How WPP’s Essence Uses Machine Learning to Improve Media Buying Results

Prior to the change in WPP’s leadership earlier in the year, Essence–a programmatic specialist media agency it acquired in late 2015–has been front-and-center of its plans to create an agency network fit for the contemporary marketplace.

This included the establishment of the media agency, which counts Google as its largest client and expertise with its Google Marketing Platform among its core competencies, as one of four pillars at the center of its turnaround strategy. Ahead of the expected announcement, Adweek caught up with Essence co-founder and chief product officer Andrew Shebbeare to hear more about its data-driven approach to media planning.

At the core of its operations is Olive, a “campaign stewardship system,” which Essence uses to help improve all aspects of its operations—including campaign planning, execution and outcome analysis, which it then uses to perpetuate efficiencies.

Olive also analyzes almost every aspect of campaign performance— from recording which ad-tech partners helped drive the best conversion rates, to how much money is spent with a particular search engine or social media platform.

“You can even link things like people’s time sheets through Olive and then see how that drives the performance of the campaign,” added Shebbeare.

From there, it’s then possible to start “crazy analyses … with meta-data,” which can result in human insights such as the optimal timeline client services teams need to produce the best results.

“You get to see lots of interesting things, like how people who work later get less done in that time, or that their campaigns work less effectively, or that people who get briefed earlier tend to get better results,” he explained. “Our hope is that from that [insight] asset we can derive lots and lots of meta-data that gets us to more predictive and interesting models for performance.”

Dating back to 2016, the GroupM unit has employed a machine-learning approach to improve how it bids on media using programmatic technologies, but more recently these efforts have included the application of AI to further scale the efficiencies it witnessed in early tests.

In 2017 the early tests were delivering demonstrable efficiencies but the process was often “labor intensive” requiring the direct oversight of expensive data scientists, reported Shebbeare.

“We ran a whole bunch of pilots that were deep learning, but they were operated by data scientists hand-building models,” he said. “And now there are literally hundreds of people in the business who are applying machine learning to their campaigns, so that’s been the hockey stick effect.”

However, the application of AI from the beginning of the first quarter of 2018 has enabled Essence to further the impact of these efficiencies, with the agency able to better meet clients’ requirements for KPIs including awareness, purchase intent and actual sales (in terms of volume and order value).

Shebbeare claims these results were achieved using the AI functions along with Olive to better assess impressions quality, i.e. whether an ad is likely to be viewed by a human, and then using software to decide what the likely ROI of an ad impressions is.

“This starts to put the signals of value into the client’s hands and they can begin to make their mind up on how they want to optimize,” he said, adding that different signals have different prominence for each brand.

“I think what’s interesting about these models is that they become more customizable and different brands will end up valuing different attributes of an impression … that potentially opens up the market by making the inventory work a little bit harder for brands.”