A 58% lift in ROAS. A 62% jump in click-through rates. Those are the numbers Adelaide reported from beta tests of its attention-based pre-bid segments in The Trade Desk. Now the same capability is expanding across the DSP landscape, and Amazon's integration changes the calculus for any brand running programmatic through retail media.
The shift matters because it addresses a problem that has quietly drained programmatic budgets for years: viewability metrics that reward inventory designed to game them rather than inventory that actually drives outcomes.
The Viewability Problem Nobody Wants to Model
Viewability was supposed to be the quality signal. An ad that meets the IAB standard (50% of pixels visible for one second on display, two seconds on video) should, in theory, represent a real opportunity to influence a buyer. In practice, Lumen Research found that 70% of viewable ads are never actually seen by a human. The ad loaded. The pixels rendered. Nobody looked.
Made-for-advertising sites exploit this gap systematically. According to 24metrics, MFA sites accounted for 21% of all programmatic ad impressions and consumed 15% of total ad spend in the ANA's transparency study. These sites are engineered to pass viewability checks while delivering zero meaningful attention. They stack ad units, auto-refresh placements, and force users through slideshow formats that generate impressions without engagement.
The financial model is simple: buy cheap traffic from content recommendation widgets, serve it pages stuffed with ad placements, pocket the arbitrage. HUMAN Security notes that MFA sites often achieve high viewability scores precisely because they're designed to do so, tricking algorithms into treating them as quality inventory.
For CMOs trying to defend programmatic spend to a CFO, this creates an uncomfortable reality. Your verification reports may show strong viewability numbers while your actual business outcomes lag. The correlation between what you're paying for and what you're getting has broken down.
How Adelaide's AU Metric Changes the Bid
Adelaide's Attention Unit (AU) metric attempts to solve this by predicting whether a placement will actually capture attention, not just whether it will technically render. As AdExchanger reported, AU scores are generated by a machine learning model that incorporates eye-tracking data, device signals, and full-funnel outcome data. The result is a 0-100 score reflecting a placement's probability of attention and impact.
The pre-bid integration means buyers can now filter inventory before the auction rather than measuring attention after the campaign runs. Select high, average, or low AU segments, and the DSP auto-optimizes toward placements that meet your threshold. No more manual optimization based on post-campaign site-level reports.
Haleon's test within Amazon DSP demonstrated what this looks like in practice. High-AU CTV placements drove 37% higher favorability, 19% better ad recall, and 9% higher purchase intent compared to low-AU inventory. On display, high-attention placements led to conversions 34% more often at a 25% lower cost per conversion.
Those aren't marginal improvements. They're the kind of deltas that change how you allocate budget across channels.
The Amazon DSP Angle
Amazon's DSP integration matters for a specific reason: retail media networks have first-party purchase data that other DSPs lack. When you can correlate attention scores with actual conversion events tied to shopping behavior, you get a tighter feedback loop between media quality and business outcomes.
Phil Jackson, Haleon's director of global digital marketing effectiveness innovation, framed the goal clearly: prove that attention leads to better campaign performance, then use that evidence to base buying decisions on attention rather than viewability alone.
The Trade Desk integration, launched in October 2024, pushed attention-based targeting into the largest independent DSP. Yahoo DSP added the capability in June 2023. Amazon's participation extends the same logic to retail media, where the stakes are higher because the path from impression to purchase is shorter and more measurable.
What This Means for Your Media Model
If you're running programmatic through Amazon DSP, the immediate question is whether to activate attention-based segments and at what threshold. The answer depends on your current baseline and your tolerance for reduced reach.
High-AU segments will shrink your available inventory pool. That's the point. You're trading volume for quality, betting that fewer impressions against higher-attention placements will outperform more impressions against mixed-quality inventory. The Haleon data suggests the trade-off is favorable, but your category, creative, and audience may behave differently.
The right approach is a controlled test. Run a holdout group against your current targeting, activate high-AU segments for the test group, and measure the delta on your actual KPIs: ROAS, cost per conversion, brand lift if you're running upper-funnel. Two to three weeks should give you enough signal to make a decision.
The IAB and MRC released standardized Attention Measurement Guidelines in November 2025, which provides a framework for evaluating attention vendors and comparing methodologies. If you're considering multiple attention providers, the guidelines offer a baseline for apples-to-apples comparison.
The CFO Conversation
Attention metrics don't replace your existing measurement stack. They add a layer of media quality intelligence that helps explain why two campaigns with similar viewability scores can produce wildly different outcomes.
For the CFO conversation, frame it this way: attention-based targeting is a waste-reduction mechanism. You're not buying a new capability; you're eliminating spend on inventory that was never going to drive results. The 25% reduction in cost per conversion that Haleon saw on high-AU display inventory is money that was previously going to placements that met viewability thresholds but failed to capture actual attention.
The sensitivity analysis is straightforward. Model your current programmatic spend, estimate the percentage going to low-attention inventory (industry benchmarks suggest 20-30% of impressions are effectively waste), and calculate the reallocation opportunity. Even conservative assumptions produce meaningful numbers.
Risks and Mitigations
Three risks to model before activation:
First, reach compression. High-AU segments exclude inventory, which may limit your ability to hit frequency targets or reach niche audiences. Monitor reach and frequency metrics alongside performance KPIs.
Second, CPM inflation. If attention-based targeting becomes widespread, high-AU inventory will command premium pricing. The current arbitrage opportunity exists because attention metrics aren't yet priced into the market. That window will close.
Third, measurement fragmentation. Adelaide's AU is one methodology among several. Lumen, IAS, and others use different approaches. EMARKETER notes that the IAB guidelines help standardize evaluation, but cross-vendor comparisons remain imperfect.
The mitigation for all three is the same: run controlled tests, document your assumptions, and build internal benchmarks before scaling. Attention targeting is a lever, not a strategy. Use it to improve media quality within a broader framework that ties spend to outcomes your CFO cares about.
