Sales cycle length is often reported as the number of days from a defined start event to close, but the average can hide a long tail. A median shows the middle completed deal, while the 75th and 90th percentiles expose slower experiences that affect staffing, forecasts, and follow-up. Open opportunities are censored—they have not yet revealed a final duration—and should not simply be treated as zero or discarded without explanation.
VISUAL LESSON
What you will learn
- 01Define reproducible cycle start and end events.
- 02Calculate median, P75, and P90.
- 03Use cohort and stage-dwell evidence to choose an improvement.

ILLUSTRATIVE WORKED EXAMPLE
Read an illustrative closed-won cycle distribution
PRACTICAL INTERFACE MAP
Move from timestamps to a bottleneck decision
Define lead created, qualified, opportunity created, first meeting, or another start and closed won, closed lost, or contract date as the end.
Use one time zone, exclude test and duplicate records, disclose reopen rules, and calculate comparable elapsed durations.
Segment carefully, review delay reasons, select one operational change, and compare a later matched cohort.
STEP-BY-STEP LESSON
Defined events → closed cohort → distribution → stage-delay improvement
THE LEAD ATLAS METHOD
Lead Atlas Data can research a fresh custom business-contact list for specific campaign markets, locations, and account categories when cycle analysis shows the pipeline needs a better-defined top-of-funnel cohort.See how custom list research works ↗Define the cycle precisely
Choose the operational start event, end event, date field, time zone, included opportunity types, owners, products, regions, acquisition channels, closed status, reopen rule, duplicate treatment, test-record exclusion, and measurement window. Write the business question the metric should answer.
Use event timestamps rather than current stage age when measuring completed cycles. Keep lead response time, opportunity cycle, contract cycle, and time to realized revenue as separate measures when their owners and causes differ.
Build the cohort
Select a cohort by start date or close date and say which one. A close-date cohort answers how long recently completed deals took; a start-date cohort follows opportunities that began together and may require longer observation. Preserve cohort boundaries and maturity.
Separate closed won, closed lost, and open opportunities. Open records are right-censored because their final duration is unknown; excluding them can make the process look faster, while treating them as zero is plainly wrong. Report their count and current age.
Calculate the distribution
For each completed record, subtract the start timestamp from the end timestamp using the chosen day convention. Sort durations, calculate count, average, median, P25, P75, P90, minimum, maximum, and missing-field rate. State the percentile method used by the analysis tool.
Inspect the underlying records behind extreme values. A reopened deal, migrated timestamp, data-entry mistake, strategic enterprise account, or genuine process delay can each create a long cycle and requires a different response.
Segment without hiding sample size
Compare market, deal-size band, product, channel, sales motion, owner, and won/lost status only when each group has enough records and uses the same event definitions. Show counts beside every statistic and avoid ranking tiny segments.
Add stage-entry and stage-exit timestamps to calculate dwell time. The total cycle can grow because qualification is slow, proposals wait for review, legal negotiation expands, or next steps are missing; each bottleneck has a different owner.
Choose and verify one improvement
Review median and tail cases, stage dwell, no-activity periods, next-step age, loss reasons, handoffs, response time, and customer dependencies. Select one operational change such as qualification criteria, proposal template, approval route, scheduling handoff, or follow-up rule.
Deliverable: metric definition, event-field dictionary, cohort rules, frozen record export, censoring statement, duration calculation, average and percentile table, segment counts, outlier audit, stage-dwell analysis, bottleneck hypothesis, process change, matched future cohort, and improvement review date.
THE TAKEAWAY
Define event timestamps and cohort rules first, show median and tail percentiles beside the average, disclose open-deal censoring, segment only when volume supports it, and improve the stage with the clearest avoidable delay.OFFICIAL REFERENCES