Compare a leading indicator with later outcomes

01

Write the hypothesis before the review

Choose a population, an early event and a later outcome. For example: among eligible inquiries arriving in one month, does a first attempt within one business day occur alongside a higher share of sales within 90 days? Define eligibility, business-day hours, the timestamp of a real attempt and the sale event. Record exclusions and their reasons.

Choose a start date and follow every eligible inquiry for the same 90-day window from arrival. An unanswered attempt remains an attempt, not a conversation. A record with no reliable attempt timestamp belongs in a separately reported unknown group, rather than automatically being classified as late.

02

Compare completed groups

In a hypothetical cohort of 100 eligible inquiries, 60 receive a prompt attempt, 30 receive a later attempt and 10 have unknown timing. After all 90-day windows close, 18 of the prompt group, 6 of the later group and 2 of the unknown group become customers. Their observed rates are 30%, 20% and 20%, respectively. The overall conversion rate is 26 divided by 100, or 26%.

The ten percentage point difference between the first two groups is descriptive. Prompt inquiries may have had different sources, needs or staffing coverage. The group counts are small: one additional sale would change the later group's rate by about 3.3 percentage points. Keep the unknown group visible and investigate its missing timestamps.

03

Check whether the comparison is fair enough to inform action

Compare the mix of source, inquiry type, qualification and arrival time across groups. Check whether records were merged, lost or counted twice. If a team gives urgent or higher-value inquiries faster attention, urgency may affect both timing and sales. Recording that difference does not by itself remove the bias.

Review multiple closed cohorts using unchanged definitions. A seasonal shift or new lead source can change the mix. Where feasible and appropriate, a controlled comparison designed before the change can offer stronger evidence, but it still needs sufficient data and care for customers' contact preferences.

04

Decide what to do with the signal

Set an owner and review date for the process measure. If faster attempts are operationally useful, check contact quality, complaints and opt-outs as well as the outcome. More activity can create noise if the team repeats unnecessary messages.

If the relationship is inconsistent, inspect the data and the proposed mechanism before abandoning or promoting the indicator. Record what was learned, the comparison's limitations and the next question. A correlation in one cohort is a reason to investigate, not a guarantee that changing response time will produce a particular sales result.