Use the register before changing the system
Budget shifts, scoring rules, lead routing and content experiments all deserve a short decision record. This register holds the question, evidence, one intended change and the conditions for judging it. It turns “we tried that” into something the next decision-maker can inspect.
Download the blank file, import it into your spreadsheet tool and keep one row per decision. The filled example is illustrative and has no measured result. It shows how to specify a test, not how to manufacture an outcome.
Write the question precisely
Ask whether a particular change improves a defined commercial or operational decision. “Improve GTM” is too broad. “Does resolving the existing CRM owner before enrichment reduce unowned demo requests?” names a change, population and observable outcome.
Choose the segment and baseline before starting. Record exclusions, sample size, dates, currency where relevant and the outcome definition. A changed denominator can produce an apparent lift even when behavior stays the same.
Fill the decision fields
| Field | What belongs here |
|---|---|
| Hypothesis | The expected mechanism and why the change could help |
| One change | The intervention whose effect you intend to learn about |
| Primary metric | Definition, numerator and denominator, not a dashboard label |
| Comparison | Baseline, holdout or cohort, with limitations |
| Minimum maturity | Time needed for the intended outcome to become observable |
| Limit | Maximum spend, workload or exposure while testing |
| Hold condition | A specific failure requiring a pause or rollback |
| Owner and review date | Who decides, and when the evidence will be read |
Record supporting measures too, but do not quietly replace the primary metric after seeing the result. If the question changes, open a new decision or explicitly amend the record.
A filled example
- Hypothesis
- Existing-owner lookup before enrichment reduces ownership delay
- Change
- Move the CRM lookup; retain all exclusion and suppression checks
- Metric
- Share of eligible requests assigned within the agreed response target
- Hold
- Duplicate owners or a customer routed as a new prospect
- Outcome
- Not run. Counts, result and decision remain blank until observed.
Close the loop honestly
At review, attach source exports or receipts and record counts alongside percentages. Mark the result as supportive, contradictory or inconclusive. Decide whether to keep, reverse, extend or retire the change and explain why. An incomplete cohort should stay incomplete rather than being promoted into a success story.
Observational comparisons can be useful without proving causation. Note changed market conditions, sales capacity, seasonality and simultaneous interventions. A controlled comparison is stronger where feasible, but a small team can still make a well-documented bounded decision.
What returns to the brain
Only the reviewed learning changes company rules. Store the accepted rule, reason, date and version; keep failed hypotheses so they are not repeated without new evidence. The live CRM or campaign system receives the approved configuration and its rollback path.
Use the measurement contract for consistent pipeline definitions, budget moves for spend decisions and score calibration for account selection. The register is an operating aid, not a claim of guaranteed commercial improvement.
Publication and source record
Published on GTMhub: . Last reviewed: .