AI customer support test delivery
Buyer: Example Buyer · Supplier: Example Studio
Buyer accepted · Agreement v1 · Submission v2
Delivered product: customer support agent v1 in the agreed test environment.
The buyer confirmed that Submission v2 meets all four agreed criteria.
The record ties the buyer’s decision to Agreement v1, Submission v2 and the evidence versions above. The parties retain the originals separately.
View standards, materials and review history
| Agreed criterion | Supplier evidence | Buyer review |
|---|---|---|
| Answer quality Test the 40 agreed questions against the agreed rubric: at least 36 must pass, and no answer in this test may invent policies or commitments. | Answer test report · r1 Supplier reports: 38/40 questions pass; no invented policies or commitments. | Answers accepted Buyer reviewed Answer test report r1 against the rubric. |
| Human handoff Following the agreed test steps, all 10 agreed handoff cases must reach the designated human support queue. | Escalation report · r2 Supplier reports: 10/10 cases covered; r1 covered 8/10. | Handoff accepted after correction Buyer reviewed Escalation report r2, including cases 9 and 10, and confirmed that all 10 handoff results meet the agreed criterion. |
| Pilot operation Under the agreed test environment, load and timing method, run for 30 minutes with five concurrent conversations; every test reply must complete within 10 seconds. | Pilot run report · r1 Supplier reports: the 30-minute run meets the concurrency and response-time limits. | Pilot operation accepted Buyer checked the timestamps and response logs in Pilot run report r1. |
| Operating guide With the required access in the agreed test environment, the designated buyer user can start and stop the agent by following the guide. | Operating guide · r1 Supplier provides operating and stop instructions. | Operating guide accepted The designated buyer user followed Operating guide r1 to start and stop the agent and confirmed access to the originals. |
Agreement v1 confirmed → Submission v1 covers 8/10 cases → buyer requests cases 9 and 10 → Submission v2 covers 10/10 → buyer accepts.