Reporting populations, denominators, and cohort definitions

“What is our conversion rate?” is unfinished until the team can say conversion of which records, entering when, to which event, and after what exclusions.

Write the population before the metric

Population: people with a first demo request in the period
Entry: form_submitted_at between period start and end
Exclude: test records, duplicates, consent-blocked records
Unit: one person per first request
Follow-up window: 30 calendar days

This definition prevents a later report filter from changing the question silently.

Denominator checks

Publish the denominator with the metric. A percentage without it is decoration.

Cohorts prevent moving goalposts

For a cohort report, freeze membership at the entry event and measure later outcomes. Do not remove a record because its current status changed. Report the outcome and the reason for missingness separately.

Platform implementation

Build the population as a saved report or view with a named owner and a test record set. Salesforce report types and HubSpot report filters can express many population definitions, though object associations and subscription features differ. Verify the current Salesforce report type documentation and HubSpot report guidance before implementation.

Define time at both ends

State the event that admits a record, the timezone, the measurement window, and the event that closes observation. Decide how to handle reopened records, late-arriving outcomes, duplicate subjects, and records that become ineligible. “Created this quarter” is not enough when the outcome can happen next quarter.

For every rate, write numerator and denominator as record-level tests. Then account for qualified, successful, unsuccessful, still open, excluded, unknown, and bad-data states. These categories should reconcile to the starting population. If they do not, the rate is not ready for a dashboard.

Validate with examples

Keep a small fixture set containing boundary dates, missing values, duplicate associations, a late outcome, and an open record. Ask the metric owner to classify each before running the report, then compare the query result. Monitor population size and exclusion reasons over time; a stable rate on a shrinking denominator can still signal a broken process.

Related: Unknown versus zero versus bad data, GTM metric dictionary template, and Dashboard design and drill-down.