Incrementality: The future of attribution in retail marketing
Just joining us? This is part 2 of a two-part series on media intelligence for attribution in retail marketing. We recommend starting with our brief history of attribution here.
For commerce marketing, 2022 was a great reset.
The arrival of a privacy opt-out on Apple’s iOS 14.5 created a chain reaction across ecommerce. The roaring attribution engine that powered Facebook and Instagram’s advertising engine was suddenly slowed. As result, direct-to-consumer brands that grew by using these tools scrambled to experiment on new platforms from new social channels to new retail media networks. In the fallout, the playbook that brought growth through a combination of Facebook marketing and Shopify stores was suddenly outdated.
At some point, Google will finally enact the deprecation of third-party cookies and, while Meta is making adjustments to introduce new advertising tools that comply with Apple’s policies, there's no going back to the old era. Advertising technology won’t stop pushing forward.
As we race toward the future, it’s important to take a step back and remember an important thing about tools.
They’re only effective when they solve the right problem.
In the last piece of our series, we covered the history of attribution, which set out to assign credit to certain advertising for a sale and enable better planning as a result. We determined those Coca-cola polar bears don’t just roam the media waves for fun - they have a real brand value for Coca-cola - but the question we left off with in our last installment was just “how much?”
Attribution was the problem. Strategies like Media Mix Modeling and Multi-Touch Attribution were built for the paradigm they operated in, whether that was TV ads or the internet.
But those models never provided a true solution.
That’s because they asked the wrong question: How do I optimize advertising?
They never asked how they wanted to optimize their business.
In an era where the ways to reach and sell to customers proliferate, it’s more necessary than ever to think about the business as a whole instead of a single campaign.
There is a need for a multilayered media model that takes into account not only how campaigns performed across a variety of channels and tactics (those Cocal-cola display ads), but also how a promotional action in one part of the chain impacted all of the others (that Coca-cola polar bear pre-video trailer or direct mailer). Brands would also like to know - as you head to the store - where there are opportunities to realize new sales opportunities that weren’t visible at the point of sale.
Meet incrementality measurement
This need has brought focus on a model that paints a more complete picture: incrementality. Incrementality measures the impact of a campaign by considering two sides of a coin: What actions would have taken place with marketing, and what happens without it? This provides an opportunity to analyze the lift that a marketing campaign provides, showing not only where it is making an impact in one channel, but whether budget that is used there may be best spent on another channel.
Why now?
Technology and marketing strategies of today must be built for a different, privacy-first era. This is already evident with the explosion of retail media, which allows advertisers to leverage first-party data collected from retailers’ purchases and loyalty programs. With this, tools that aggregate data from multiple sources while protecting privacy and proprietary information, such as data clean rooms, will become more powerful in reaching across walled gardens.
There are echoes here of the probabilistic approach used by mixed media modeling (MMM) and lift studies, but the pace of decision-making is different now. Thanks to advances in analysis with machine learning and compute speed with the cloud, it is now possible to reduce what used to take a month with MMM down to minutes. Brand marketers need a better understanding of their data and how to take action at the speed of commerce.
It would be a mistake to assume that any modeling change is only required because of Apple’s policy decision, or that last-touch attribution was perfect. Instead, it’s necessary to go back to the reason that new attribution models were explored in the first place. The problem was that all of the measurement was stuck at the bottom of the funnel, even as the ability to build audiences and retarget helped advertising move up closer to the top. Add in the fact that customer acquisition costs were rising even before iOS 14.5 arrived, and it all points toward the idea that change was coming anyway.
After all, a marketer’s job isn't to hit metrics. It's to drive her business. To do that, she must figure out the best way to judge the impact and effectiveness of an entire strategy, considering not just what’s right for individual channels, but how they interact with each other…
Let’s return to the story of Coca-Cola’s polar bear. On a typical day, a consumer may have interacted with Coca-cola’s marketing multiple times. Imagine - a woman wakes up to check the news and sees an in-feed Coca-cola ad. She heads to work on the train listening to her favorite cooking podcast where she hears an advertisement for Coca-cola’s latest launch. At lunch, she watches a short clip her sister sent her on Youtube, bookended by ads - one of which is by Coca-cola. Finally, at the end of the day, she stops by the grocery store on the way home and spies Coke’s Polarbear-themed endcap but forgets her items in-store. Her children will be disappointed so at home, she checks her daily Walmart coupon email and sees a discount for Coca-cola's holiday Polarbear-themed twelve-pack and places an order.
Broadly, the key strengths that created the Facebook conversion machine were the platform’s ability to understand the attributes and motivations of the people on the platform, and how advertising in one place led to action in another. There may not be a way to recreate this deterministically, but, given the limitations of last-touch attribution, there never was. Instead, we must look further to the past. The new focus on incrementality lends itself more to the probabilistic models that powered marketing mix modeling while still being able to model external factors. Comparing actions that were taken with and without marketing heightens our understanding of how people are behaving in response to a particular campaign outside the context of just that platform. In turn, the ability to analyze campaigns across multiple channels can help a marketer understand where their dollars are best spent and allocate budgets accordingly.
That’s how brands can optimize their business, and not just campaigns.
Brands that embrace these modeling advances will be best positioned to navigate an era of commerce that brings about a growing number of retailing and marketing options. While there may not be an exact replacement for Facebook’s engine at the height of its efficiency, the rise of retail media signals a world in which advertising and retail sales are taking place on the same platform, creating new approaches to leverage data for growth. The trick is to understand where these work together, and put it in the larger context of media investments and the business.
You wouldn’t want to give all the credit to the endcap when you can factor in the power of the polar bear. The end goal is to sell a product to a customer, build a relationship that keeps them coming back and make a healthy margin in the process. It will continue to be that way, even if direct attribution becomes but a memory of a bygone era.
