American Express has been investing in measurement technology to try and give credit where credit is due.
The increase in regulatory scrutiny, the end of third-party cookies and the paucity of other identifiers means that “what used to work doesn’t work anymore,” said Abhi Juneja, VP of performance marketing and ad tech data science at Amex.
And attribution “has been one of the biggest challenges of all,” he said. (Just ask Meta.)
“Optimizing your marketing budget and investing properly relies on the ability to stitch the user-level journey together across different touch points through to the point of conversion, and then being able to apply logic on top of it,” Juneja said. “It’s hard to do that without access to data.”
Which is why American Express has been moving away from multi-touch attribution (MTA) and embracing marketing mix modeling (MMM) and machine learning to power its approach to measurement.
MMM vs. MTA
A multi-touch attribution model aims to assign credit across all of the different touch points in a consumer journey so a marketer can, in theory, see how much influence each channel had on a sale.
The problem with that technique, Juneja said, is that, without access to tracking data tied to some form of addressable media, MTA “becomes an impossible task, and you have to start making assumptions.”
And so, in late 2019, Amex read the writing on the wall and evolved its approach to look toward MMM and aggregate data as a more reliable way to evaluate the success of its channel mix. MMM analyzes the statistical relationships between the various factors that can influence sales, such as seasonality, promotions and market trends.
Amex had already developed its MMM muscle in order to measure the effect of its brand marketing dollars on channels that don’t always have built-in direct-response mechanisms, such as television and out-of-home.
“We began to apply the same concept to measure the efficacy of digital campaigns, including display ads and social,” Juneja said.
In other words, Amex started experimenting with treating digital as a non-DR environment from a measurement perspective, which you could argue is pretty much becoming the case in light of signal loss.
Although, to be fair, even with signal loss, digital does have a data advantage over channels like TV, Juneja said, which means there’s still a decent amount of data at the aggregate level to feed Amex’s model.
Amex now uses MMM for planning purposes and to inform its budget optimizations by channel.
“The more we did it and the more we got used to it, the more granular we could get with our insights – and the more tactical we could get with our decisions,” Juneja said. “This is really an evolution from last-touch to a more hybrid approach primarily powered by MMM.”
For example, MMM has helped Amex unlock audio. Back when Amex used last-touch, the channel wasn’t even considered from a performance perspective. Because people have to leave an audio environment and head to another channel if they want to actually apply for an Amex product, it rarely received credit.
But after trialing audio as part of a test-and-learn experiment and using MMM to measure the value, Amex realized audio could “work quite well” for acquisition, Juneja said.
American Express is now applying the same approach to test the value of influencer marketing.
Amex has also been able to spend smarter on remarketing, a perennially effective acquisition tactic for the brand. But using MMM, Amex realized part of the reason remarketing looked so good in some cases is because the brand was underestimating the value of its upper-funnel activity.
“Some user segments come to us on their own, which is something the MMM model validated for us,” Juneja said. “And it’s also led to a better customer experience, because people don’t want to be chased around by remarketing if they’ve already made their purchase or are planning to soon.”
No small measure
Although Amex has made a lot of progress on its attribution journey, there’s still work to be done, including running more experiments faster across more channels.
The end results are more worth it, Juneja said. But marketers that want to embark on a similar course of exploration need to be ready to invest the right amount of time and effort both upfront and on an ongoing basis.
For example, Amex is mature enough now to automate parts of the process, which leads to quicker results – but it’s always possible to get even quicker, Juneja said.
“We’re on a path to refine what we’re doing and automate the modeling so we can turn things around faster,” he said.
Amex is only able to think about aspects like speed to value because it took time at the beginning to get its data house in order.
“Aggregating all of the data from different marketing activities can be quite cumbersome, especially for an organization of our size, but that is where we spent a lot of our time initially – getting the right data together and making sure we were validating it with other sources,” Juneja said. “A model is only as good as the data you put in.”