Attention Measurement Theater
Viewability did not solve the accountability problem in digital advertising. So the industry built something that sounds closer to an answer. It isn't.
Attention-based metrics were gaining momentum in DOOH and the pressure to participate was real. So we reached out to an attention measurement vendor. They engaged. And what happened next is the part that does not make it into the case studies.
Before any measurement begins, the supplier has a meaningful opportunity to select the conditions of the test. Which locations. Which screens. Which creative. The vendor provides guidance on what tends to perform well in their methodology. You pick your best assets, your most favorable environments, your strongest creative execution. Then you run the test.
We did not like the first results. So we went back, took the vendor’s feedback on creative selection and location choice, and ran the measurement again. The second time, the numbers were better.
I want to be precise about what happened there. The vendor did not falsify anything. The methodology was applied correctly both times. But the test was not designed to find out whether our DOOH network held attention. It was designed to find a version of our network that scored well enough to justify a partnership. That is a fundamentally different exercise, and it produces a fundamentally different kind of number. The score that came back was real. What it measured was our best case, selected and optimized specifically because it was our best case.
This is not unique to the vendor we worked with or to DOOH. It is the structural reality of any measurement exercise where both parties benefit from a positive outcome. The DOOH company needs good numbers to justify the partnership. The attention vendor needs good numbers to justify the integration. Nobody in that room has an incentive to run a test that might come back unfavorable. So the test does not get designed to come back unfavorable. And the number that emerges from it gets reported as though it reflects the network, not the curated selection that produced it.
That experience is what I mean when I use the phrase attention theater. And it did not start with the test. It started long before, with what attention products actually are.
Here is what attention products actually are.
Most attention metrics are not measurement tools. They are predictive models trained on data collected from opted in eye-tracking panels. A group of people agreed to have their eye movements tracked, typically in a controlled environment, while viewing content and ads. That data trained a model. The model now scores every impression in a live campaign with an attention probability, a number between zero and one that represents the likelihood that a viewer paid attention to the ad.
What you are buying when you buy attention is not a confirmation that attention occurred. It is a prediction, produced by a model that has never seen the specific viewer, watching the specific ad, in the specific context, on the specific device, in the specific moment where your campaign actually ran. The panel was opt-in, which means it skews toward people willing to be observed. The controlled environment means viewing behavior differs from a real consumer watching TV while checking their phone. The model extrapolates from that sample to your entire media buy, and the output looks like measurement.
“It is not measurement. It is a well-constructed estimate dressed in measurement’s clothes.”
Even if the score were perfect, it would still not tell you what you need to know.
Attention is a condition of advertising effectiveness. It is not a measure of it. An ad can hold full visual attention and move nothing. I often test this with my friends while we are out. I point to billboards and talk to them about them, wait 15 minutes and ask them to tell me the 3 ads they saw tonight. The relationship between attention and outcome is real but it is not linear, not uniform across categories, and not something a probability score captures.
When the industry sells attention scores as evidence of performance, it is selling a necessary condition as though it were a repeatable methodical one. That specific substitution is the equivalant of jazz hands. Not a lie, but a claim that has traveled further from the evidence than the language around it admits.
The vendor incentive is not a conspiracy. It is a structure.
A body of research exists showing correlation between high-attention inventory and brand lift or downstream sales outcomes. Some of that research is rigorous. Much of it is funded by attention vendors, conducted on inventory selected to perform well, and reported without the context of what the lower-performing inventory looked like. Correlation between an attention score and a positive outcome is not proof that attention caused the outcome. High-attention inventory also tends to be premium placements with richer creative environments, better content adjacency, and less competitive clutter. Any of those factors could drive the lift. The attention score is present in the room but not necessarily responsible for what happens there.
The research gets used to sell the methodology. The methodology generates scores. The scores get used to justify budget allocation. The budget flows toward high-attention inventory. The high-attention inventory performs well, partially because it was already premium inventory. The attention vendor publishes another case study. The cycle continues, and the original question, whether the attention score caused the performance or merely accompanied it, never gets a clean answer.
Attention is a better question than viewability. It is not the answer. The industry will keep producing better questions dressed as answers until buyers start demanding the one thing the ecosystem is not structured to deliver, a test designed to show the ad did not work.
Business Idea: Imagine starting a measurement company designed with the only goal of measuring all the ad campaigns that did not work for you? Golden!
Curious whether others on the buy side have been through a similar process: have you run attention measurement on your own inventory or campaigns, and did the test design give you confidence the number reflected reality or the best available version of it? Reply or drop a comment.