Every analytics script has a hole in it. Independent estimates put the share of visitors blocked from measurement at somewhere between 15 and 30 percent, published by analytics vendors including Sleek Analytics and Kissmetrics, and swinging hard by audience. A developer tool loses far more than a recipe site. Ours has the same hole, and serving the script from your own domain does not close it.
Most tools mention this in a support article and move on. There is one instrument that still works when a script is blocked, and it is old fashioned. Ask.
One question, fixed answers
A post-purchase survey is a single question on the confirmation page. How did you hear about us. A short list of answers you wrote yourself, and no free text box.
The no-text-box rule is not squeamishness. A text field on a checkout confirmation page is a machine for collecting things you did not want and now have to store: names, order numbers, email addresses, complaints. Fixed choices answer the attribution question and collect nothing you have to defend later.
Keep it to one question. Response rates fall off a cliff at two, and the second question is never worth what it costs you on the first.
The distinction that makes it useful
Here is where most implementations stop short, and where the interesting part actually is.
Suppose a customer answers Google and you saw no click. Google clicks are the kind you should have seen, so that gap is measurement loss. An ad blocker, a stripped parameter, a broken redirect, an untagged campaign. Something to go and fix.
Now suppose a customer answers a podcast. There was never a click to see. Nobody taps an ad in a podcast. That is demand arriving through a channel click tracking cannot observe even in principle, and it is not a bug in anything.
- measurement lossClaimed a channel that should have produced a click, and none was seen
- a fixWhat that number is: tagging, redirects, or accepting the blocked share
- unmeasurable demandClaimed a channel that never produces a click at all
- a findingWhat that number is: spend working where clicks cannot look
Added together they become one meaningless number called unattributed. Kept apart, one is a repair job and the other is a budget insight.
Every tool we looked at while building this adds those two together and reports the sum as unattributed. That single decision is why survey data usually reads as a vague second opinion instead of as evidence.
What to do with the answers
Show both numbers. Not a blend. If your click data credits Google with 40 percent of conversions and your customers credit it with 55, the honest output is both figures and the gap between them, not a third number you invented by averaging.
Self-reported memory is evidence, and it is a different kind of evidence from a click. People misremember. People say Instagram when they mean a creator they saw on Instagram. Neither source is ground truth, and pretending one of them is has been the field's favourite mistake for a decade.
This matters most for creator campaigns, where discovery routinely happens without a click at all. If that is where your budget is going, the definition is worth five minutes: what is paid amplification. The rules we hold ourselves to when reconciling the two sources are on how we measure.