An agency's conversion problem isn't finding friction. Experienced people are good at that. It's proving that removing it helped, on a client's site, with a client's traffic, in a way that survives a quarterly review.
Two structural difficulties make this hard, and both are solvable.
Difficulty one: your findings aren't repeatable
Two consultants audit the same site and produce two different lists, ordered differently. Neither is a baseline. When the client asks "is the site better than in March," the honest answer is "in our judgment, yes," and judgment is exactly what a nervous client discounts.
A deterministic measure fixes this. If the same page produces the same score every time, then a score that rises means the page changed — not that a different person looked at it. It converts your work from opinion into evidence, which is what renews retainers.
Difficulty two: client traffic rarely supports testing
Most client sites can't reach significance on an A/B test in a quarter. Agencies know this, and the usual responses are both bad: run the test anyway and report an underpowered result as a win, or skip measurement and report activity.
The better answer is to change the standard of evidence deliberately, and say so. Fix defects that don't need proof. Measure at the step you changed rather than the total. Watch direction across weeks. Tell the client in advance which claims will be provable and which won't — a client who was warned respects the distinction; a client who wasn't suspects everything.
What scales across a client book
A per-client baseline. One number per site, computed identically, tracked over time. It makes a portfolio review possible in an hour rather than a week.
One fix at a time, per site. Not because you can only do one, but because a fix that can be credited teaches you something transferable to the next client. Five simultaneous changes teach you nothing you can reuse.
A finding library. Friction patterns repeat across clients — competing CTAs, costs revealed late, forms with a phone number nobody needed. An agency that records which fixes actually moved which steps is compounding an asset. One that reports activity is not.
The uncomfortable part
Some of what agencies sell as optimization is unproven. That's not a scandal; it's the state of the field. Best practices are priors, not results.
The agencies that will look best in three years are the ones drawing a visible line between "we fixed a defect," "we applied a well-established principle," and "we measured a lift." Clients can tell the difference, eventually, and they remember who told them first.
Defrixa gives a deterministic per-page score and one ranked fix, free per URL. With the snippet installed on a client's site, it measures which step loses visitors and whether the change you shipped moved it.
Common questions
Report the defect fixed, the step it affected, and direction over weeks — and say plainly that significance isn't available at this volume.
Guaranteeing a percentage you cannot control is a promise the data will eventually break. Guarantee the process and the measurement.
Usually one action made unmistakably primary, and a cost moved earlier in the flow.