How to find which pages lose you the most customers

Why exit rate is the wrong sort

Exit rate answers “where do sessions end”, and sessions ending is not in itself a problem. Your confirmation page has a near-total exit rate and is working perfectly. A guide that answered someone’s question completely has a high exit rate and did its job.

The related trap is bounce rate on entry pages. A high bounce on a page that receives cold paid traffic tells you about the ad, the audience and the promise, and very little about the page.

What you want to know is which page, if it worked better, would put the most money back. That is a question about volume and transition rate together, and neither number answers it alone.

Templates, not URLs

This is the change that most often relocates the answer. On a Shopify store the product page is one template rendered for every item; the cart is one template; a collection is one template with a filter set. Ranking individual URLs scatters one problem across hundreds of rows, each too small to notice.

The exception worth carving out is a URL that behaves unlike its template: one collection with a broken filter, one product with a variant picker that fails. Rank by template first to find the systemic loss, then look for outliers inside the template to find the local one.

In practice the two have different fixes. A template fault is one change that pays across the catalogue. A single-URL fault is usually a content or configuration mistake and is cheaper still.

Converting a drop into a figure

The arithmetic is deliberately simple: sessions that reached the page, times the share that failed the transition above what a comparable page achieves, times revenue per session on that path. The result is a monthly number you can rank against every other page.

Two cautions make the difference between a useful figure and a flattering one. First, the comparison has to be like for like: a collection page reached from paid social is not comparable to one reached from a returning customer’s bookmark, and mixing them will price a traffic-quality problem as a page problem.

Second, resist summing the whole list and calling it recoverable revenue. Losses overlap, some are the same shopper counted twice, and no fix recovers all of a loss. A ranked list is for choosing what to work on. It is not a forecast, and a tool presenting it as one is selling you something.

When the page is not the problem

Often the expensive page is innocent and the page before it set a false expectation. A collection that advertises a low starting price, leading to product pages where nothing comes close to it, produces a product-page exit that is really a collection-page fault.

Speed is the other displaced cause. A page that paints slowly on a phone has lost a share of its traffic before any of its content had a chance to be the problem, and no amount of copy work on that page will show up in a test.

The tell is the shape of the loss. Friction on the page itself produces engaged sessions that fail: scrolling, clicking, hesitating. An expectation or speed fault produces short sessions with almost no interaction. They look identical in an exit-rate report and need different fixes.

Doing it without building the report

If you are assembling this by hand in GA4, the path is: filter to human sessions, build a funnel exploration per template with the transition each page owes, export the step counts, and do the revenue arithmetic in a spreadsheet. It is an afternoon, and it is stale within a fortnight, which is why few people repeat it.

Liftable’s version, once Shopify is connected and the pixel is on, is a list of opportunities rather than a page report. Each names the surface (home page, collection, product page, cart) and the problem, with the sessions it was seen in, and gets an estimated monthly range with its inputs, lookback period and calculation shown. It is a modelled estimate to put the list in order, not a measured result.

Before the pixel is on, the free scan works from a crawl: it finds faults on your own pages, such as dead product-card links or an add-to-cart button below the fold on mobile, and prints no dollar figure. There is no heatmap or recording to browse in either case. The pixel records the steps a shopper took, not a video.

The ranking, in order

1. 01

   Take the bots out of the denominator first

   Automated traffic lands on pages unevenly: it hammers the home page and a few product URLs and never reaches the cart. Leave it in and your entry pages look far worse than they are. Filter before you rank, not after.

2. 02

   Group by template, not by URL

   Two hundred product URLs are one product page with two hundred sets of copy. Ranked individually each looks trivial; ranked as a template the loss is often the largest single number on the list, and it is one fix rather than two hundred.

3. 03

   Define the next step each page owes

   A page is not failing because people left it. It is failing because they did not do the one thing it existed to make easy: reach a product from a collection, add to cart from a product, reach checkout from a cart. Measure that transition, and only count a session at a step if it cleared the step before.

4. 04

   Turn the gap into money

   Multiply the sessions that failed the transition by your revenue per session for that path. This is what converts a percentage into a ranking you can act on, and it is the step almost everyone skips.

5. 05

   Split every row by device

   Do this before you draw any conclusion. Phones are usually most of the sessions and a smaller share of the orders, so a mobile-only fault sits inside the average looking like a mild general one.

6. 06

   Look at what shoppers did on the top row’s page

   Now, and only now, look at behaviour on the worst page: dead clicks, abandoned form fields, repeated taps. You are looking for the cause of a loss you have already sized, not fishing for something interesting.

Questions