Weekly Eval Report Template

Communicate evaluation results to stakeholders clearly and consistently. Fill in this template each week.

Template For: PM, Ops, Leadership Est. time: 15 min / week

📋 Report Header

Report Period Week of [DATE] – [DATE]
System / Product [Your AI system name]
Report Owner [Name, Role]
Overall Status 🟢 Green / 🟡 Yellow / 🔴 Red
Model Version [e.g., v2.3.1 + gpt-4o-mini]

📊 Key Metrics

Metric Target This Week Last Week Trend Status
CW Accuracy ≥ 92% 93.2% 91.8% 🟢
Faithfulness ≥ 4.0/5 4.2 4.1 🟢
Hallucination Rate ≤ 3% 2.1% 2.4% 🟢
Human Escalation Rate ≤ 12% 14.3% 11.9% 🟡
P95 Latency ≤ 3s 2.4s 2.3s 🟢

Highlights

  • CW accuracy improved 1.4% after prompt refinement for product category queries.
  • New regression tests added for 8 previously failing edge cases.
  • [Add your highlights here]

⚠️ Risks & Issues

Issue Severity Impact Action Owner ETA
Escalation rate spiked for billing queries 🟡 Med 2.4% increase in human review volume Investigate billing doc retrieval pipeline @eng-lead EOW
[Add risk]

📡 Drift Indicators

Signal Threshold Current Status Notes
Query distribution shift KL-div < 0.15 0.08 🟢 Stable
Embedding similarity > 0.85 cosine 0.91 🟢 No drift detected
New topic clusters < 5% new 3.2% 🟢 Minor uptick in pricing questions

📌 Action Items for Next Week