Funnel analysis: find the step where people leave

Funnel analysis counts how many people reach each step toward a result, such as product, cart, checkout and purchase, so you can see the step where most of them leave. A conversion rate says "sales are low"; a funnel says "half the people who start checkout never finish", which tells you what to fix.

This guide covers how funnels are counted, the four settings that change the answer, which drop-offs are normal, and a checklist for reading one honestly.

What a funnel is#

A funnel is an ordered list of steps toward one result, and a count of how many people reached each step:

Example numbers, made up for illustration.

The two percentage columns answer different questions. Of starters is the overall conversion to that point. Went on is the step's own pass rate, and it is the one that points at a problem: here, half the people who start checkout do not finish.

The settings that change the answer#

Two funnels over the same data can disagree by a factor of two. These settings are why.

Who is counted: people, visits or events

Most funnels count people (or browsers), once each. A person who opens the cart three times is one person at that step. Counting visits instead makes every step look worse for products people research over several days, because the visit that buys is often not the visit that browsed.

Closed or open

In a closed funnel, people must start at step 1. In an open funnel, people can enter at any step, so someone who lands straight on the cart from an email counts from there. GA4 offers both (Google). Closed funnels are easier to read; open funnels show entry points you forgot existed.

In order, and how strictly

Steps must happen in order. Most tools also let you choose whether a step must come directly after the one before or at any point after it, with other pages in between. Directly-followed funnels are almost always too strict for websites, where people wander.

The time window

How long someone has to get from step 1 to the last step: one visit, a day, a week, 30 days. A short window undercounts considered purchases. A long window with a short date range overcounts nothing, but makes this week's funnel look worse than last month's, because this week's starters have not had the time to finish yet.

Pick the window that matches how long your customers take, and compare periods that have each had the full window.

Settings for common questions

Which drop-offs are normal#

Every step loses people. The question is which losses are unusual for your business.

• Product to cart is usually the biggest drop on a store, because most visitors are browsing.
• Cart to purchase is the best-documented one. Baymard Institute averages 50 studies to a cart abandonment rate of about 70 percent (Baymard, checked October 2026). In its survey of US shoppers, the most common reason after "just browsing" was extra costs such as shipping, tax and fees.
• Form view to form sent on a lead site drops with every field you add.

The most useful comparison is your own funnel against itself: last period, by source, by device. A checkout that loses 50 percent on desktop and 80 percent on phones has a phone problem, whatever the industry average says.

How to read a funnel without fooling yourself#

1. Read the step's pass rate, not the bar height. The last bar is always small. A step is a problem when its own pass rate is low or falling.
2. Split it by source. A falling funnel is often a change in who arrives, not a broken page. Paid social visitors finish less often than people who searched for your brand, and a new campaign changes the mix.
3. Split it by device. Phone checkouts break in ways desktop testing never shows.
4. Look at time between steps as a median. A few people who come back after three weeks drag an average far from the typical case. GA4's elapsed time is an average (Google); a median says what a typical person does.
5. Check the step definitions when a number jumps. A renamed page, a new checkout domain or a form that stopped sending its event looks exactly like people leaving. If a step suddenly drops to zero, suspect the tracking first. See cross-domain tracking for checkouts on another domain.
6. Watch sample size. Fifteen people at a step is too few to judge a change of a few points.

What to do with the leaky step#

Look at the people, not only the percentage. Watch where those who left came from and what they did last: a coupon field that sends them searching for codes, a shipping cost they first see at checkout, a form that asks for a phone number they do not want to give.

Then change one thing, and read the same funnel, with the same window, for a period long enough to reach a few hundred starters.

Where Sporetag fits#

From Sporetag's funnels code and help: a site gets a suggested funnel built from its goals and the pages people really visit. Steps are pages (with * for any part), events or goals, 2 to 8 per funnel, up to 10 funnels a site, on paid plans. Each gap shows how many left and the median time the next step took. Split by compares sources, campaigns or devices. Steps count within one visit, or within 1, 7 or 30 days across visits with 90-day tracking, and then the number who left opens their profiles.

Common questions#

What is a conversion funnel?

An ordered list of steps toward one result, such as product view, cart, checkout and purchase, with the number of people who reached each step. It shows where people leave, which a single conversion rate cannot.

What is the difference between an open and a closed funnel?

In a closed funnel, people must start at the first step. In an open funnel, they can enter at any step, so someone who lands on the cart from an email counts from the cart.

What is a good funnel conversion rate?

It depends on what you sell, the price and the traffic. Compare each step with its own past and across sources and devices rather than with an industry average.

Why does my funnel show fewer people than my goal count?

The funnel counts only people who did the earlier steps first, in order and within the window. Goals count everyone who reached the result, however they got there.

Sources#

• Google Analytics Help: Funnel exploration. Checked October 2026.
• Baymard Institute: Cart abandonment rate statistics. Checked October 2026.
• The reading checklist is our own recommendation from building funnel reports. Figures marked as examples are made up for illustration.