SLA metrics that matter for a QA testing vendor

Updated 2026-09-24

Useful software testing SLA metrics measure things the vendor controls: regression turnaround after a release candidate, time to report a blocking defect, report delivery, staffing continuity and the stability of the automated suite. Metrics that depend mostly on your developers or your product, such as total defects found or defects in production, are better used as shared indicators than as service levels with credits. Every service level needs a definition, a target, a data source, a measurement period and a consequence when it is missed.

What to check

  1. For each proposed service level, write down who controls the outcome. Keep as service levels only those the vendor controls.
  2. Check that each service level has a definition, target, data source, measurement period and reporting owner.
  3. Check that the data comes from your systems, such as your CI pipeline or issue tracker, not only the vendor's reports.
  4. Check that missed service levels carry a service credit or other remedy, and that repeated misses lead to a termination right.
  5. Remove words such as timely, reasonable, prompt and as soon as possible from any service level text.

Ask the vendor

  • Which measurable outcomes would you agree to be held to, how are they measured, and what happens when one is missed?
  • How exactly do you define and count your defect metrics, such as severity, defect leakage and reopen rate?
  • What service credits apply when an agreed service level is missed, and how are they calculated and applied?
  • Which of the service levels you propose depend on things we provide, such as environments or builds, and how do you want that recorded?

Start from what the vendor controls

A service level only works if the vendor can meet it through its own effort. US federal rules on performance-based acquisition ask buyers to describe the required results, set measurable performance standards, and state how performance will be assessed. That discipline is useful for any buyer: if you cannot say how a number will be measured, or the vendor cannot move it, it is not a service level.

EU financial entities under the Digital Operational Resilience Act (DORA) have a specific duty here. Contracts for ICT services that support critical or important functions must include full service level descriptions with precise quantitative and qualitative performance targets. Even if DORA does not apply to you, it is a reasonable standard to aim for.

Metrics the vendor can control

These are good candidates for service levels with consequences. Use the contract clause builder to turn them into a schedule.

Service level Definition to agree Data source
Regression turnaround Time from release candidate available to regression results reported CI pipeline timestamps, report timestamp
Blocking defect notice Time from finding a blocking defect to notifying your named contact Issue tracker created time, chat or email record
Report delivery Status and completion reports delivered on the agreed day with agreed contents Report dates, checklist of required sections
Defect report quality Share of defects returned by your developers as not reproducible or incomplete Issue tracker status history
Staffing continuity Unplanned changes to named team members per period Team register, timesheets
Suite stability Share of automated test failures caused by the tests rather than the product, and time to fix a flaky test CI results, flaky test log
Response to requests Time to acknowledge and estimate a change request Change request log

Fill in targets with your own numbers in brackets during negotiation. The targets depend on your release cadence, team size and the model you are buying, so do not copy them from another contract.

Metrics that look good on paper

Some measures appear often in proposals but make poor service levels.

Number of test cases executed or defects found. These reward volume. A vendor can raise both without reducing your risk.

Automation coverage percentage. Meaningless without a stated base: which features, which test levels, which environments. Even with a base, a high percentage of brittle tests is worse than a smaller stable suite.

Defect leakage or escaped defects. Useful to track, but the result depends on what your team built, what was in scope, and what you chose to release with known issues. Credits attached to it lead to arguments about causes.

Pass rate. A high pass rate can mean a good product or weak tests. It says nothing about the vendor's performance on its own.

Delivery performance metrics. The DORA research programme's measures of software delivery performance, such as change failure rate, describe your whole delivery system. They are good shared indicators for a service review, not measures of the testing vendor alone.

What each service level needs

Every service level in the contract should state:

  1. what is measured, in one sentence;
  2. the target;
  3. the data source, preferably your own systems;
  4. the measurement period, such as monthly;
  5. who reports it, and when;
  6. exclusions, such as delays caused by your environments being unavailable;
  7. the consequence when it is missed.

Weak and stronger wording side by side:

Weak: "Supplier will provide timely regression testing for each release."

Stronger: "Supplier will report regression results for [the agreed suite] within [X] working hours of a release candidate being available in the staging environment, measured from the pipeline timestamp. Target: met for [Y] of releases in each calendar month. Releases where the staging environment was unavailable for more than [Z] hours are excluded."

Remedies that give service levels weight

Without a consequence, a missed service level becomes a discussion. The UK Model Services Contract ties service levels to service credits and to a termination right for serious or repeated failure. For a testing contract, agree:

  • a service credit per missed service level, deducted from the next invoice;
  • a statement that credits are a price adjustment and do not limit other rights;
  • a remediation plan after a missed period;
  • a right to terminate for cause after repeated misses in a set window.

Make the data yours

Measure from your systems where you can: your pipeline, your issue tracker, your repository. Ask for read access to the vendor's test management data at all times. See how to read a vendor's sample report for what the reports behind the numbers should contain, and put your service levels into the RFP so every vendor prices against the same targets. The staffing continuity measure links directly to how QA vendors staff projects.

Tools for this

Questions buyers ask

Should defect leakage be a service level?
Usually not with credits attached. Defects that reach production depend on what was built, what was in scope and what your team chose to release. Track it together at service reviews and use it to adjust scope and risk, rather than to penalise the vendor.
How many service levels should a testing contract have?
Few enough that both sides actually measure them every period. A short list of measurable service levels with consequences is more useful than a long list nobody reports on.
Do service credits limit my other rights?
They can, if the contract says credits are the sole remedy. Ask for wording that says credits are a price adjustment and do not limit your other rights, including termination for repeated failure.

Sources

  1. FAR Subpart 37.6: Performance-Based Acquisition, US Federal Acquisition Regulation
  2. Regulation (EU) 2022/2554 (DORA), including Article 30 key contractual provisions, European Union
  3. Model Services Contract, UK Cabinet Office
  4. DORA research program: software delivery performance metrics, DORA
  5. Flaky Tests at Google and How We Mitigate Them, Google Testing Blog