Attribution is twice as blind to a recommendation as it is to an ad
Every performance marketer has had the same argument. The dashboard says a third of revenue is “direct”. Nobody believes it. Someone suggests brand strength, someone else suggests broken tracking, and the meeting moves on, because there has never been a way to settle it.
There is now. For the Zigpoll Attribution Index we took 3.3 million post-purchase survey responses and put two things side by side that are normally kept apart: what the customer said when asked how they found the brand, and what the store's own analytics recorded about that same order.
The failure is not evenly distributed
Overall, 51.8% of orders carried no usable source. But the interesting number is not the average — it is the rate within each channel. When a customer says they found you a particular way, how often did the analytics know?
| Channel | Untracked | Responses | Stores |
|---|---|---|---|
| YouTube | 60.2% | 133,733 | 1,015 |
| Word of mouth | 58.7% | 490,512 | 2,420 |
| Press / editorial | 56.4% | 22,719 | 249 |
| TikTok | 56.3% | 110,149 | 1,148 |
| Creator / influencer | 54.6% | 33,528 | 264 |
| Podcast | 51.8% | 22,858 | 106 |
| 50.8% | 560,766 | 2,111 | |
| Search | 50.1% | 405,008 | 2,271 |
| 40.1% | 488,516 | 1,706 |
YouTube at 60.2% and word of mouth at 58.7%, against Facebook at 40.1%. A recommendation is roughly one and a half times as likely to vanish from your reporting as a Facebook visit, and YouTube is worse still.
Why the ordering looks like that
This is not a random failure. It is a systematic one, and the logic is simple: attribution captures what arrives with a tag attached and loses what arrives because one person told another.
A paid social click carries its own paperwork. The platform appends identifiers, the pixel fires, the visit is documented from both ends. A friend sending someone a link does not generate any of that. Neither does a mention in a video, a podcast read, or a recommendation made out loud.
A channel's visibility in your reporting is basically a function of how much documentation it attaches to the visit. Which means the channels that look weakest in a last-click report are systematically the ones doing the least trackable work — not the least valuable work.
The customers are not confused
The obvious objection is that untracked customers simply do not remember. They do. Only 8.3% of them chose “other” or “don't remember”, against 7.3% of customers whose source was recorded. That difference is noise.
Whatever is happening in the untracked half, it is not amnesia. People know how they found you. The measurement system has nowhere to put the answer.
What to do about it
There is no clever model that recovers a signal the browser never sent. Modelled attribution and media mix modelling both help, and both are inferences drawn from the same incomplete record.
The one source of evidence that does not depend on the tracking layer is the customer. They were there. Asking them is a crude instrument — people simplify, they forget the first touch, they name the last thing they remember — but it is a genuinely independent one, which is exactly what a measurement stack full of correlated blind spots is short of.
Start with your own direct bucket and ask those customers what they say. If the answer is mostly “I don't remember”, your tracking is fine and the bucket really is brand equity. If it looks like the numbers above, a meaningful share of your budget is being allocated from a report that cannot see the channel doing the work.
Full tables, method and the underlying data: the Zigpoll Attribution Index.
