Attribution when there is no click
When an assistant describes you and the person never visits, no attribution model will find it. The workable response is to stop trying to allocate the outcome and start measuring what arriving buyers already knew, which is qualitative, cheap and more persuasive than a modelled figure.
Attribution has always been a simplification that everybody agreed to pretend was a measurement. It worked while most influence left a trace. A growing share of influence now happens inside a conversation with an assistant, and leaves none at all.
The shape of the problem
A buyer asks an assistant who does a particular kind of work in their market. Four companies are described. They read the descriptions, form a view, and three weeks later type one company's name into a browser. That session is recorded as direct traffic, and the moment that decided it is invisible.
This is not new in kind — word of mouth and private messages have always done this — but it is new in scale, and it is happening to exactly the top-of-funnel activity that attribution models were used to justify.
The channel that decided it will be recorded as direct. It always was.
Three responses, in order of usefulness
One required field on the first call or the enquiry form: how did you come across us, and what did you already know? It is unfashionable, it is the most accurate instrument available for this, and it takes a week to implement.
Direct sessions and branded searches rising while nothing else changed is what influence looks like from the inside. It is circumstantial and it is real.
Ask your sales people to note, for each new conversation, whether the buyer arrived already understanding your method, your positioning or your pricing. A rising share of pre-informed buyers is the outcome this work produces.
What not to do
- Invent a multiplier. Assigning an estimated value per assistant mention puts a made-up number into a model that then gets treated as real.
- Attribute all direct traffic to AI visibility. Direct traffic has a dozen causes and always did.
- Abandon measurement entirely. "It cannot be measured" is how budgets get cut.
- Buy a tool that reports a confident figure for something nobody can observe, and then present it to the board.
How to put it to a finance director
Not as an attribution argument, because you will lose it. As a portfolio one: some spend is measurable and some is not, this has always been true of brand, and the response to an unmeasurable channel is to bound it rather than to zero it.
| Question | Honest answer |
|---|---|
| How many leads did this produce? | Unknown. Here is the share of new conversations that started already informed, and how it has moved |
| What is the return? | Not separable. Here is the cost, which is small, and here is what would be lost by stopping |
| How do we know it works? | Assistants name us on questions where they did not a year ago. Here are the recorded answers |
| When will we know more? | The reading improves as the series lengthens. It will not become a click |
Questions this settles
Partly. Some assistant visits carry a referrer and can be segmented; many do not and land as direct. Any count built this way is a floor.
You cannot, because you do not control the link. What you can do is watch the referrers that do arrive and treat them as a sample.
Individually no, in aggregate and over time yes. It is the standard instrument for measuring influence that leaves no trace, and it outperforms a model built on data that does not exist.