The identifier that never reaches checkout

You tag the visit, the person buys, and the sale arrives with no marker. We measured two different causes in real accounts — and both are fixed by changing one thing in how the link is assembled.

Anyone selling through a third party checkout knows the problem: the visit is tracked properly, the person buys, and the sale reaches your system with no clue where it came from. The path exists — most checkouts accept a free tracking field that comes back in the webhook — but it fails silently in two ways.

Both were measured in real accounts, and neither shows up in any report. You only find them by counting.

Cause 1: the field is truncated, and your marker sits at the end

In an account with 2,819 approved sales over thirty days, the session marker appeared in 1,708 of them. That looked reasonable, until we read the contents: only 753 carried the full identifier. Another 786 arrived with exactly 34 characters of something that should have 36.

The detail that closes the diagnosis: in those 786, the tracking field had length 283 — the same value as both minimum and maximum. A fixed length across hundreds of rows is not a coincidence, it is a cap. The checkout truncates the field, and whatever sat at the end dies.

This is not random loss. It is positional loss: whatever is at the end of the string is always the first thing cut.

The typical tracking template assembles UTMs first and the session identifier last, because it was the last thing added. With long campaign names, the cap arrives before the end — and more than half of the markers that were present arrived unreadable.

The fix

Move the session identifier to the front of the field. It is one line in how the checkout link is assembled, and it changes who gets sacrificed by the cut: the tail of the UTMs, which you already have by other means, instead of the key that ties the sale to the visit.

Cause 2: the marker was never added

On the same platform, another campaign from the same group had 145 approved sales, all with the tracking field filled, and zero with the session marker. The longest field was 250 characters — comfortably under the cap.

There was no truncation here. The template simply never included the identifier: it assembled conversion, campaign, content, source, medium, term and page path, and stopped there.

The practical lesson is that the two causes need different diagnoses, and confusing them costs weeks. The discriminator is length: if the rows without a marker all hit the same length, it is truncation; if they are short and varied, it is absence.

The check that separates them in one query

SELECT
  position(tracking, 'cl_ussid') > 0   AS has_marker,
  count()                               AS sales,
  min(length(tracking))                 AS shortest,
  max(length(tracking))                 AS longest
FROM approved_sales
WHERE day >= today() - 30
GROUP BY has_marker

Shortest equal to longest on the rows without a marker is the signature of a cap. Varied lengths, all below a known limit, is the signature of absence.

Why this matters more than it looks

There is an alternative attribution path that does not depend on the marker: matching the sale to the person by the buyer's email, against the identity graph. It works, and it is what holds the result together when the marker fails.

But it has a ceiling of its own: it only reaches people whose email was already known before the purchase. In one of the accounts we measured, 92.4% of buyers existed as a known person in the graph — a good number, and still not 100%. The marker is the deterministic path; email is the safety net.

• Marker present and intact: direct attribution, no ambiguity.
• Marker missing and email known: identity attribution, very good and not perfect.
• Neither: the sale lands in your revenue and in no origin report at all.

The third line is the one that disappears from discussions, because it does not look like an error: the revenue is right there. Only the origin is gone.

In CrazyLeads the session marker is injected into checkout links by the pixel itself, and attribution uses both paths — deterministic first, identity as the net — with a count of which one explained each sale visible on screen.