Trends, segments, and sample size

A rate can move because the process changed, the population changed, or the sample is small. Your report should help a reader tell those apart.

Show the count with the rate

Accepted handoff rate: 7 / 9 (78%)
Previous period: 18 / 31 (58%)
Open records: 4
Population definition: unchanged

The count exposes the size of the comparison. The open count exposes that the current period is not finished.

Segment only for a decision

Choose a segment because an owner can act on it: team, workflow version, source, region, or customer tier. Avoid slicing until a dramatic difference appears. That creates a story with no prior question.

Segment checklist

• same unit and denominator in each segment;
• same entry and outcome window;
• enough records to inspect the underlying rows;
• missing and suppressed records shown separately;
• segment ownership named; and
• a next action if the difference persists.

If two segments use different process definitions, label the comparison as descriptive. Do not present it as a fair performance comparison.

Trend review path

1. Confirm the metric version and population.
2. Compare counts before percentages.
3. Check open, unknown, and bad-data states.
4. Inspect the record-level rows behind the change.
5. Decide whether to monitor, investigate, or change the system.

Compare like with like

Use the same population, definition version, measurement window, and maturity across periods. A recent cohort may look worse simply because outcomes are still open. Mark policy, routing, product, territory, and source-system changes on the trend so readers do not interpret a measurement break as business movement.

Show the numerator and denominator with every rate. Prefer a longer period or pooled view when volume is low, and avoid ranking small groups whose difference may be one record. Use uncertainty ranges where the audience can interpret them; otherwise state plainly that the sample is too small for a conclusion.

Investigate segments without manufacturing a story

Choose segments tied to a decision, not every available dimension. Check unknown segment values and minimum counts before comparing performance. Treat a surprising slice as a question, then inspect records and test it in another period. If the segment was discovered after exploring many cuts, label the finding exploratory until it repeats.

Related: Reporting populations and denominators, Unknown versus zero versus bad data, and GTM operating review template.