Renewal risk on sample data

Scenario

You are the operations lead at a fictional company with 120 accounts in a sample CRM. Leadership asks: which of our top 20 accounts are at renewal risk next quarter, and why? Everything here runs on the provided sample dataset — nothing of yours is touched.

Image: the sample dataset overview: accounts, orders, interactions.

Definition of done

• One forecast ranking the top 20 accounts by renewal risk.
• Every ranked account traceable to evidence you can read.
• One recorded correction that visibly improves the rerun.

Steps

1. Connect the sample source. Install it from the catalog like any source, but note the difference: sample data is fixed and documented, so surprises are lessons, not incidents.
2. Describe the decision. "Rank our top 20 accounts by renewal risk next quarter, with reasons." Read the proposal as a scope check: target, time range, audience.

Image: proposal with target, time range, and audience highlighted as the three scope answers.

3. Shape the slice. Define the account and order entity types, the places-relationship, and the identifiers — then map, resolve identity, and materialize. Two accounts in the sample are near-duplicates on purpose; catch them here.

Image: the duplicate pair and the identity rule that merges them.

4. Dry-run skeptically. Check the drivers against the sample notes: support tickets and late orders should dominate, exactly as the data was built to show.
5. Commit and defend. Pick the riskiest account and walk its evidence chain end to end, then record one correction the preview missed.

Image: evidence chain expanded from the riskiest account down to tickets, orders, and the runs that built each link.

6. Rerun and compare. Confirm the correction moved the result. That visible change is the loop working — the skill every tutorial teaches.

Image: before/after ranking with the corrected account moved and the reason attached.

What good looks like

• You predicted the known-risk accounts the sample was built around.
• You can explain any ranking from evidence alone.
• The rerun visibly accounts for your correction.

Next

• Do it with your data: Getting started.
• Definitions behind every step: Reference.
• Day-to-day depth: Operators.