Experimentation maturity model
Experimentation Maturity Model: 5 Stages From Ad Hoc Tests to Compounding Learning
Maturity is not the number of tests a team runs. It is the ability to make a reliable decision today and reuse the learning tomorrow.
By Atticus Li · Updated July 28, 2026
Use the model to choose the next constraint
The right next step for a young program is often a consistent test record, not a sophisticated dashboard. The right next step for a larger program may be evidence retrieval, not another idea generator. This model makes that choice explicit.
1. Ad hoc
Testing is reactive. Evidence, decisions, and ownership live in separate documents or in people’s memory.
Next move: Create one experiment record and a basic definition of done.
2. Repeatable
The team has a workflow, but fields and quality checks vary by analyst or project.
Next move: Standardize the hypothesis, metric hierarchy, MDE, sample plan, QA, and decision fields.
3. Managed
Program standards exist and important work can be inspected, but learning is not yet consistently reused.
Next move: Make historical tests searchable before backlog prioritization and link reports to evidence quality.
4. Compounding
Research, experiments, reporting, and planning refer to the same evidence base. Repeated work declines.
Next move: Focus on comparability, cross-team adoption, and known gaps in research or instrumentation.
5. Evidence-led
High-stakes decisions use governed, comparable evidence and explicit limits. Replication and synthesis are earned, not assumed.
Next move: Maintain the system, test transferability, and avoid letting dashboard incentives lower quality.
Maturity assessment
Assess the operating system, not the team's ambition.
Score each area from 1 to 5 based on what the team can do consistently today. This is a planning diagnostic, not a certification or benchmark.
Keep the models separate
This model measures the whole experimentation program. The test-library category is narrower: it measures whether historical evidence can be found, checked, and reused. A team can have a mature repository and weak experiment design, or a rigorous design process with poor knowledge retention.