | +25% | 1 | 1 | 0% | 4,257 | 7,407 | +74% | 0 | 0 | — |
case-03 | fail→fail | 30,730 | 43,171 | +40% | 1 | 1 | 0% | 4,320 | 8,501 | +97% | 0 | 0 | — |
case-04 | fail→pass | 17,226 | 16,992 | -1% | 1 | 1 | 0% | 2,302 | 4,458 | +94% | 0 | 0 | — |
case-05 | fail→pass | 20,009 | 20,814 | +4% | 1 | 1 | 0% | 2,981 | 5,094 | +71% | 0 | 0 | — |
case-06 | fail→pass | 20,736 | 20,555 | -1% | 1 | 1 | 0% | 2,942 | 4,951 | +68% | 0 | 0 | — |
case-07 | pass→pass | 23,315 | 33,928 | +46% | 1 | 1 | 0% | 3,389 | 7,090 | +109% | 0 | 0 | — |
case-08 | fail→fail | 29,834 | 33,399 | +12% | 1 | 1 | 0% | 4,467 | 6,968 | +56% | 0 | 0 | — |
case-09 | fail→fail | 17,615 | 20,502 | +16% | 1 | 1 | 0% | 2,537 | 5,129 | +102% | 0 | 0 | — |
case-10 | pass→pass | 19,609 | 23,805 | +21% | 1 | 1 | 0% | 2,694 | 5,395 | +100% | 0 | 0 | — |
case-11 | pass→pass | 20,821 | 33,166 | +59% | 1 | 1 | 0% | 3,306 | 7,154 | +116% | 0 | 0 | — |
case-12 | pass→pass | 20,249 | 32,897 | +62% | 1 | 1 | 0% | 2,687 | 6,959 | +159% | 0 | 0 | — |
case-13 | fail→pass | 17,460 | 29,942 | +71% | 1 | 1 | 0% | 2,428 | 6,504 | +168% | 0 | 0 | — |
case-20 | fail→pass | 21,985 | 31,888 | +45% | 1 | 1 | 0% | 2,961 | 6,471 | +119% | 0 | 0 | — |
case-21 | fail→fail | 22,961 | 27,799 | +21% | 1 | 1 | 0% | 3,157 | 6,041 | +91% | 0 | 0 | — |
case-22 | pass→pass | 21,950 | 32,448 | +48% | 1 | 1 | 0% | 2,982 | 6,461 | +117% | 0 | 0 | — |
case-14 | pass→pass | 18,239 | 20,298 | +11% | 1 | 1 | 0% | 2,736 | 5,060 | +85% | 0 | 0 | — |
case-15 | pass→pass | 17,589 | 24,597 | +40% | 1 | 1 | 0% | 2,498 | 5,709 | +129% | 0 | 0 | — |
case-16 | pass→pass | 21,549 | 26,951 | +25% | 1 | 1 | 0% | 3,098 | 5,672 | +83% | 0 | 0 | — |
case-17 | fail→pass | 23,146 | 32,517 | +40% | 1 | 1 | 0% | 3,225 | 6,746 | +109% | 0 | 0 | — |
case-18 | pass→pass | 25,129 | 35,989 | +43% | 1 | 1 | 0% | 3,405 | 7,401 | +117% | 0 | 0 | — |
case-19 | fail→pass | 20,360 | 35,115 | +72% | 1 | 1 | 0% | 2,992 | 6,934 | +132% | 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 +32 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.