Measure every step, not just the end
A single conversion rate hides where people leave. Write your funnel down as the steps a buyer actually takes: the visit, the first meaningful action, the lead or trial, the offer seen, the purchase. Then measure how many people reach each one.
Google Analytics has a funnel exploration built for exactly this: it shows completion and abandonment at each step you define, and lets you break the result down by device or channel. The step with the largest loss, in real numbers of people, is where to look first.
Audit the whole path, in order
Walk the funnel the way a buyer does and check each part:
Traffic. Does the promise in the ad, post or search result match what the page says?
The first offer. Is the lead magnet, trial or free tool something this visitor wants, or something the team wanted to make?
The landing page. One clear message, one clear next step, and the proof a sceptic needs.
Follow-up. Does the first email arrive while the visitor still remembers you, and does it continue the same story?
The offer. Can a buyer tell what they get, what it costs and what happens next?
Checkout. Surprise costs, forced account creation, long forms and missing payment options all lose buyers at the last step. Try it on a phone.
Measurement. Are the events firing? A leak you cannot see is still a leak.
The most expensive leak is a broken one
Some leaks are not about persuasion at all. A buy button that links nowhere, or opens the wrong product's checkout, loses every buyer who reaches it, and nothing in your analytics will call it a bug. We found both on our own sites by hand in September, which is why we now test our buy buttons continuously.
Fix one leak at a time
Pick the step that loses the most people and is cheapest to change. Change one thing, and measure it properly before moving on. The Lab's application on running a conversion-rate test you can trust covers how to tell a real improvement from noise. Repeat the audit every quarter; funnels drift as offers, pages and channels change.