Can AI build a supplier risk register?
Yes, but it still needs field-level verification. In PAIR-20's July 2026 T04 run, most platforms flagged hidden source-status risks, while Copilot either buried Weiss or fabricated a Dravec expiry.
Yes, AI can draft a supplier risk register. The July 2026 PAIR-20 T04 task asked platforms to build one from a synthetic ERP spend extract and supplier master: nine suppliers, total category spend, source status, audit recency, contract timing, and backup-qualification context.
That is a reasonable AI use case. The output shape is clear: supplier, commodity, spend, verified source status, risk scores, rank, and recommended action. A structured brief also makes the register easier to audit because it forces the model to show the dimensions it used.
The July run still shows why the draft cannot be treated as finished. ChatGPT, Claude, and Gemini reached past a spend-only view and flagged hidden vulnerabilities. Claude explicitly called out Weiss Logistics' lapsed backup. Copilot broke in two different ways: on the generic prompt it ranked Weiss low and missed the lapsed-source issue; on the structured brief it invented an expired backup issue for Dravec that the file did not support.
A register is only as good as the verified field. Spend concentration is visible and tempting. Source status is where the task lives. If the ERP file says a backup exists but the supplier master says the backup has lapsed, the verified status should override the label.
The summary can preserve the error. In the PAIR run, the invented Dravec issue survived into the VP-ready summary. That matters because senior readers usually do not re-open the source files. They act on the compressed version.
Use AI for structure, then check the risk-bearing fields. The practical workflow is to let AI build the first register, then manually verify source status, backup expiry, contract proximity, audit recency, and any field that changes the rank. Do not only review the top line. In T04, the supplier that mattered was easy to bury if the model trusted spend more than source status.
Where this comes from
- PAIR-20 July 2026 run of record, task T04 — single dated runs, screenshots on file
- Procurement AI Reality Check, July 2026
Last checked Sat Jul 04 2026 00:00:00 GMT+0000 (Coordinated Universal Time). Evidence comes from dated, single-run platform sessions with screenshots on file — read each finding as “this happened,” not “this always happens.”
Work this yourself — from the course
Related questions
See what the platforms caught — and missed
Twenty procurement tasks, four AI platforms, real dated runs. Lesson 2 is free to read, no account needed.