DORA Metrics: the complete guide to the 4 DevOps performance indicators

DORA Metrics are four indicators drawn from research conducted by Nicole Forsgren, Jez Humble and Gene Kim within the DevOps Research and Assessment program, published notably in their book Accelerate (2018) and in the annual State of DevOps Report. Unlike internal metrics specific to each team, they provide a standardized basis for comparison across organizations.

The four metrics, in detail

Metric	Definition
Deployment frequency	How often the organization deploys code to production over a given period
Lead time for changes	Time elapsed between a code commit and its deployment to production
Change failure rate	Proportion of deployments that cause an incident or require an immediate fix
Mean time to restore (MTTR)	Time needed to restore service after a production incident

Reference performance levels

Level	Deployment frequency	MTTR
Low	Less than once a month	More than a week
Medium	Between once a month and once a week	Less than a day
High	Between once a week and several times a day	Less than an hour
Elite	Multiple deployments per day	Less than an hour

How to measure these metrics in practice

Instrument the deployment pipeline to automatically timestamp every release.
Link each production incident to the deployment that potentially caused it.
Calculate these four metrics over a rolling period (for example the last 90 days) rather than a single point in time.
Share these metrics with the team as a steering tool, not as an individual evaluation.

What these metrics do not measure

DORA Metrics measure the speed and reliability of software delivery, not directly the quality of the product or user satisfaction. An organization can show excellent DORA metrics while shipping features that are barely useful: they should be complemented by product metrics, not replaced by them.

DevOps & Operations Diagnostic·See a diagnostic preview

For an organization discovering these metrics for the first time, the priority is not reaching the elite level overnight, but measuring them once without dressing up the numbers, then tackling the actions that will improve deployment reliability the fastest.