Traffic comes in, but the people who ultimately leave their info or place an order are always fewer than expected — which level is the problem? A lot of teams only worry about the total number, without looking at how much leaks at each level of the funnel. A conversion funnel breaks the path from click to deal into stages, calculates retention at each stage, and pinpoints the level where the most traffic is lost. This article explains how to build a funnel that actually works, and how to fix it once you’ve located the problem.
First, break the path into countable stages
The first thing in a funnel isn’t looking at data — it’s drawing the path. From the user’s first click to the final conversion, whatever mandatory steps sit in between become your stages. A typical lead-gen funnel is: landing page visit, key content read, form shown, form submitted, deal closed. Each step needs a trackable event, otherwise that stage is a black box.
I once took over a site where between landing page visit and form submission there were three popups and a long article, with zero event tracking in between — nobody could say at which step users were leaving. After adding the events, we found that 70% of users left at the second popup. Before the tracking was in place, that funnel stage simply didn’t exist.
Calculate retention and leakage at each stage
For each stage, use the previous stage as the denominator to calculate retention rate, and this stage as the numerator. For example, a landing page gets 1,000 visits, 30 form submissions — overall conversion is 3%. But broken down: visit to content read retains 60%, read to form shown retains 50%, shown to submitted retains 40%. The biggest loss is at the read-to-form-shown stage — it swallows half of the remaining people.
| Funnel stage | People at this stage | Retention vs previous | Problem assessment |
|---|---|---|---|
| Landing page visit | 1000 | — | Entry, baseline |
| Content read | 600 | 60% | Normal |
| Form shown | 300 | 50% | Heavy loss zone |
| Form submitted | 120 | 40% | Can optimize |
| Deal closed | 90 | 75% | Final push |
The criterion for定位 is simple: whichever stage has a retention rate明显 lower than its neighbors is the priority fix. Don’t spread effort evenly — concentrate resources on the stage with the heaviest loss, and the return is highest.
Common causes of loss by stage
Loss at different stages has different causes and different fixes.
- Entry-stage loss: usually intent mismatch — users search for A and land on B, so they leave immediately. Fix: make the first screen directly answer the search term.
- Middle-stage loss: mostly trust or patience issues — content too long, too many popups, too slow to load. Fix: cut steps, reduce distractions, speed things up.
- Submission-stage loss: the form is too long or asks for too-sensitive information. Fix: reduce fields to a minimum — just email or WeChat is enough to start.
- Deal-stage loss: often stuck on response speed — leads sit for two days before anyone follows up. Fix: write follow-up timeliness into the process.
Investigation order: work backward from the stage with the heaviest loss. First confirm the event tracking for that stage isn’t missing, then look at page behavior and user feedback at that stage, and only then change the design.
A quick reminder: causes and stages don’t always map one-to-one. Behind one heavy-loss stage there might be several causes stacked together. Fix the one with the lowest cost and fastest effect first — for example, remove one popup or shorten one form, see if retention moves, then decide whether to touch the bigger structure. Small, fast iterations make it easier to tell which change actually worked than one big overhaul.
How to verify you fixed it right
After making changes, don’t say it worked based on feeling. Record the retention rate, bounce rate, and conversion before the change as a baseline, then observe for two to four weeks after. The criterion: under roughly the same traffic volume, has the target stage’s retention rate recovered? If traffic also rose, pin the traffic volume separately and only compare conversion rates — then the conclusion is clean.
I once cut a form from nine fields to three, and moved its position from the end of the article to just below the first screen. Retention from form-shown to form-submitted went from 40% to 61%. Landing page visits didn’t change, but deals closed went from 90 to 137. That’s the return from concentrating effort on the stage with the heaviest loss.
Don’t split stages too fine or too coarse
There’s a degree to splitting stages. Split into two or three and you can’t see where the problem is; split into ten and you’ll see small losses at every stage, making it hard to tell what matters. I generally keep a main path to three to six stages, each corresponding to a real action node rather than artificially chopped micro-steps. If there’s a long stretch between nodes with no events, that stretch is a blind spot — prioritize adding tracking there rather than splitting further down.
Look at versions by source and device
The same funnel often loses traffic at different positions depending on source and device. Users coming from organic search vs. paid ads have different intent and lose at different stages. Mobile loses more at the form stage — often because input fields are too cramped on small screens or the captcha is too annoying. I run the funnel separately by channel and device. The stage you定位 might be normal on desktop but collapsed on mobile — in that case the fix targets mobile, not a site-wide redesign.
Once I looked at the overall funnel and thought the form stage was fairly stable. Splitting it down to mobile revealed that shown-to-submitted was only 20%. Turns out the submit button on that site’s form was blocked by a floating window on narrow screens — users literally couldn’t click it. The overall average had covered up the problem. Looking at separate versions is a necessary step for locating loss.
Summary
The value of a conversion funnel isn’t in that final percentage — it’s in pointing out which stage is leaking. First add tracking so every stage can be counted, keep it to three to six stages, then calculate retention to locate the heaviest-loss stage, confirm it by source and device versions, push the changes in, and finally verify with before-and-after comparison. Once the path is clear, growth stops being a matter of luck.
