Using machine learning in outdoor for good
It’s now 2 years since the Sustainable Development Goals (SDGs or “Global Goals”) were launched. As they are basically the 17 top priorities for the planet, how do we ensure that we keep them on the agenda of the business world as well as the wider public?
As part of our commitment to Common Ground — the advertising and marketing industry’s collaborative initiative to tackle the Global Goals — we wanted to make a contribution to Global Goals Week: the third week in September during which the United Nations General Assembly meets to get an update on progress. We wanted to raise the awareness of the Global Goals but also showcase the power of advertising: in this case, we showcased new digital out-of-home measurement technology that uses machine learning to test and tailor the most effective advertising for an audience.
Dentsu Aegis Network, in support of Project Everyone and the Bill & Melinda Gates Foundation, brought together a coalition of businesses to amplify the Global Goals during Global Goals Week in 2017. This coalition included Perfect Day — an award-winning creative agency approached to deliver the collateral — and Quividi, an agency that measures audience engagement within digital out-of-home advertising (DOOH).
The campaign focused on the “goalkeepers” — the celebrities, artists, state and business leaders who are achieving exceptional progress on the Global Goals — and encouraged everyone to become a goalkeeper themselves. A variety of pro bono partners, including agency Posterscope, helped to reach millions of people with the key message, but our project attempted something that hadn’t been done before.
Quividi proposed to amplify the natural effectiveness of DOOH advertising by using an AI-driven process which dynamically assembles ads using a genetic algorithm that optimizes creatives for maximum attention time. This tool is called Campaign Genius and this is the very first time that this tool was deployed on a real-time, evolutionary DOOH campaigns on a large scale.
One of Quividi’s important partners, Westfield, kindly donated screen time to the campaign and it ran from 18–24 September in 3 of its shopping centres. The ad that was played was a 10-second video composed of 4 clips of 2.5 second each as developed by Project Everyone. The first 3 clips were Goalkeepers selected and sequenced by Campaign Genius and the last clip was a common branded call-to-action outro. To the audience, this looked like a regular DOOH campaign. Underneath the hood, however, the local attention dynamics as measured by Quividi were being fed to the genetic algorithm. The algorithm recombined the sequences that fetched the most attention together while discarding the rest allowing for the most optimized sequences of #goalkeepers17 messaging for each particular screen, its environment, and its audience.
Our campaign delivered some interesting insights in testing this new technology. Firstly, some combinations drew more attention and tended to persist across many iterations. Other combinations worked better depending on the order in which they were displayed. Secondly, messaging appearing on successive warm colour backgrounds drew more attention. Thirdly, the location that optimized the best had a massive 32% increase in attention time (from 4.2 seconds to 5.5 seconds of attention) between the beginning and the end of the one-week campaign. And lastly, in most cases, the optimized recombination of content pushed the attention time up 20 % over the course of the campaign, from 2.3 to 2.7 seconds. These results are helpful for any subsequent campaign, both paid-for and pro-bono, which uses digital out-of-home as a medium.
