97 Suppliers Automated Health Check: Real Business Data Driven

¥485,504, Not Demo Data

Inventory Manager Employee inspects the real ERP stock table (stk_stock_setup) and returns: 4 stock items, 4 items below the waterline (50), estimated replenishment amount ¥485,504. The data source is the real ERP, not fake data stuffed into a demo database.

Supplier Manager Employee conducts a full performance review for 97 suppliers: 7 low-scoring, 89 with missing data, 1 normal. HTML report auto-generated. For the 7 low-scoring suppliers, collaboration tasks are automatically created and assigned to procurement employees for re-quotation.

How does this chain run?

1. Inventory Manager queries stk_stock_setup (read-only inspection, no write).

2. Identifies shortage items → Human-in-the-loop confirmation (paused via requestConfirmation).

3. After confirmation, generates a PR (writes Part object) + assigns collaboration task to procurement employee (taskQueue).

4. Supplier Manager simultaneously performs health checks on 97 suppliers; low scores automatically trigger re-quotation.

Key Design: Strict separation of read-only queries and write actions (readOnly does not trigger approval); write actions must be confirmed via human-in-the-loop.

Another real number: The '89 with missing data' behind the 97 suppliers

The most glaring issue in the health check report is the 89 suppliers with missing data—incomplete supplier master data prevents performance calculation. This number is not polished; it goes directly into the report. One value of AI employees is to lay bare these 'usually ignored' accounts.

Boundaries, stated as-is

• The PR real write (writing Part) and procurement task assignment paths in the resume phase; this test only covered Phase1 read-only inspection; the write path awaits environment testing with a confirmation channel.
style="padding-left:20px;margin:12px 0;">
• Supplier diagnose (LLM deep diagnosis) is disabled by default—diagnosing all 97 would be 40+ sequential calls, which would bog down the process. It is recommended to normalize to 'diagnose only the low-scoring/missing subset.'

BossAgents: Digital employees drive real business objects, not demo data.

All data in this article can be traced in the BossAgents technical documentation repository · BossAgents · Private Intelligent Agents for the Boss.