Many companies measure external engineering teams the way they were measured in the staffing era: hours logged, tickets closed, commits per person. These numbers are easy to collect and say almost nothing about whether the team is delivering value. Worse, measuring and steering individual engineers pulls the client into a line-management role that does not belong in a result-oriented engagement. A better approach measures the team as a delivery system: the DORA software delivery metrics for throughput and stability, a small set of flow metrics such as cycle time, and acceptance data from the sprint process. This analysis explains which metrics to use, how to read them, and how to avoid the common traps.

The numbers: what the latest DORA research shows

The research behind these metrics has shifted its focus to AI, and the findings matter for anyone measuring an external team. The DORA 2025 report, based on nearly 5,000 technology professionals, found that 90% use AI at work, with a median of two hours a day, and that more than 80% believe it has increased their productivity. Yet 30% report little or no trust in AI-generated code, and the report links AI adoption to higher throughput but also to lower delivery stability. The 2024 report had estimated that rising AI adoption was associated with a 1.5% drop in throughput and a 7.2% drop in stability. Stack Overflow's 2025 Developer Survey of more than 49,000 developers confirms the gap between use and confidence: 84% use or plan to use AI tools, but 46% do not trust their accuracy. The lesson for buyers is direct: activity is rising, so only outcome metrics at team level show whether delivery is actually improving.

The DORA metrics in their current form

The DORA research programme, now run within Google Cloud, has published software delivery metrics for over a decade. The current model groups five metrics into two categories.

MetricCategoryDefinition
Change lead timeThroughputTime from a code change being committed to it running successfully in production
Deployment frequencyThroughputHow often changes are deployed to production
Failed deployment recovery timeThroughputTime to recover when a deployment causes a failure in production
Change fail rateInstabilityShare of deployments that cause a failure in production
Deployment rework rateInstabilityShare of deployments that are unplanned fixes for production issues

Two changes in recent years are worth knowing. In 2023, the older mean time to restore was redefined as failed deployment recovery time, limited to failures caused by changes. In 2024, deployment rework rate was added as a fifth metric. If a provider reports the classic four keys only, that is not wrong, but ask how they define each one.

Why DORA metrics suit external teams

DORA metrics describe the delivery system, not the people in it. That property is exactly what an external engagement needs. They answer the question a CTO actually has: is working software reaching production at a sustainable pace, without breaking things? They are also largely derived from systems the client already owns, such as the Git platform, the CI/CD pipeline and incident tooling. This makes them hard to game and easy to verify.

There is a contractual dimension as well. In a Werkvertrag or Dienstvertrag, the provider organises its own team. If the client starts monitoring individual engineers' output and assigning work person by person, the engagement moves toward the integration and direct instruction that characterise Arbeitnehmerüberlassung under §1 Abs. 1 Satz 3 AÜG. Team-level outcome metrics fit the contractual model. Per-person activity metrics do not.

Adding flow and acceptance metrics

DORA metrics start at commit. For an external team, the time before commit and the acceptance step afterward matter as well. Three additions cover the gap:

  • Cycle time: time from work starting on an item to that item being accepted. This captures waiting for decisions, reviews and acceptance, which are often on the client's side.
  • Review wait time: time from pull request opened to first review. In distributed teams, this is often the single largest delay.
  • Sprint acceptance rate: share of committed sprint items accepted at the end of the sprint without blocking defects. This connects metrics to the contractual acceptance process.

Add one quality indicator from after acceptance: defects found in production or in later sprints that are traced back to an accepted increment. This is the honest counterweight to a high acceptance rate.

How to read the numbers

Metrics are most useful as trends over several sprints, not as snapshots. A practical reading pattern:

  1. Look at throughput and instability together. Rising deployment frequency with a rising change fail rate is not progress.
  2. Decompose cycle time. Split it into active work, review wait, client decision wait and acceptance wait. In many engagements, a significant share of cycle time sits with the client, and that is where the improvement lies.
  3. Compare against your own baseline. Published benchmark clusters are useful context, but the team's own trend over time is the more meaningful reference.
  4. Discuss anomalies, do not punish them. A spike in recovery time after a complex migration is information, not misconduct.

What not to measure

  • Hours as a performance indicator. Hours are a cost input, not an output.
  • Lines of code or commits per person. Both reward the wrong behaviour.
  • Story points as productivity. Points are a planning tool within a team. Comparing them across teams or over long periods invites inflation.
  • Individual rankings. They damage collaboration within the team and, in an external engagement, push the client into a supervisory role.

A monthly delivery review

Bring the metrics together in a short monthly review with the provider's project lead and your engineering lead. A simple agenda:

  1. DORA metrics and flow metrics, trend over the last three months.
  2. Cycle time breakdown: where is waiting time concentrated, and who owns it?
  3. Sprint acceptance rate and post-acceptance defects.
  4. One or two improvement actions, each with an owner on the responsible side.

Keep the dashboard in a tool both sides can see, fed from the client-owned systems. When both parties look at the same data, the conversation shifts from defending numbers to improving the system.

Practical setup

Most of these metrics can be derived from the Git platform, CI/CD logs and the issue tracker with modest effort, and several tools offer them out of the box. Agree on definitions in writing before you start: what counts as a deployment, what counts as a failure, when an item starts and when it is accepted. Without shared definitions, metrics create disputes instead of resolving them.

Whichever provider you use, the approach is the same: measure the team as a system, read trends rather than snapshots, and let contractual roles determine who acts on what.

The remarks on contract types and the AÜG are general information, not legal advice.

Team-level delivery metrics turn a vendor relationship into an evidence-based one. In niogod™ pods, the Projektleiter prepares a monthly delivery review from your own systems' data, and improvement actions are tracked as backlog items like any other work. If you want to measure an external team as a delivery system rather than in hours, it is a conversation worth having early.

Talk to niogod™

Planning engineering capacity in DACH? Book a 30-minute consultation with the niogod™ team.

Book consultation
← All insights