Semogram Docs
Tutorials

Renewal-risk exercise

Use your own account history to build a probability forecaster, preserve evidence cutoffs and evaluate renewal outcomes without leaking future information.

Evaluate renewal risk on a small test dataset you provide, with account IDs, renewal dates, support history and historical outcomes. Keep the data available at each prediction cutoff separate from the outcomes used to score it.

  1. Connect or upload the dataset and verify ingestion.
  2. Define Account and relevant relationships, then configure bindings or mapping/materialization. Verify identity on any known duplicate records.
  3. Author, validate, and publish an evidence query for one account at a cutoff. Keep post-cutoff outcome data out of forecast inputs.
  4. Create a probability forecaster using that query release, clear renewal-risk event definition, input/output schema, prompt, model, and evaluation policy.
  5. Run for named subjects and explicit evaluation dates. Inspect evidence, warnings, confidence, and probability on each result.
  6. Record outcomes when the window permits them and review the evaluation report.
  7. Change a query, prompt, or data defect deliberately, version it, and compare a new execution. Do not expect a review record alone to change predictions.

Success means traceable inputs and measured outcomes, not a guaranteed account ranking. See Get your first forecast.