DORA metrics, honestly
The four metrics that tell you whether your delivery is actually getting better, plus what the 2025 research changed, and the ways teams quietly game them.
Google Cloud DORA research
There are only two things you care about with software delivery: how fast you ship, and how safely. DORA splits each into two measurable pieces. The genuinely important finding underneath a decade of research: speed and stability are not a trade-off. The best teams are better at both, at once.
Deployment frequency
How often you ship to production. The best teams deploy on demand; struggling ones ship every few weeks in a tense, ceremonial release.
Lead time for changes
How long from a commit to that commit running in production. It measures the friction of your pipeline, not how fast people type.
Change failure rate
What share of deployments cause a problem that needs a fix. Not bugs in general, deployments that break things.
Failed deployment recovery
When a deployment does break something, how fast you are back to healthy. (Renamed from MTTR to be more precise.)
Where teams land
The classic four bands, as a directional reference, not a target to force your team to hit regardless of context. A stable, mature product has different healthy numbers than one racing to product-market fit.
| Metric | Elite | High | Medium | Low |
|---|---|---|---|---|
| Deployment frequency | On demand | Daily–weekly | Weekly–monthly | Monthly or less |
| Lead time for changes | Under a day | A day to a week | A week to a month | Over a month |
| Change failure rate | ≈ 5% | ≈ 10% | ≈ 15% | Higher |
| Failed deployment recovery | Under an hour | Under a day | A day to a week | Over a week |
D-SIGN.IN · DORA four keys
The 2025 report dropped the Elite/High/Medium/Low tiers in favour of team archetypes, a single ranking flattened teams that were strong in different ways. It also added a fifth metric, rework rate: how much of your work is unplanned fixing of things you thought were done. And the headline finding: heavier AI adoption correlated with more delivery instability, not less. Faster code generation with the same review discipline just means you ship your mistakes faster.
- Gaming deploy frequency, one deploy becomes five, split artificially. The number rises; nothing improved.
- Hiding failures, if failures are punished, they stop being reported. A suspiciously perfect change-failure rate is a culture problem, not a triumph.
- Watching the number, not the trend, moving from monthly to weekly deploys is winning, even if “weekly” isn’t “elite.”
Based on the DORA (DevOps Research and Assessment) program, now published by Google Cloud. Metrics and findings are freely referenced; benchmark figures are directional and drawn from published reports.
Want the version tuned to your actual situation?
Book a call