Two very different problems produce the same symptom. Either the right people are arriving and something stops them, or the wrong people are arriving and behaving correctly by leaving. The fixes have nothing in common, so establish which one you have before spending a month on the other.
Is the traffic even candidate traffic?
Segment by source and look at the shape of behaviour, not the conversion rate.
Wrong-audience traffic looks uniform. Short sessions, shallow scroll, similar bounce across every page, and a conversion rate that doesn't move no matter what you change. Broad paid campaigns, viral social links, and informational keywords that read as commercial all produce it.
Candidate traffic looks lumpy. Some pages hold attention. People reach the middle of the funnel. There's a specific step where they stop.
If the pattern is uniform, no page change will help. Your leverage is upstream: targeting, keywords, ad copy, and the promise being made before the click.
If they are candidates, find the step
Stop reading the sitewide rate. It averages over every location and therefore reveals none.
Instead: for each transition in your main flow, what fraction continues? Landed → engaged, engaged → started, started → completed. One transition will lose far more than the rest. That cliff is your answer, and it will be obvious the moment you stop averaging.
Then look at the same transitions on mobile separately. A large device gap at one step isolates interface friction, because everything else was held constant.
The most common leaks, by position
Early: the first screen doesn't say what this is or who it's for, or two co-equal actions force a choice the visitor didn't come to make.
Middle: the page that was supposed to answer the objection didn't. Price, proof, fit.
Late: a form asks for more than the transaction needs, or a cost appears for the first time near the end.
Late-stage leaks are the cheapest to fix and the most immediately profitable, because everyone there had already decided.
Check that you're measuring anything at all
Before diagnosing a conversion problem, confirm the conversion is being recorded. Broken tracking, a tag that fires twice, a thank-you page reachable without purchase, or a snippet missing on one template — all of these produce a "conversion problem" that no page change will fix.
It's a dull check and it resolves a surprising share of alarming numbers.
Getting more from the traffic you have
You don't need more visitors to earn more. Fix the widest transition, and every visitor you already pay for becomes worth more — including the ones from campaigns you can't easily scale.
Change one thing. Watch that transition, not your revenue. Changing five and watching the total is how a quarter passes with no knowledge gained.
Defrixa's free scan reads any page's structural friction and names the single biggest issue, so the first change isn't a guess.
Common questions
Enough that a step's drop isn't a handful of people. Below that, watch direction across weeks rather than chasing significance you can't reach.
It's usually the more expensive path. Every improvement to a leaking step compounds across all future traffic; more traffic through a leak just costs more.
Unknowable without context. The number that matters is the gap between two adjacent steps in your own funnel.