A baseline is not simply last week's dashboard. It is a documented view of normal behaviour before exposure to a change. Good baselines account for seasonality, campaign activity, operational incidents and shifts in customer mix.
Start with the outcome the release is expected to influence. Record its current level, variation over time and differences between important segments. Add guardrail measures that could reveal an unintended cost, such as slower task completion or increased support contact.
Mark the rollout plan clearly. Staged exposure can create a natural comparison group, but only when assignment and eligibility are understood. If a clean comparison is not possible, document the limitation before results arrive.
The goal is decision confidence, not statistical theatre. A concise baseline agreed by product, engineering and commercial owners is more valuable than a complex model nobody trusts.