A funnel leaks at every joint. That's normal — not everyone who lands should buy. The skill is not sealing every joint; it's knowing which leak is worth patching and in what order.
Rank leaks by recoverable people, not by percentage
A step that loses 60% sounds worse than one that loses 20%. But if the 60% step is early, where visitors include everyone who arrived by accident, most of that loss was never recoverable. The 20% at the payment step is composed entirely of people who intended to buy.
So rank by recoverable people: the number reaching the step, times the share who intended to continue, times how much of that loss is plausibly caused by friction rather than fit.
In practice this means later leaks usually pay more, because commitment has already been demonstrated. Fix from the bottom up.
Distinguish leaks from filters
Some drop-off is a filter doing its job. A pricing page loses people who can't afford the product; that's information, not damage. A qualification field loses people you didn't want.
Ask of each leak: would I want these people back? If the honest answer is no, the step is working. If it's yes, it's a leak.
This one question saves teams from "optimizing" away the filters that protect their sales time.
Find the joint, then find the cause
The funnel tells you the joint. It doesn't tell you the cause, and the causes at each joint are stereotyped:
- Losing people before engagement: the promise didn't match the click, or the first screen didn't answer "is this for me."
- Losing them at the decision page: an objection went unanswered — price, proof, fit, or what happens if they're wrong.
- Losing them in a form: a specific field is too expensive, hostile, or premature.
- Losing them at payment: a surprise cost, an absent trust cue, or a missing payment method.
Match the joint to its stereotype, then verify by walking the step yourself, on a phone, as a stranger.
Patch one, then measure that joint
The discipline that separates teams who learn from teams who don't: change one step, watch that step.
Watching total conversion after three simultaneous changes tells you the total moved. It never tells you which change moved it, so the next decision is another guess, and the guessing never compounds into knowledge.
When the funnel itself is wrong
Sometimes the leak isn't in a step — it's that your funnel isn't the path visitors take. You built landing → pricing → signup; they actually go landing → docs → pricing → back to landing → signup.
A funnel that doesn't match behaviour reports leaks that aren't there and hides the ones that are. Before optimizing steps, confirm the steps are real.
Defrixa suggests a starter flow from your site's structure immediately, and once the snippet is live it compares that flow against how visitors actually move — flagging where the real path differs from the assumed one.
Common questions
Usually not. Fix the drop with the most recoverable people, which is often a smaller percentage at a later, higher-intent step.
Few enough that each one is a real decision. A twelve-step funnel produces twelve small numbers and no clarity.
Fix the later one. Its visitors have already demonstrated intent, so more of that loss is recoverable.