Why a campaign has to earn the word verified

Some conversions cannot be traced to a post. Someone watches a video, taps nothing, and types your name into a browser four days later. The click that would have carried the code never happened.

This is not a bug in anyone's tracking. It is a real customer journey that leaves no machine-readable trail, and it is a large enough share of short-form's actual effect that no honest tool can ignore it.

What most tools do

They model it. A statistical model distributes the untraceable conversions across your content according to some assumption: proportional to reach, weighted by recency, fitted against historical correlation. The output is a complete-looking report where every conversion has a source.

The appeal is obvious. Complete reports are easier to read, easier to present, and easier to sell. A dashboard with no gaps feels like a dashboard that knows things.

What is wrong with that

A modelled number and a measured number look identical once they are in the same column. You cannot tell, looking at a report, which figures were observed and which were allocated by an assumption you never saw and cannot inspect.

That matters because the two justify different decisions. "This campaign produced eleven paying customers, verified" supports cancelling something else to make more like it. "This campaign was allocated eleven customers by a model" supports considerably less, and you should know which one you are holding when you make the call.

There is a second-order problem too. Once a model fills the gap, nobody has any incentive to close it. The report looks complete, so the fact that most of your traffic is arriving untagged never surfaces as a thing to fix.

What we do instead

Every campaign carries a confidence grade, and the grade is capped by the evidence rather than by optimism. There are four, and they mean exactly what they say:

• Verified. A tracked link was followed, or your product reported an event naming this campaign. Something happened on your side that points back at this content.
• Strong estimate. The post was matched and its figures were read, but nobody arrived through a link carrying its code.
• Possible influence. The post exists and that is all we know.
• Insufficient data. Not even that.

The rule that does the work is the one about the top grade: nothing reaches verified without a tracked link or a named event. A rise that lines up beautifully in time is possible influence, however convincing the chart looks, and it is never promoted for looking convincing.

Why a grade rather than a number

Because the grade survives being read quickly. A number carries no memory of how it was produced, and by the time it reaches a slide nobody remembers which half was measured. A campaign labelled possible influence cannot be mistaken for a measured result, even by someone who was not there when it was recorded.

It is less satisfying than a full report. A campaign sitting at possible influence reads as an admission, and we have had to resist the urge to make that rarer by being cleverer rather than by helping people tag more of their links.

But it is useful in its own right. Watching campaigns move up the grades as you code more destinations is the most direct feedback on setup quality we have found, and a model would have hidden it completely.