Where funnels leak
Every multi-step journey loses people at every step, and the losses are not evenly spread. Some are expected: visitors who arrived on a product page to compare prices were never going to buy today, and their departure is not a leak. Others are the funnel failing: a visitor who added an item, started the checkout, entered an address and then left at the delivery options was ready to buy and was stopped by something on the page. Funnel work is about distinguishing the second kind from the first and fixing the second.
The leaks cluster at predictable places. The transition from browsing to intent — product view to add to basket, service page to form start — leaks because the offer or the page has not done its job, which is landing page territory. The transitions inside the checkout or the form leak because of friction: account walls, long forms, surprise costs, limited payment options, errors with no way back. The final transition, payment to confirmation, leaks for technical reasons that are invisible unless someone looks: declined cards, failed redirects, timeouts.
Each of these shows up as a step-to-step rate in an analytics funnel, and the first task is to measure them. The marketing funnel guide explains the model and its limits; this section assumes the model and gets to the measurement.
Analysis and checkout: the core guides
- Funnel analysis — a step-by-step tutorial that defines the steps from events, builds the funnel exploration in GA4 as Google documents it, compares open and closed funnels, segments by device and source, compares periods, finds the biggest drop and turns it into a hypothesis. It includes an interactive drop-off explorer.
- Checkout optimisation — a guide to the friction points in an e-commerce checkout: guest checkoutLetting a shopper complete an order without creating an account, with the account offered afterwards when the details are already captured. and account walls, address and form entry, shipping and cost surprises, payment options and trust signals, and error handling and recovery.
The pages the funnel starts on — landing pages, product pages, service pages — have their own section at landing pages, and the discipline as a whole, with research and testing, is introduced in the conversion rate optimisation guide. A funnel finding usually becomes a test; the testing section covers how a test is planned, run and read.
Further guides
Further guides cover the other steps a funnel passes through, from the page where a plan or a product is chosen to the page after the conversion, and how to keep a funnel working after the site changes:
- Designing a pricing page — who visits, plan cards and comparison tables, complete costs, a call to action per plan, and research and testing.
- Category and listing pages — how listing pages help shoppers choose: product cards, filters and sorting, the default order, load more or numbered pages, empty results, phones and measurement.
- Sign-up and trial-start flows — what to ask now and what later, verification, card or no card, consent, the first screen and what to measure.
- Redesigning a long application form — a tutorial for applications that cannot be short: every question with its reason, eligibility first, what to have ready, short steps, saved progress, documents last and help where people stall.
- Booking and appointment flows — availability first, the fewest steps, full prices and deposits, confirmations, reminders and measurement.
- Upgrade paths for existing customers — the path from a plan that no longer fits to the next one: where the need shows, prompts that inform, the upgrade screen, billing, fair exits and measurement.
- Confirmation and thank-you pages — confirming and setting expectations, offering one next step, asking one question and counting each conversion once.
- Checking conversion paths after a release — a tutorial for walking the paths that earn money after the site changes: choosing devices, browsers and payment methods from the site's own data, marked test records, following each result to the order or lead and the person who must act, checking the count and logging the walk.
The biggest-leak rule
With the steps measured, the question is where to work. The rule is to work on the step whose improvement would add the most conversions, which is not the same as the step with the lowest rate or the largest absolute drop.
| Step | What a low rate usually means | Why it may not be the priority | Why it may be |
|---|---|---|---|
| Landing to product or service view | Traffic quality, message mismatch, slow page | Many arrivals were never going to continue; part of the drop is expected | It carries the most people, so a small improvement is many conversions |
| Product view to add to basket | Offer, price, proof, stock, page design | Comparison shopping is normal here | The intent signal is strong and the fix is on one template |
| Basket to checkout start | Delivery cost shown late, basket distractions, forced login | Some baskets are wish lists | People here have decided; losses are almost all real leaks |
| Checkout steps to payment | Form friction, address entry, delivery options, cost surprises | Multi-step checkouts spread the loss thinly | Everyone here wants to buy; fixes are concrete and testable |
| Payment to confirmation | Declines, errors, redirects, timeouts | The rate is usually high, so it looks fine | A small loss here is pure revenue loss and often a bug |
The arithmetic is simple: multiply the number of people entering a step by the improvement a fix could plausibly make to its rate, carry the result through the downstream rates, and compare across steps. The conversion rate calculator turns step counts into rates; the comparison is done by hand. In practice the late steps win more often than their rates suggest, because every person there has already declared intent and the fixes are on pages the business controls fully. The same reasoning applied beyond a single funnel, to production workflows, sales capacity and fulfilment, is the theory of constraints and its focusing steps.
Recovery alongside fixes
Fixing the funnel reduces the leak; recovery goes after the people who leaked anyway. The two are complementary and should run together, because the best-designed checkout still loses people to interruptions, comparison and cold feet. The cart abandonment rateShare of shopping carts created that did not become an order., read beside drop-off at each checkout step, shows how much of that loss is browsing and how much is friction. Recovery means identifying the shopper who left, which requires an email address or a login captured before the drop, and reaching them with a reminder of what they left, timed and worded to the step they left at. The abandoned cart flows tutorial builds the flow; the funnel work here decides what it says, because the message to someone who left at delivery options (the cost, the timing) is not the message to someone who left at the basket (the product, the proof).
Recovery has a trap: incentives. A discount in the recovery email teaches shoppers to abandon, and the uplift it produces is partly cannibalised from people who would have returned anyway. Fix the leak first, recover without incentives, and reserve incentives for a tested policy. Average order value (AOVAverage revenue per order.) is the guardrail: a recovery programme that lifts the conversion rate (CVRShare of visits or clicks that convert.) while shrinking baskets may be costing money.
A homeware retailer's leak map
A homeware retailer builds a closed funnel from its GA4 events for one month: 240,000 product views, 31,000 adds to basket, 14,500 checkout starts, 9,100 reaching payment and 8,200 purchases. The lowest step rate is product view to basket at 13%, and the team's instinct is to work there. The arithmetic says otherwise. A plausible fix on the product page might lift that rate by a point, worth roughly 2,400 extra baskets and, at the downstream rates, about 630 orders. Checkout start to payment is 63%; a plausible fix on address entry and delivery options — the two steps where recordings show people stalling — might lift it by five points, worth about 680 orders from far less traffic and on pages the team controls fully. Payment to purchase is 90%, which looks healthy until the team finds that nearly half of the loss is a single card provider's redirect timing out on mobile: a bug, fixed in a day, worth about 400 orders a month.
Common funnel problems
| Symptom | Usual cause | Guide |
|---|---|---|
| The funnel in analytics shows more people at a later step than an earlier one | Open funnel read as closed; a step event missing on some templates | Funnel analysis |
| Step rates look fine but overall conversion is low | Loss spread thinly across many steps; a long checkout | Checkout optimisation |
| Mobile converts far below desktop at the same step | Form and address entry on small screens; payment options; layout | Checkout optimisation, funnel analysis |
| A large drop at the first checkout step | Forced account creation; login wall | Checkout optimisation |
| A drop at delivery or payment options | Delivery cost or timing shown late; a missing payment method | Checkout optimisation |
| Payment to confirmation is lower than expected | Declines, redirects, timeouts; the confirmation event not firing | Funnel analysis |
| The overall rate fell sharply from one day to the next | Tracking stopped, a new traffic source, a browser bug, failing payments, a price change or bots | Conversion rate dropped suddenly |
| Recovery emails lift orders but revenue is flat | Incentives cannibalising full-price orders; order value falling | Abandoned cart flows |
| Nobody agrees which step to fix first | No leak arithmetic; priorities set by opinion | The biggest-leak rule above, then funnel analysis |
| Shoppers reach a category page and open few products | Product cards without the deciding attributes; filters that do not match buying decisions | Category and listing pages |
| Many visitors reach the pricing page and few choose a plan | Plans not led by who each is for; costs incomplete; one call to action for every way of buying | Designing a pricing page |
| Sign-ups rise but few become active or paying | Sign-ups counted rather than what happens next; too much asked before the first session | Sign-up and trial-start flows |
| Visitors start a booking and leave before choosing a slot | Personal details asked before availability is shown | Booking and appointment flows |
| Purchases or leads are counted twice when the confirmation page reloads | The conversion event fires on every load, with no transaction identifier | Confirmation and thank-you pages |
| Applicants start a long application and few submit it | Questions asked before eligibility is clear, no way to save progress, documents demanded midway | Redesigning a long application form |
| Customers reach a plan limit and contact support instead of upgrading | A prompt that blocks without explaining the step up, or billing that is unclear | Upgrade paths for existing customers |
| A release went out and some visitors may be unable to complete a purchase or an enquiry | A change broke one device, browser, payment method or entry point and no one walked the paths afterwards | Checking conversion paths after a release |
Step rates are the KPIA metric tied directly to an objective. set for this section. A fix is proved by its step rate moving in a controlled test, not by the overall rate moving in a month, because the overall rate moves with traffic mix and season whether or not anything was fixed.
If a step rate or the overall rate has suddenly changed, the conversion rate troubleshooting section separates tracking faults and traffic shifts from genuine losses inside the funnel.
