Report Generator

Overview

This skill produces structured sales and performance reports for store managers. Reports are data-first, anomaly-aware, and always include a recommended action.

Depends on: POS/ERP data connection configured in report_config (Step 05). If no data source is connected: return empty report with setup guidance.

Report Types

| Type | Trigger | Covers | Default Schedule | |------|---------|--------|-----------------| | Daily | "今天怎么样" / 18:00 cron | Current day vs. yesterday | Daily 18:00 | | Weekly | "本周报告" / Monday 09:00 | Mon–Sun vs. prior week | Monday 09:00 | | Monthly | "月报" / 1st of month | Full month vs. prior month | 1st @ 09:00 | | On-demand | Any data question | Specified time range | — |

Report Structure

Every report follows this 5-section format:

Section 1: Headline Numbers

The top 3 KPIs the manager cares about most:

Total revenue (实际 vs 目标, percentage completion)
Transaction count
Average transaction value

Section 2: Product Performance

Top 5 best-selling SKUs (by revenue)
Any items with zero sales (flag if > 3 days)
Low stock warnings (pull from inventory data if available)

Section 3: Anomalies & Alerts

Auto-detect and flag:

Revenue > 20% above/below yesterday → flag
A category drops > 30% vs. prior period → flag
Escalation rate spike → flag
Unusual peak/valley in hourly distribution → note

Section 4: Staff Performance (if data available)

Transactions per staff member
Average ticket per staff
(Never names individuals negatively in customer-facing channels)

Section 5: Recommended Actions

1–3 specific, actionable items the manager should do today/this week. Examples: "补货SKU001（库存仅剩3件）", "关注女装区下午时段表现"

Data Source Integration

Pull data from configured sources in report_config:

| Source | Data Retrieved | |--------|---------------| | POS API | Transactions, revenue, returns by time period | | ERP API | Inventory levels, restock status | | Manual input | If no API: manager pastes today's numbers |

If partial data: generate report with available data, mark missing sections clearly.

Reference: data-connectors.md

Delivery

After generating, deliver via configured channel:

report_config.delivery_channel (wecom / telegram / etc.)
report_config.recipients (manager IDs)

Format for channel:

WeCom: Use card format with emoji headers for scannability
Text channels: Markdown tables

On-Demand Queries

For ad-hoc questions like "上周卖得最好的是什么？":

Identify time range and metric from the question
Query data source
Use scripts/generate_report.py --period custom --start X --end Y --metric top_products
Present inline (no delivery to channel)

Reference: metric-definitions.md — standard metric formulas.