Data Analytics report
8月28日转人工率偏高诊断
UMall AI 客服截至 11:37 的转人工率诊断。
核心结论
偏高来自重复会话,不是更多客户需要人工。 截至 11:37,28日会话口径为 12/17(70.6%Source: UMall production conversation sessions),27日同期为 8/14(57.1%Source: UMall production conversation sessions);客户口径则同为 7/12(58.3%Source: UMall production conversation sessions)。重复转人工从 1Source: UMall production conversation sessions 次增至 5Source: UMall production conversation sessions 次,恰好解释新增的 4Source: UMall production conversation sessions 次。
3Source: UMall production conversation sessions 个客户贡献 8/12 次转人工;单个取消订单产生 4Source: UMall production conversation sessions 次。另有 afterSaleInfo 白名单拦截 11/17 次、图片引用失败 2Source: UMall production conversation sessions 次。建议复用同一客户/订单的人工工单并修复附件引用;当前样本仅 17Source: UMall production conversation sessions 个会话,全天结束后再复核。
Scans production UMall session JSONL records, applies the same 11:37 JST cutoff to Aug 27 and Aug 28, detects transfer markers, and aggregates only de-identified counts.
Scans production UMall session JSONL records, applies the same 11:37 JST cutoff to Aug 27 and Aug 28, detects transfer markers, and aggregates only de-identified counts.
| Date | Rate | Denominator |
|---|---|---|
| Aug 27 | 0.57 | Sessions |
| Aug 27 | 0.58 | Unique customers |
| Aug 28 | 0.71 | Sessions |
| Aug 28 | 0.58 | Unique customers |
Sources
- UMall production conversation sessions
Scans production UMall session JSONL records, applies the same 11:37 JST cutoff to Aug 27 and Aug 28, detects transfer markers, and aggregates only de-identified counts.
- CustomerAi after-sale status checks
Fresh read-only checks of after-sale status for the orders represented in Aug 28 transferred sessions.
SQL query
python3 scripts/customer_ai_query.py afterSaleInfo --order-id <order-id>