A team's eval set was "everyone's responsibility." It didn't grow. It didn't get reviewed. Quality drifted.
Eval needs an owner. The right role depends on the org.
The three options
Product Manager. Owns what good means. Authors cases that reflect product intent. Reviews drift trends.
Engineering. Owns the eval infrastructure. Adds cases mined from production. Maintains the test pipeline.
QA. Owns the eval-as-test-suite framing. Ensures coverage. Authors edge cases.
Each can work; each has trade-offs.
When each wins
- PM-owned: when product clarity is the bottleneck.
- Eng-owned: when infrastructure is immature or the eval needs deep technical investment.
- QA-owned: when the team has a strong QA function and clear product specs.
For early-stage startups: PM-owned often works (PM has product clarity, engineering supports). For larger orgs: dedicated eval engineer or ML-ops role.
Reviewer ritual
The owner:
- Drives quarterly reviews.
- Surfaces eval-related concerns to leadership.
- Owns eval-set growth.
- Makes the call on threshold changes.
A real org
A team we work with:
- PM owns each feature's eval set.
- Engineering owns the infrastructure (CI, dashboard, mining tooling).
- Quarterly review attended by PM, lead eng, leadership.
The split works because each role has clear accountability.
Trade-offs
- PM-owned: product alignment, may lack technical depth.
- Eng-owned: technical depth, may lack product-quality lens.
- QA-owned: systematic, may lack product authority.
The right choice depends on the org's maturity and which role can execute.
What we won't ship
Evals as "everyone's responsibility."
Owners without authority to make decisions.
Owners without time allocated to eval work.
Skipping the org-level review of eval health.
Close
Eval ownership is org design. PM, eng, or QA. Each can work. The team picks based on capabilities and stage. Skip the ownership question and the eval atrophies.
Related reading
- What makes an eval good — what owners protect.
- Building your first eval set — what owners do.
- Eval taxonomy — surrounding context.
We build AI-enabled software and help businesses put AI to work. If you're choosing eval owners, we'd love to hear about it. Get in touch.