Evaluate policy changesbefore live rollout

Use logs from your current production policy to estimate whether a proposed policy is worth a live test.

Best fit: repeated decisions with clear actions and outcomes that can be linked back later.

What CounterFact returns

Each run combines a policy estimate with an evidence verdict and a recommended action.

The proposed policy can start as a scored column, a fixed action, or a plain-English rule draft reviewed before evaluation.

  1. Outcome estimate

    Estimate the target outcome under the proposed policy and compare it with the logged baseline.

  2. Evidence verdict

    Check whether the logs support the estimate, support only a directional read, or need better instrumentation.

  3. Recommended action

    Plan a monitored test, use the result for prioritization, improve logging, or reframe the policy question.

Good estimates need the right logs

Historical logs can be incomplete. They may miss which actions were available, when a policy changed, or whether enough time has passed to measure the outcome. CounterFact checks these issues before reporting an estimate.

If the logs are not ready, CounterFact shows what to capture next so a future evaluation can be stronger.

When CounterFact is a good fit

CounterFact works best when the same decision repeats many times, the action set is small, and the outcome can be linked back to the original decision.

Common examples include renewal offers, save treatments, routing choices, onboarding prompts, and intervention policies.

See what your policy logs can support.