Playground XYZ won the ‘Most Effective use or Data or Insights’ at The Drum Awards for Marketing 2021 with its campaign for Under Armour. Here, the team behind the winning entry reveal the secrets of this successful project…
Originally posted on The Drum, 9th July 2021.
Playground xyz won the ‘Most Effective use of Data or Insights’ at The Drum Awards for Marketing 2021 with its campaign for Under Armour. Here, the team behind the winning entry reveal the secrets of this successful project…
The challenge
In partnership with Digitas and Under Armour, Playground xyz was tasked with driving awareness of the brand’s cold gear apparel/clothing range.
Media consumption has reached an all-time high. We’re hooked on a proliferation of screens feeding us an unprecedented amount of content. Accompanying this content are advertisements. Lots of them. Forbes reported that people see up to 10,000 a day. But how many are paid attention to?
And what is attention? In digital advertising, metrics like viewability measure whether an ad made it onto the screen. The problem is that an ad can be on the screen, and not actually be looked at.
It’s obvious each person needs the right length video based on their attention span, ensuring that the entire message is delivered no matter the length of attention captured, without losing the crucial brand narrative and its all-important call to action.
We dreamt that we could break free of panel-based approaches and instead use AI to build a scaled attention measurement product that works across all platforms.
The strategy
The first step in our journey was data collection to build and train an eye gaze model. To do so, our machine learning engineers applied a deep convolutional neural network (CNN) architecture to a training set of 3,000+ opt-in, crowd-sourced users who played a purpose-built mobile game and consented to their face images being captured and analysed.
In total we collected 754,600 photos with an associated known point of gaze. Through iteratively adapting the model’s layer weights to maximize model accuracy (known as convolution), the model developed an implicit understanding of the non-linear relationships between the photographic features captured by the camera and the user’s estimated point of gaze.
The eye gaze model was then integrated with the web browsing app that was used for market research tasks and to ensure that all data and creative delivery is inherently cross platform. It is worth noting that the technology was built so that there is no longer the need to take images of participants’ eyes off the device, instead now only sending the X/Y coordinates to our servers, an important feature in today’s privacy conscious landscape.
We recruited 7,000 panelists, making it the world’s biggest eye tracking research project, to our opt-in app and had them browse around websites. When their eye gaze intersected with ads we captured their Attention Time. We also capture 40 additional data signals that occur alongside looking – scroll velocity, page length, and many, many more.
We fed 70,000 sessions of this data to a convolutional neural network – a machine learning framework that finds probabilistic patterns. Our specific goal: we wanted to know if it could help us understand if users were looking at the ad… without using eye tracking.
Why is this so important? If we can tell how long someone looked at an ad without eye tracking, we can remove the shackles of panel-based approaches and apply this measurement to any impression on the web. It’s privacy-compliant, doesn’t use cookies and above all else, is scalable.
The solution
Playground xyz’s revolutionary Video Switch product, powered by our Attention Intelligence Platform (AIP) offers our clients the ability to do a number of things, in their advertising efforts, that have never before been possible.
The first is measurement and reporting. Using our AI, we can tell our clients which of their ads, placements and partners get the most attention – and which don’t. This is rapidly turning Attention Time into a valuable metric and currency by which their media is assessed.
Next is media optimisation. We trained our AI to quickly learn how a brand’s creative is performing across different articles – scoring and ranking each page for its Attention Time. What’s more, it now uses historical data to predict which future pages might be the best fit for a particular brand and/or campaign.
We feed this data into the client’s DSP and they watch their ads automatically appear on the pages where they’re actually getting looked at more – this delivers valuable extra seconds of looking that drive their brand forward.
Lastly, there’s creative optimisation. We asked ourselves – how would storytelling change if our AI could predict how long the user would look at the ad? The answer was simple – tailor the creative based on the user’s attention appetite. Every user can now get the right length story suited to their attention span. It’s another game changer in a year that’s been full of them.
The Most Effective Use of Data or Insights Award-winning results
In a whitepaper we published with Kantar, we proved that Attention Time was 7.5-times more powerful at driving Awareness than Viewability and 5.9-times stronger at driving brand recall. What’s more, for each extra second a user looks, awareness and recall went up an additional 11% and 7% respectively.
To action perfect storytelling we needed to know how long consumers are looking at different ads to establish Attention Time. And, we need to know this before we choose what length storytelling video to serve to them.
Once we know this, our AIP feeds this signal to our ground-breaking product, the Video Switch, which chooses the right length video to serve to each user. Small attention spans get the shortest video, medium get a little more, and those with the highest attention span get the longest. We call it Dynamic Creative Optimisation for Attention (DCO-A). DCO-A fundamentally changes the way our clients both develop and serve creative.
DCO-A was used by Digitas across Under Armour’s cold weather clothing campaign through the Video Switch placement, allowing real time optimisation via our AIP to constantly analyse and improve on the campaign results.
Dynamically serving the right video length, at the right time, to the right user meant that results far exceeded expectations. For Under Armour, this delivered a 93% video completion rate across their campaign, more than 11 seconds of time in view and a 0.2% click through rate, all metrics which ultimately worked to deliver significant increases in business outcomes for the brand.
The Attention Time delivered through this campaign was an 11-times increase on what the brand could have expected to receive if not using the Playground xyz platform. Playground xyz’s Attention Intelligence Platform tracks delivery against a control group in order to measure the Attention uplift per campaign.
Through AIP, we were also able to pinpoint where target audiences were most engaged with the Under Armour ads; during the working day. Media optimisation through AIP also ensured the results were improved upon each and every week throughout the six week campaign, serving the ads in the best performing contexts in real time.
