POV by Hannah Pavalow, ThinkMedium Client Advisor

In our previous post, we discussed the ARF AUDIENCExSCIENCE coverage of cross-media currency efforts; this post will focus on attention.

Day 2 of the conference had a dedicated Attention track, along with two mainstage presentations on the topic. While “Attention” has long-been a focus of the ARF, the industry’s renewed interest has revitalized the discussion. The research track had excellent presentations exploring biometric measures of attention, but it wasn’t until the mainstage presentation that two points of contention were raised, unfortunately without time for resolution. Similar to the currency conversation, the day’s content went deep on a specific dimension – biometrics measures – and in this post, we’ll focus first on the context necessary to understand the broader landscape and then on implications to industry and consumers.

SOME CONTEXT
Attention has long been an area of academic research, with numerous industry attempts to commercialize it. During the early 2000s, the measure succeeded in creative pre-testing via biometric methods like facial coding and neuroscience measures. While there were rumors of attempts at scaled attention metrics in the late 2010s, the remnants of which are scattered over the Internet, no product materialized, and the rise of viewability measures quickly became the preferred quality measure, while direct sales signals dominated optimization.

However, three things have changed in the past 2-3 years, causing attention to gain renewed…attention:

1. The loss of sales signals for optimization caused by privacy restrictions, leading to a decrease in performance for direct response advertising, and a gap for a proxy-based measure.
2. Recognizing that viewability is not enough; this has led to the current trend of combining proxy signals to create a new metric.
3. TV stakeholders pushing on metric standards. TV networks have long used biometrics as performance indicators for content and as a way to differentiate their medium from their digital counterparts. For them, attention is not a new metric, but their recent participation in driving cross-channel measurement has motivated discussion of a biometric-based method as a quality measure.

THE WHAT AND HOW
With the above context, the ads and media ecosystem needs to align on two issues concerning attention: (1) What is it? And (2) How can it be used? Without some agreement on whether attention within the industry will be a biometric-based metric or an onsite behavioral metric (i.e., the inputs), there will be no consistency in the results (outputs) of this measure, leading to an inability to use it across channels. The second decision is the role of attention within the ecosystem, assuming that attention becomes a scaled measure, should it be applied to optimization? Despite significant demand for attention optimization from the buy side, without specific precautions, the optimization may not drive sales and could negatively impact consumer safety.

Let’s first cover the two flavors: biometrics and proxies. While the ARF’s attention track focused almost exclusively on biometric methodologies, the ARF’s Attention Measurement Validation Initiative explicitly called out the two methodological approaches. Even the attention panel at the event, which was nearly all representatives of vendors that use biometric measures, recognized there is confusion in the marketplace between academic measures of attention (which are physiologically based) and the demand for a behavioral-based sales proxy. Bill Harvey, an industry leader, concluded, “Let’s not call it attention but rather impression quality. If we’re trying to predict sales, let’s find what best predicts sales.”

When comparing the two methods, the biometric method is still struggling with scaling, though some bio-based vendors are incorporating machine learning to try and overcome that, as well as disagreements in the exact definitions and methods to use. In contrast, the predominantly proxy-based vendors, led by third-party solutions like IAS (which does incorporate eye-tracking as a sub-component) and DoubleVerify, have newer attention metrics, and are still building out proof points and refining their methodologies. Ultimately neither method is a clear winner at this point, and the customers for the two are largely different (TV vs. digital), meaning that we will see both versions in the market, with different media types using each, leading to fragmentation and a failure of attention as a cross-channel currency.

THE DEBATE
Regarding the second question of the role of attention as a pre-bid solution or optimization, the division on this is aligned with the division in the methodology above. The biometric vendors agreed that while attention is a valid effectiveness measure (the ad achieving a desired goal), it should not be used for optimization. There are several reasons for this:

1. Research shows that biometric attention (for lack of a better differentiator between the approaches) is impacted most significantly by context and environment, even when the business outcome is constant. An ad that achieves the same level of brand recognition in-feed vs. in-stream will have wildly different attention scores because consumer behavior is inherently different by format and activity. Sophie MacIntrye of Meta and Bill Harvey Consulting presented a study examining attention across devices and platforms, demonstrating this effect. The implication is that baseline biometric attention is not a meaningful predictive measure to compare outcomes by device or platform where the ultimate format experiences differ.
2. Consistent with all biometric attention methods is the need for user-level normalization to get meaningful results. People’s baseline physiologies are wildly different, and you can’t compare raw biometric data from one person to another and expect to gain insight into the stimulus (e.g., ad being tested). Instead, you will most likely pick up underlying differences between people. Marc Guldimann of Adelaide, one of the biometric-based vendors with a scaled measure, noted, “When you take this simplistic measure of attention time or duration in the wild, you see that a metric that is good as analysis, fails when you start to optimize it,” referencing research that showed older or intoxicated people pay attention to ads longer. Without user level inputs, or at least normalizing all inputs to as granular a level as possible, this measure fails to be predictive of the stimuli’s impact and instead is optimizing for user behaviors that are not tied to business outcomes (like Click-Through Rate optimization delivering “clicky” people or Video View optimization delivering “viewy” people). Guldimann further referenced Goodhart’s Law: “When a measure becomes a target, it ceases to be a good measure.”
3. The most significant risk with the industry’s shift toward attention is the question of consumer safety. Assuming that the industry can create a scalable attention metric with which to optimize without building the metric to be as close to an individual causal model as possible, we are motivating the proliferation of clickbait and sensationalist content. When we look at aggregated proxy metrics for engagement, we see that the more extreme the content, the higher the probability of engagement; this sets a dangerous model for the quality of ads and, in the case of publisher optimization, content. Any pre-bid solution or optimization ought to provide transparency into their models and explicitly address these concerns before going to market.

While the biometric attention vendors aligned on the overarching point of view that attention is not for optimization, the lack of pre-bid attention vendors on stage made it hard to know how they are addressing these issues. While none of the above are impossible to handle, they require explicit consideration within the products, something that the proxy-based attention vendors have yet to discuss publicly. This makes the ARF’s Attention Measurement Validation Initiative so important, as it can begin to unpack some of the thornier issues surrounding this controversial, multi-faceted metric.