📊 AI engineering productivity scorecard

Example data — four fictional teams over two quarters, so you can see how the scorecard works. Replace them with your own numbers, or paste a CSV below.

Cycle time: commit → production if you have it (that is what DORA’s lead-time bands describe). Review time: PR opened → approved. Percentages are 0–100.

How the bands and diagnosis work

The band edges below follow the ranges DORA publishes for its performance clusters; the sourced DORA table appears here once it is published.

• Cycle time bands: Elite (< 1 day) ≤ 24 h · High (1 day – 1 week) ≤ 168 h · Medium (1 week – 1 month) ≤ 720 h · Low (1 – 6 months) ≤ 4,380 h.
• Deploy frequency bands: Elite (on demand, daily or more) · High (daily – weekly) · Medium (weekly – monthly) · Low (monthly – every 6 months).
• Change-failure bands: Elite level (≈ 5%) ≤ 5% · High / medium range (10–20%) ≤ 20% · Low level (≈ 40%) ≤ 40%.
• Band edges follow the ranges in DORA's 2024 cluster table (see the DORA items). DORA's clusters are survey results, not standards; the band edges here are our simplification for comparison.

Diagnosis rules illustrative assumption — use your own numbers

• Quality gates: change-failure rate up ≥ 2 pts while AI-authored share up ≥ 5 pts.
• Review is the bottleneck: adoption ≥ 50%, PRs per engineer up ≤ 5%, and review time ≥ 24 h or up ≥ 20%.
• Delivery foundations: adoption ≥ 30% but cycle time ≥ 168 h or fewer than 1 deploy per week.
• Adoption: adoption below 30%.
• Gains showing: PRs per engineer up ≥ 10% without a change-failure rise.

The diagnosis thresholds are illustrative assumptions, not published benchmarks — adjust them to your organization. With only one period, the scorecard can flag “watch” items but not trends. Use it at team level only — never to rate individuals.