A funnel measures how an eligible population progresses through ordered, observable stages within a defined time window. Each stage needs an event rule, identity rule, and sequence rule. Report both counts and conversion rates, preserve the original entry cohort, and investigate instrumentation or mix changes before interpreting a drop as customer behavior.
Who this is for: Product, marketing, sales, and operations analysts measuring progression through a multi-step customer journey.
- Define stages from meaningful completed events, not page views that may not represent progress.
- Keep eligibility, ordering, identity, and conversion windows explicit so rates are reproducible.
- Compare stage losses by segment and volume while checking tracking quality before recommending action.
Specify the journey and entrant
Begin with the decision the funnel will inform. A checkout funnel may seek payment failures, while an onboarding funnel may identify where new customers fail to reach first value. Choose an entry event that creates a meaningful eligible population. Landing-page traffic is too broad if only signed-in trial users can complete later setup.
Write each stage as an observable event with required properties. 'Interested' is not measurable until represented by a qualified action. Keep stages necessary and ordered. Optional branches, such as using either an import or a manual setup path, need explicit union rules rather than forcing every person through one fictional sequence.
Control identity, order, and time
Choose whether the unit is a person, account, order, lead, or session. Identity stitching can overcount anonymous-to-known visitors or merge shared accounts. Document how identifiers are joined and what happens when they change. The denominator at each stage should be traceable to the qualifying units from the prior stage.
Require events to occur after entry and in sequence. Set a conversion window based on the normal decision cycle, such as seven days after trial creation or ninety days after a qualified opportunity. An unlimited window rewards old cohorts with more observation time. For fair comparisons, use matured cohorts or clearly mark incomplete ones.
Handle repetition and branching
Decide whether first attempts, latest attempts, or any successful attempt count. A customer may revisit pricing five times or retry payment. Counting events rather than unique eligible units inflates progression. Preserve timestamps for time-to-stage analysis even when the primary funnel records only one qualification per unit.
Real journeys branch and loop. Use separate funnels for materially different paths or define a shared milestone that both routes reach. A single funnel chart is useful for ordered stages, but it can conceal path choice. Supplement it with path tables, failure reasons, and distributions when sequence complexity matters.
Diagnose changes responsibly
Show stage counts beside rates. A stable conversion percentage with much lower entrants still produces fewer outcomes. Segment by acquisition source, device, geography, plan, and new versus returning status only when sample sizes support comparison. Look for shifts in who entered before attributing the change to a stage experience.
Audit event volume, missing properties, release dates, and source-system latency. A sudden drop at one stage may be a renamed event. Annotate product and tracking changes on trends. Use the funnel to locate a question, then inspect qualitative evidence or run a controlled test before declaring a root cause.
Measure activation after trial signup
A collaboration product wants to know why fewer trial accounts become active teams, but users can invite colleagues before or after creating a project.
- Set the unit to trial account and entry to a verified trial creation, excluding employees and duplicate test tenants.
- Define activation as completing both a project and a colleague invitation in either order within fourteen days.
- Create intermediate milestones for project created, invitation sent, colleague joined, and first shared edit while retaining valid branches.
- Report entrant counts, stage rates, completion time, and matured weekly cohorts by company size and acquisition source.
- Check event coverage around releases before investigating the largest segment-specific loss with session research.
Funnel definition sheet
Use this specification to make a funnel query reproducible across tools.
- Purpose and unit: decision, outcome, entity counted, eligible population, and exclusions.
- Stage contract: event name, required properties, sequence rule, optional branch, and deduplication rule.
- Timing: entry timestamp, conversion window, cohort period, maturity rule, and timezone.
- Identity: anonymous and known identifiers, stitching policy, account membership, and duplicate handling.
- Validation: expected event volumes, release annotations, segment checks, and source reconciliation.
Common mistakes
- Using all website sessions as the denominator for a stage available only to qualified account holders.
- Allowing later stages to occur before entry or outside any consistent conversion window.
- Reading a tracking outage or acquisition-mix shift as evidence that the product experience worsened.
Try one
Design a three-stage application funnel for a loan prequalification flow without implying that applicants must be approved.
A satisfactory design identifies an eligible application unit, uses observable stages such as application started, required information submitted, and prequalification decision delivered, and specifies order plus a reasonable window. It distinguishes completion from approval, protects sensitive segmentation, reports counts with rates, and includes checks for duplicate applications, identity changes, system errors, and cohorts that have not had enough time to mature.
Sources
- Microsoft Power BI funnel chart guidanceOfficial Microsoft guidance for constructing and interpreting funnel visuals.
- NIST exploratory data analysis handbookNIST methods for exploring data, checking assumptions, and revealing structure.