Funnel Analysis: How to Measure and Read Drop-Off

Funnel analysis is the measurement practice of mapping the sequential steps a user takes toward a conversion goal and quantifying where they drop off. It answers what share of users advance from each step to the next, where the largest leaks sit, and how long the journey takes. It is a diagnostic method, not an optimization playbook.

What Is Funnel Analysis and What Question Does It Answer?

Funnel analysis is the act of defining a linear or branching path of user actions and counting how many people complete each step. The core output is a set of conversion rates: step-to-step rate (the share of people who reached step N who also reached step N+1) and step-to-entry rate (the share of people who entered the funnel who reached step N). Together these tell you both the local leak at each boundary and the cumulative leak from the top.

The question it answers is "where do we lose people, and by how much?" That is different from asking "which cohort stays?" or "which channel gets credit?" Cohort analysis follows a fixed group of users across time to measure retention and decay. Attribution analysis distributes credit for a conversion across touchpoints. Funnel analysis sits upstream of both: it measures the path itself, before you worry about who stayed or who gets the win. A team that conflates the three usually reports a "funnel problem" when it actually has a retention or attribution problem.

How Do You Define Funnel Steps Correctly?

A funnel is only as trustworthy as its step definitions. Start by naming the action, not the page. An event like checkout_started is a step; a page view of /checkout is a proxy that fires when users bounce before interacting. Use the event that proves the user did the thing.

Decide four things per funnel:

• The entry population: who is eligible to enter (all visitors, signed-in users, users from a campaign).
• Required vs optional steps: a checkout funnel requires payment, but a saved-cart step is optional.
• Ordering: can a step be reached out of sequence, and does that matter for your question?
• Event naming: consistent, past-tense, action-based names that survive schema changes.

A step definition is wrong when it uses a proxy event, when it conflates a page view with an action, or when it includes steps a meaningful share reach by a different route. The most common error is counting page_view on a confirmation screen as the "purchase" step, which inflates the bottom of the funnel by including sessions that never transacted.

What Is the Difference Between Open and Closed Funnels?

A closed funnel requires users to pass through every prior step in order; anyone who skips a step is dropped from the counts. An open funnel allows users to enter at any step, so someone who lands directly on step three is still counted there. The toggle changes your numbers and your story.

Consider a worked hypothetical with 10,000 sessions that reach step one, 3,400 that reach step two, 900 that reach step three, and 210 that reach step four.

The step-to-step rate says verification converts 26.5% of those who started it, which frames the leak as moderate. The step-to-entry rate says only 9.0% of all entrants ever verified, which frames the leak as severe. Use step-to-step rate when you are debugging a specific boundary and want a clean local number. Use step-to-entry rate when you are sizing total opportunity or reporting to leadership, because it reflects the real cumulative leak. Never mix the two in one chart without labeling the denominator.

How Do You Run Funnel Analysis in Common Tools?

Most teams use one of three tool classes. The GA4 Explorations funnel exploration is the fastest entry point: drop your steps into the tab, toggle open vs closed, add a breakdown dimension, and read elapsed time between steps. It is event-scoped and good for a quick read, but re-defining history after a schema change is limited.

Product analytics tools such as Amplitude, Mixpanel, and PostHog let you build funnels on top of tracked events with flexible segmentation and retroactive property changes. They shine when you need to slice by plan tier or behavior cohort repeatedly. Warehouse SQL gives you the most control: you define steps as queries over an events table and can join any dimension.

A short generic SQL sketch for a closed funnel:

1. Create a CTE that marks each user's first timestamp for each step event.
2. Filter to users whose step timestamps are non-decreasing in order.
3. Count distinct users per step and divide to get conversion rates.

This pattern works in BigQuery, Snowflake, or Postgres with minor syntax changes, and it is the only approach that lets you redefine steps after the fact without re-instrumenting.

How Should You Segment and Read a Funnel?

An aggregate funnel hides the real problem. Slice by channel, device, new vs returning, and plan tier before drawing conclusions. A blended 2.1% activation rate might be 5% on paid search and 0.4% on a partner channel, and the fix is different for each.

This is where Simpson's paradox bites in plain terms: a change that looks like an improvement in the total can be a worsening in every segment, or vice versa, because the mix of segments shifted. If mobile conversions rose while desktop fell, but more mobile traffic arrived, the blended number can move opposite to both. Always read segment funnels side by side with the aggregate, and report the segment that changed the mix, not just the headline.

For teams building a longer-term retention lens on the same users, pairing this with cohort retention analysis shows whether funnel leaks are a first-session problem or a come-back problem.

What Are the Common Mistakes and Traps?

• Conversion window too short: users who convert on day three are dropped if the window is one day.
• Session-scoped vs user-scoped funnels: a session-scoped funnel splits one user's journey across visits and undercounts completion.
• Sampling and thresholding: sampled reports and minimum-cell thresholds quietly hide small but important segments.
• Small-sample noise on late steps: a late step with 40 users swings wildly week to week; wait for volume.
• Tracking gaps mistaken for UX problems: a sudden drop is often a broken event, not a bad button.
• Comparing funnels across a tracking change: a schema migration makes before-and-after funnels non-comparable.

Before blaming design, confirm the events fired. A quick check is to compare the step count against a source-of-truth like raw server logs or a billing table for the final step.

How Do You Turn a Funnel Finding into a Prioritized Action?

Rank leaks by impact, not by rate. Impact equals step volume times a realistic lift. A 50% drop on a step with 10,000 users upstream is worth far more than an 80% drop on a step with 200 users. Write each candidate as a test with a cheap-first ordering.

1. List each leaky boundary with its entry volume and current rate.
2. Estimate a realistic improvement based on comparable work, not best case.
3. Multiply volume by estimated lift to get expected incremental conversions.
4. Order tests by expected impact divided by implementation cost.
5. Run the cheapest high-impact test first and re-measure the funnel.

This keeps the team from over-investing in a dramatic-looking but tiny leak. When a leak is large and the test is cheap, that is where the first sprint goes. The broader stage tactics for fixing leaks live in the conversion funnel optimization guide; this post is about measuring the leak so you can prioritize it correctly.

Key Takeaways

• Funnel analysis measures sequential step conversion, drop-off, and time to convert, separate from cohort and attribution analysis.
• Define steps from actions, not page views, and decide entry population, required vs optional steps, ordering, and naming up front.
• Closed vs open funnels and step-to-step vs step-to-entry rates give different answers; label the denominator you report.
• Segment before concluding, because aggregate funnels hide Simpson's paradox reversals between channels and devices.
• Most "leaks" are tracking gaps or window and scoping errors, so verify events before blaming UX.
• Prioritize by impact (volume times realistic lift), running the cheapest high-impact test first.

Frequently Asked Questions

What Is Funnel Analysis in GA4?

Funnel analysis in GA4 is built inside Explorations using the funnel exploration template, where you add ordered steps from your events and choose open or closed funnel mode. GA4 computes step-to-step and step-to-entry conversion and lets you add a breakdown by dimension such as source or device, plus an elapsed-time view between steps. It is event-scoped and good for fast reads, though retroactively redefining historical steps is limited once your schema has changed.

What Is the Difference Between Open and Closed Funnel?

A closed funnel counts only users who passed through every prior step in order, dropping anyone who skipped a step, while an open funnel lets users enter at any step and still counts them there. Closed funnels show a stricter, more linear path and are common for checkout-style flows. Open funnels reflect reality when users can legitimately enter mid-path, such as returning directly to a dashboard. The choice changes your conversion rates, so pick based on whether skipped steps are valid or invalid for your question.

How Do You Calculate Funnel Drop Off?

Funnel drop-off is one minus the conversion rate at a boundary, so if 3,400 of 10,000 users reach step two, the step-to-step drop-off is 66 percent. You can also express cumulative drop-off from entry as one minus the step-to-entry rate, which for 210 of 10,000 activated users is 97.9 percent. Always state which denominator you used, because step-to-step and step-to-entry drop-off describe different things and are easily confused in a report.

Why Does My Aggregate Funnel Look Fine but Revenue Is Down?

An aggregate funnel can look healthy while revenue falls because the mix of segments shifted, a Simpson's paradox effect where each segment worsened but the growing segment had a higher rate. It can also happen when a high-value channel's funnel degraded while a low-value channel grew its volume, lifting the blended number. Segment the funnel by channel, device, and plan tier, and watch the mix, not just the headline rate, before deciding what to fix.