Funnel analysis: how to find where your sales funnel really leaks

Funnel analysis is how you find the exact step where buyers give up. You define the sequence a customer goes through, such as product view, add to cart, checkout and purchase, count the unique users who complete each step, and compare the step-to-step rates. The step with the largest unexplained drop is where your next fix, and your next test, should go.

Funnel analysis is the measurement of how many users complete each step of a defined sequence of actions, and what share of them drops off between one step and the next, within a set time window.

How do you define funnel steps that match real behavior?

Most broken funnels are broken before the first number appears. Someone draws the journey they designed on a whiteboard and turns it into a report, and the report then measures a path that few real customers take.

Start from behavior instead. Look at the pages and events that actually precede purchases on your site. If half your buyers land straight on a product page from a search ad and never see the home page, a funnel that begins with “Home page” will understate your conversion and blame the wrong page.

A good step has three properties:

• It is a meaningful commitment. “Viewed product”, “Added to cart” and “Started checkout” each represent a decision. “Scrolled 50%” usually does not.
• It is tracked reliably. Use an event that fires once per real action. If the “add to cart” event fires on page load in some templates, your step two is fiction.
• It is shared by most buyers. A step only some buyers pass through, like “Applied coupon”, belongs in a segment or a separate funnel, not the main path.

For a typical online store, four or five steps are enough: product view, add to cart, checkout started, shipping details entered, purchase. For a SaaS sign-up it might be landing page, pricing page, sign-up started, account created, first key action. Write the event name behind each step next to it, so anyone reading the report knows what is being counted.

What is the difference between step conversion and overall conversion?

Two rates come out of every conversion funnel, and they answer different questions.

• Step conversion = users who completed step n ÷ users who completed step n − 1. It tells you how well one transition works.
• Overall conversion = users who completed the last step ÷ users who entered the first step. It tells you how well the whole funnel works.

Here is a worked example. The numbers are hypothetical: a small online store over one month, counting unique users in a closed funnel with a seven-day window.

Overall conversion is 1,008 ÷ 40,000 = 2.52%. You get the same number by multiplying the step rates: 0.12 × 0.50 × 0.70 × 0.60 = 0.0252.

That multiplication is the most useful fact in funnel analytics. Because overall conversion is a product, a 10% relative improvement at any step produces a 10% improvement overall. Lift step five from 60% to 66% and purchases rise from 1,008 to 1,109. Lift step two from 12% to 13.2% and you get the same 1,109 purchases. At an average order of $40, either change is worth about $4,040 a month in this example.

So the biggest absolute loss (35,200 users at step two) is not automatically where to work first. Step two always loses the most people, because browsing is cheap and buying is not. The better question is which step is worse than it should be, and which is cheapest to improve. A payment step that converts at 60% when shoppers have already typed their address is a stronger signal than a 12% add-to-cart rate, which may be normal for your category. Compare each step with its own history and with its own segments, not with the other steps.

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

In a closed funnel, a user is counted only if they enter at step one. Anyone who arrives at checkout from a saved link, without viewing a product in the window, is ignored. In an open funnel, users can enter at any step, so step three can include people who skipped steps one and two. Google’s documentation for the GA4 funnel exploration describes the same split: in an open funnel “users can enter the funnel in any step.”

Use each for a different job:

• Closed funnels answer “of the people who started this journey, how many finished it?” That is the right view for the step and overall rates above, and for most conversion work.
• Open funnels answer “how much traffic reaches each step from anywhere?” They help when people enter mid-journey, such as returning customers going straight to the cart, or email links that land on checkout.

The trap is mixing them. In an open funnel a later step can have more users than an earlier one, and a “step conversion” above 100% is a sign you are reading an open funnel as if it were closed.

A related setting is step order. Tools let you require that steps happen in sequence with other actions allowed in between, or that each step immediately follows the previous one. GA4 calls these “indirectly followed by” and “directly followed by”; Amplitude offers “this order”, “any order” and “exact order”. For most sales funnels, “in this order, other actions allowed” matches how people actually shop: they view three products, compare, then add one to the cart.

How long should a funnel conversion window be?

The conversion window is the time a user has to go from the first step to the last one and still count as converted. Mixpanel’s documentation frames the funnel question this way, as the share of users who convert “within a particular time window”. Set it too short and slow but real buyers are counted as funnel drop-off. Set it too long and you credit the funnel with purchases that had nothing to do with the journey you are studying.

Pick the window from your own data:

1. Take users who purchased in the last few months.
2. Measure the time from their first product view to their purchase.
3. Look at the distribution and choose a window that covers most of it, for example the point where around 80 to 90% of buyers have converted.
4. Use the same window every time you compare periods, campaigns or tests.

A single session suits low-price, impulse products. A furniture store or a B2B sign-up might need 14 or 30 days. If you change the window, treat the result as a new series, not a continuation of the old one.

Which funnel analysis mistakes hide real drop-off?

These five errors are common, and each one makes a funnel look healthier, or sicker, than it is.

1. Too many steps. A ten-step funnel splits the journey into transitions so small that random noise looks like a problem. It also multiplies tracking failures: if any one event misfires, every step after it is wrong. Keep the main funnel to the few commitments that matter and build sub-funnels for detail, such as the checkout form fields.
2. The wrong conversion window. A one-session window on a product people research for a week turns most buyers into drop-offs. A 90-day window on an impulse product hides a checkout that is genuinely failing. Measure time-to-purchase before you choose.
3. Mixing sessions, devices and users. If the funnel counts sessions, a shopper who adds to cart on mobile tonight and buys on a laptop tomorrow appears as one abandoned session and one purchase with no history. Count unique users, and use a tool that can connect a user’s visits across devices, or your cross-device buyers will show up as drop-off.
4. Ignoring segments. A blended rate hides the step that is broken for one group. Say the hypothetical store’s 60% purchase rate is really 1,080 mobile users converting at 50% and 600 desktop users converting at 78%. The average looks acceptable; the mobile payment page is the problem. Split by device, channel, new versus returning, and campaign before you decide where to work.
5. Counting bots. Crawlers, uptime monitors and scrapers inflate the first step and drag every rate down. Some tools exclude known bots automatically; Google says GA4 does this and that you cannot see how much was removed. Bots that imitate users are harder to catch. A sudden jump in step-one traffic with no matching change in later steps is a sign to check.

How do you build and segment a funnel in AdPix?

AdPix includes a full conversion path report, the funnel view of your data, that follows every step from first visit to sign-up or purchase and shows where users drop off. The sequence below works whether you are setting it up for the first time or rebuilding a funnel you no longer trust.

1. Install the tag and confirm your events. Add the AdPix code to your site or import your Google Tag Manager container; setup takes under five minutes. Check in the realtime view that each event you plan to use as a step fires once per action.
2. Define the steps. Use the four or five commitments you chose above, in order, and keep the event names consistent with the rest of your reporting.
3. Read step and overall conversion side by side. AdPix reports are not sampled, so small segments and low-volume steps show the counts you actually recorded, not an estimate.
4. Count people, not visits. AdPix connects a user’s visits across devices and channels into one journey, which addresses the third mistake in the list above.
5. Split by source. Acquisition reports show channels, source / medium and UTM-tagged campaigns, so you can see whether one campaign brings visitors who never add to cart. Consistent tags matter here; the guide to UTM parameters covers a naming convention that holds up.
6. Segment the leakiest step. Segments are part of the Starter plan and up. Use them to compare device, new versus returning users, or a campaign against the rest of your traffic. Plan details are on the AdPix pricing page.

If you sell online, order analysis is available on every plan, and from the Starter plan up the online sales and revenue reports add revenue by product and campaign, so you can turn a step-rate change into money, as in the $4,040 example above. A purchase is also worth more than its first order when customers come back, which is why a checkout fix often pays off beyond the month you measure; the guide to customer lifetime value shows how to put a number on that. Invalid traffic detection is included from the Starter plan up; on the Free plan, watch for the bot pattern described in mistake five. AdPix can also run alongside Google Analytics while you compare the two funnels; the Google Analytics alternative guide explains how to run that comparison fairly, and the setup steps are in the AdPix documentation.

Once you know which step leaks, the next question is whether your fix actually works. Change one thing at that step, such as fewer form fields, a clearer delivery cost, or a different payment layout, and test it against the current version. Size the test before you start: a step with 1,680 users a month needs a much larger effect, or a much longer test, than a step with 40,000. The guide to A/B test sample size explains the arithmetic, and Kohavi, Tang and Xu’s Trustworthy Online Controlled Experiments is the standard reference on running tests you can trust.

Finally, give the funnel a place in your regular reporting. Overall conversion belongs among your core marketing KPIs, and step rates belong next to it so a drop is visible in the week it happens. When a campaign looks strong at the top of the funnel but weak at purchase, check how you credit it with attribution models before you move budget. Pick the one step that looks worse than its own history this month, segment it by device and channel, and start there.

A/B test calculator

Sample size and duration for your next experiment. The free plan covers 100,000 events a month, no credit card. Setup takes under five minutes.

Start freeTry it free