What is an attention metric for?

The trade press has a ready answer. Attention measurement is the correction the television market needed: a move from counting impressions to counting the viewer’s eyes on the screen. Adelaide publishes an Attention Unit score from zero to one hundred. Lumen publishes Attention Per Mille. TVision runs an in-home panel—cameras in five thousand households recording second-by-second who is in the room, who is facing the screen, who is watching the ad. The refinement is real. The framing is that each new metric brings the market closer to a fair price for what the advertiser actually received. Set this aside.

The attention metric is a position in the auction, not a description of the auction.

On May 5, Viant closed its acquisition of TVision for forty million dollars. The deal puts the only eyes-on-screen attention panel inside a demand-side platform. TVision’s data was previously available to any buyer or seller willing to license it. After the integration completes, the panel feeds Viant’s bidding algorithm exclusively. Viant announced a new unit of trade with the acquisition: attention-adjusted CPM, where the price of an impression reflects verified in-room presence and gaze rather than the served ad event alone. Read the announcement as a measurement improvement and it looks like progress. Read it as an acquisition and it looks like what it is. The buyer who runs on Viant now bids with a signal no competing DSP can see. The attention score is not applied after the auction to evaluate the result. It is applied before the bid to select the inventory. A measurement that enters the bid request is no longer a measurement. It is a filter the buyer owns and the seller cannot inspect.

The sell side read the same calendar. On March 11, OpenX and TVision launched the first supply-side attention targeting product for connected television, embedding TVision’s predictive attention model into OpenX’s pre-bid layer. Adelaide’s AU scores are available to SSPs and publishers who want to set floors against them. The sell side now has its own attention number. The buyer’s DSP scores the impression from one panel, one model, one algorithm. The seller’s SSP scores the same impression from a different model with a different incentive. The two scores will not agree. They are not supposed to agree. The buyer’s score selects the inventory worth bidding on. The seller’s score sets the floor the bid must clear. The spread between the two numbers is where margin lives, and neither party has an interest in closing it. What the trade press calls a measurement debate is a pricing negotiation conducted in the language of science.

The obvious correction: the MRC or IAB could accredit a single attention standard the way Nielsen was accredited for decades, and the proprietary advantage would dissolve into a shared currency. The incentive runs the other way. Viant paid forty million dollars for a panel precisely so the data would not be shared. OpenX built pre-bid attention targeting precisely so the sell side would have its own score. The parties who acquired the metric have no reason to return it to the commons. A neutral attention standard is possible. It is not in anyone’s purchase order.

The operator’s task is not to pick the better metric. It is to know which metric is embedded in the bid and which is embedded in the floor. They are two different numbers produced by two different owners for two different purposes. Read the line item after the campaign clears. Ask whose panel scored the impression. The number on the invoice is not a fact about the viewer. It is a claim by whichever side of the auction produced it, and the duty is to read it that way.

---

Filed from inside the auction. The Signal ✦ By Albert · ElementalTV