Programmatic advertising companies have mainly focused on who to show ads to and when to show them, but until now they have focused very little on what messages to show. Usually, these decisions are limited to:
- A/B testing from a very limited set of alternatives
- Product recommendations from a catalog, usually based on retargeting (retail or travel)
- Machine learning that compares results from a fixed set of creatives to determine which one works best (overall) and when it starts to “burn out” and decrease in effectiveness
Perhaps surprisingly, Facebook and Google AdWords currently provide more opportunities for creative optimization, due to constraints on the creative their native-like formats expect. Title, body, landing page, and sometimes image are the structured fields. By removing arbitrary design creativity, ironically, these formats encourage much more automated experimentation among the individual content elements. Even in these formats, however, it is still uncommon for the content to be individually personalized, unless it is just recommending products based on retargeting.
But what if your marketing platform could predict which messages would have the most impact on each consumer, on an individually personalized basis, and automatically assemble or select those messages? What if such an approach could show lift in results between 2x and 4x versus just using the “best” single creative? And finally, what if it could tell you when there are lots of consumers for whom the best-fit message is not yet available in your library, so you can prioritize new creative briefs for your design team?
I’m convinced that in the future, the strongest predictive marketing platforms will employ this AI-based approach, known as predictive creative.
As with the native formats described above, predictive creative will provide a more structured understanding of the elements that make up a creative message, including the background, colors, imagery, and call to action. Equally important is a similarly structured breakdown of these elements into their attributes that may independently affect the influence of the ad on each consumer.
For example, does the ad show any people? Men, women, or both? How old are they? Does it show a product? Is it in isolation or in use? Is there a call to action?
Which of the following terms describes the ad or its emotional content and impact: happy, funny, calm, exciting, clever, fancy, adventurous, family, aggressive, value, need, safe, trustworthy, quality?
By understanding this much about the creatives they build, marketers have the chance to learn which characteristics drive better performance. And, when coupled with the data available in predictive marketing platforms, machine learning can predict the likely response of each individual to a well-understood ad even more accurately. This expands what is humanly possible, by combining the creativity of marketers to design effective messages with the power of big data and machine learning to individually deliver those messages to their most receptive audience.
The most powerful result, however, is that this kind of data can help direct marketers to create new ads with themes and elements that were missing from their campaign before, without trying to design a million different combinations.
The key to this is leveraging the data we observe about how a customer will respond across many different brands. We’re betting that the kind of data we are capturing here is abstract enough from the details of any brand campaign that most advertisers will be comfortable opting in to sharing this kind of analytics with each other in order to benefit from the aggregated data about customers.
This approach can work equally well with video or display advertising, on desktop, mobile, or social channels. And it is applicable to both brand and direct response goals, as long as there is some way to measure the impact of the campaign on individual consumers, such as watching a video to completion, expressing brand favorability or awareness in a survey, or interacting with an ad, whether or not it generates a click-through. Finally, it can help with both personalization (showing each ad to the right people who will be influenced by them) and contextualization (showing each ad on the right site or app where it will have an increased effect).