By Mike Follett, CEO at Lumen Research
Psychologists tell us that attention is driven by two quite different sets of factors:
- Bottom Up (stimulus-driven, automatic, about the world around us)
- Top Down (goal-driven, deliberate, about the aims and purposes of people)
We can parse the ground truth data from Lumen Research’s years of eye tracking to isolate the contribution of different factors, understand how they interact with each other, and so predict attention.
And the existing ad tech ecosystem already provides us with lots of data signals for each of these factors – ready for co-option or ‘exaption’.
Bottom Up factors include:
- Prominence of stimulus: (e.g. size of ad, % of pixels on screen) = Integral Ad Science
- Permanence of stimulus: (e.g. time in view) Time in view = Integral Ad Science (again)
- Context of stimulus: (e.g. clutter and competing messages) = sincera.io or Seedtag
- Creative design: (e.g. visual elements like edges or straight lines that ‘automatically’ get attention) = Lumen Research’s SPOTLIGHT tests
Top Down factors include:
- Audience demographic or intent (e.g. Experian, LiveRamp)
- Prior purchase history (e.g.Amazon, Tesco Media Network)
- Prior brand familiarity (e.g. Kantar)
- Prior brand exposure (e.g. Brand Metrics, Audience Project)
- Creative relevance e.g. (the meaningfulness of the ads drives attention, as measured by Lumen’s SPOTLIGHT tests again)
If you have the core ground truth attention data, then all these Top Down and Bottom Up signals can be combined and weighted and hey presto: accurate attention predictions that can be mapped to increases in memory or money.
But here’s the problem: Creative falls into both categories.
The same visual cue can be a Bottom Up attention trigger while also being a Top Down reinforcement of meaning and relevance.
It’s like the classic duck/rabbit visual illusion: it’s not one, or the other. It’s inescapably BOTH AT THE SAME TIME.
What does this mean for predicting attention?
- Embrace the ambiguity. The overlapping, polyvalent nature of the different factors means that they will never ‘add up to 100%’. An absolutely perfect prediction of attention is impossible.
- Good enough to get going. PwC’s analysis on an early, ‘Bottom Up’-only version of the Lumen model suggested that it predicted visual attention with 70% accuracy – with the remaining 30% of the variation in attention accounted for by Top Down factors. This is a good start – but there’s plenty of room for improvement by adding extra factors.
- Mess with the mix. Lumen already partners with both ‘Top Down’ and ‘Bottom Up’ partners – but there are many more signals out there. Get in touch if you think we can collaborate.