Sep 20, 2018 - Technology

Ad industry leans into machine learning

A new study from one of the world’s biggest ad firms, its digital ad agency and a programmatic ad tech company suggests that digital ad campaigns optimized by machine learning tools outperformed campaigns managed by humans over the course of one month.

Data: Magna Global; Chart: Axios Visuals

Data: Magna Global; Chart: Axios Visuals

Why it matters: Although advertising has traditionally been a creative industry, stakeholders — like agencies, ad tech firms, and even brands — are pushing the effectiveness of their automation and machine learning tools, to lure clients that are focused on cost-efficient data-driven ad campaigns.

The study was conducted by:

  • IPG Media Lab, the media advisory arm of Interpublic Group (one of the world's biggest ad holding groups)
  • MAGNA, the investment and innovation unit within IPG Mediabrands
  • and true[X], a programmatic ad tech company.

It suggests that humans and machines need similar exposure to an ad campaign to learn, absorb patterns and predict incomes. Given that same input, the machine-learning programs performed better than humans at optimizing the campaign — basically, putting the most effective ads where they will have the most impact.

  • For example, the average number of ad impressions across all three experiment campaigns that it took to reach a consumer with a human-managed ad campaign was 4.13, while the average number of ad impressions that it took a machine to reach a person was 3.08.
  • Overall, the study found that all three brands achieved higher levels of brand familiarity, brand interest and purchase consideration when their campaigns were optimized by machine learning instead of human campaign managers.

The bigger picture: The integration of automated ad technology (called programmatic advertising) has been proven to improve the outcomes of digital advertising when properly managed by humans, but it will require a new set of skills and jobs to implement.

If machine learning takes over routine tasks — like allocating ad impressions to parts of a website where they will perform better, or swapping an ad's creative elements with more compelling art — humans will only need to step in to solve complex problems and make big decisions.

  • That bodes well for experienced managers, but potentially poorly for people who perform easily automated tasks.
  • The addition of these jobs and resources will likely displace some advertising jobs that are no longer necessary. For example, fewer sales people are needed to perform small-level direct sales jobs that can be done digitally.
"Smart media companies no longer invest as heavily in digital media web site sales because the return is now lower .... Instead they hire programmatic to optimize their sites for machines who want to buy. This in theory frees sellers up to sell "experiences." So yes — web sites are barely mid-funnel marketing tools. They are “bottom funnel,” mostly. 
— Ryan McConville, President & COO, Kargo, a mobile advertising company

Yes, but: Almost every new ad tech firm, marketing agency and tech company is touting awesome machine learning products, but right now none of them performs wildly better than the human-built systems we have today.

Bottom line: Advertising and marketing jobs are increasingly becoming more automated, but it will take time and experimentation to best understand what the balance should be between automation and human engagement.

Go deeper: How media companies lost the advertising game

Go deeper