Half your orders say “direct”. Here is what those customers actually say.
“Direct” is the most comfortable number on an attribution report. It can mean almost anything you want it to mean. Brand strength. Loyal repeat customers. Type-in traffic from people who already know you.
We tested it. Across 3,303,402 post-purchase responses on 3,237 ecommerce stores, 51.8% of orders arrived with no usable source recorded — direct, unknown, or simply absent. For every one of those orders we also have the customer's own answer to “how did you hear about us?”
They are not forgetting
The comfortable assumption is that untracked customers cannot remember. Only 8.3% of them chose “other” or “don't remember”.
For comparison: among customers whose source was tracked, 7.3% said the same thing. The two groups are equally sure of themselves. The difference between them is not what the customer knows. It is what the analytics caught.
What is actually in there
The untracked group skews hard toward channels that produce no click. Compared with customers whose source was recorded, they were:
- 32% more likely to name word of mouth — 19.3% of their answers against 14.6%
- 40% more likely to name YouTube — 5.4% against 3.8%
And the pattern runs the other way for the platforms that stamp an identifier on every visit. Facebook took a substantially larger share of stated answers among customers whose source was tracked than among those whose was not.
So the direct bucket is not a random sample of your customers with the labels missing. It is a biased sample, and the bias runs consistently toward conversations and video.
Why this costs money
Because the untracked group is disproportionately driven by word of mouth, creators and video, the channels most likely to be under-credited in a last-click report are also the ones least able to defend their own budget in a planning meeting.
Spend flows toward whatever can prove itself. That is not the same thing as whatever is working. Optimise hard enough on last-click and you will eventually defund the thing that was filling the funnel — and the dashboard will look fine right up until it doesn't.
In a tightening market the asymmetry gets worse, because the reflex is to cut what cannot evidence itself first.
What “direct” should mean
There is a version of the direct bucket that really is brand equity: returning customers who typed your name in because they already know you. That exists in our data too — returning customers show up in the untracked group at a modestly elevated rate.
But it is a small part of the whole. Treating the entire bucket as brand strength is a decision to not look. The cheapest way to find out which version you have is to ask the customers in it, on the thank-you page, in one question.
Full tables, channel-by-channel rates and method: the Zigpoll Attribution Index. Related: which channels your analytics loses and AI-assisted discovery, up 5.4x in ten months.
