Finding Funnel Leaks by Stage: The Working Guide

A leak is a stage where visitors who intended to continue do not. Finding leaks is a locating exercise before it is a fixing one: compute the step rates (the foundations funnel), compare each stage with its neighbours and its own history, then diagnose the stage with the largest unintended loss using recordings, surveys and support evidence. This bench gives each stage a guide - arrival, browsing, product consideration, the add-to-cart step, the cart, checkout start, inside checkout, and after the sale - because the causes, the evidence and the fixes differ by stage, and stage-blind advice ('improve your funnel') fixes nothing.

Not every drop is a leak. Browsers who were never buying, visitors in the wrong country, researchers pricing a future project - their departure is the funnel working. Leakage is the loss of visitors whose intent the earlier stages had already established: the person who chose a product and could not find the size, who reached the cart and met a cost that should have been stated an hour earlier, who started checkout and was stopped by a form. The diagnostic art is separating the two, stage by stage, because repairing intended loss is impossible and ignoring leakage is expensive.

Arrival leaks are audience and promise problems. Browsing leaks live on category pages and in navigation. Consideration leaks are product-page information failures. The add-to-cart step leaks on mechanics and feedback. Cart leaks are hesitation with the total visible. Checkout-start leaks are doorway friction: accounts, surprises, commitment shock. Inside-checkout leaks are forms, payment and errors. Post-sale leaks are quieter: delivery disappointment and the missing second order. Each guide on this bench follows the same pattern: the stage's signature, the causes in the order they actually occur, the evidence that separates them, and the repair sequence.

Simultaneous repairs across stages destroy attribution: fix the product page and the checkout in the same month and the overall rate moves without telling you which repair earned it - or which one hurt. Work the stage with the largest unintended loss, verify the repair on that stage's step rate, then move on. The exception is the trivial fix: a defect found anywhere (a broken filter, a dead button) gets fixed on sight regardless of the current stage, because defects are not experiments. The prioritisation method on the foundations bench keeps the queue honest across stages.

Three instruments cover most diagnoses. Step rates locate the stage. Session recordings at the stage show the behaviour - the hesitation, the rage click, the form abandoned at one field. And the voice of the visitor - support messages, post-purchase surveys, abandoned-cart replies - names the cause in words numbers cannot. A stage diagnosed with all three rarely stays mysterious. A stage diagnosed with step rates alone invites the classic error: redesigning the page when the leak was a price, a stock-out or a traffic change wearing the page's clothes.

Monthly, refresh the five-line funnel by segment and scan for stage movement; quarterly, take the worst stage through a full diagnosis with recordings and visitor voice. Log every repair with its stage and its read rule. Over a year the log becomes a map of where your funnel personally tends to break - most stores have a characteristic weakness (an information gap, a mobile step, a cost-clarity habit) that recurs in new costumes, and recognising the costume early is the experienced operator's whole advantage.

Illustration: a store's monthly funnel shows sessions 8,000, product views 3,900, adds 310, checkout starts 240, orders 150. The stage everyone complains about is checkout (240 to 150, a 62 percent completion that feels low), but the map says otherwise: the arrival-to-product step and the view-to-add step together discard far more intent, and a recordings sample at the product stage finds visitors hunting for a delivery date the page never states. The quarter's work follows the map: the delivery answer moves onto the product page, the add rate is the read, and checkout - the loudest stage - waits its turn with its 62 percent, which turns out to be the chain's healthier link. Maps beat loudness; that is the bench in one paragraph.