WTF is attribution modeling?

Why did they buy when they did?

Why did they buy it there?

And most importantly, did any of the TV spots, direct mails, banner ads or podcast readouts we bought influence them?

To answer those questions — for their own peace of mind, to satisfy the curiosity of budget-minded finance departments, and to optimize future media investments — marketers have employed a variety of measurement frameworks and statistical models.

Recently, marketing mix modelling (MMM) has come back into vogue. But it’s not the only measurement regime available. In fact, it’s often used by marketers alongside another framework called attribution modeling.

According to the IAB, 88% of marketers already use some form of attribution modeling, while 61% of marketers expect to rely on it more than they did last year, higher than those (56%) expecting to increase their lean on MMM.

OK, so what is it?

On a basic level, attribution modeling takes data collected by an advertiser used to track a customer’s path through their online footprint (a minimum sample might contain 500 sales), and connect it to the various places the brand has been running marketing campaigns.

The end point could be a sale, or it could be some other action — a newsletter signup, say, or a web page visit.

Rhys Jackson, EMEA technical solutions director at media agency PMG, described the model as an attempt to “figure out what did the customers do in the days prior to those sales? Which ads did they see? Which keywords did they search? Which formats did they view? And work backwards from there.”

Done right, that modeling is used to assign credit (in dollar amounts) to the channels or marketing activities which actually influenced the customer. Done over a long enough period of time, it’ll allow marketers to sort baseline and seasonal sales patterns from the incremental ones driven by their discretional marketing activity.

That sounds like last touch attribution, no?

That’s not too far off. There’s a lot of overlap between attribution modeling and multi-touch attribution, in that they try to account for each moment a customer might have been influenced to make a sale.

Because they approach that problem by figuring out a customer’s path to purchase, there’s an inherent bias to the “last touch” — the last time a customer saw an ad from a brand, even if they’d been influenced to buy long before that stage.

And because that’s likely to be a search ad rather than a 30-second spot shown halfway through last night’s prime-time drama, attribution modeling can be biased in favor of performance marketing channels. “You’re allocating a lot of credit towards bottom-of-funnel marketing, and it makes it hard to justify your upper-funnel marketing,” said David Walters, director analyst at Gartner.

It’s not the only limitation of attribution modeling. The data used to track a customer’s buying journey would previously have been derived from third-party cookies, as well as an advertiser’s own (first-party) data. 

“The practicality of being able to get access to the required data has changed quite substantially,” since Apple triggered the yearslong deprecation of the cookie with its 2021 iOS 14.5 update, said PMG’s Jackson.

Some attribution tools (like one offered by Adobe) offer probabilistic modeling — using estimates based on historic data to fill in the gaps left by the absence of third-party cookies.

With those flaws, why not use MMM?

Because time is of the essence. While MMM might be able to provide a snapshot once a week, attribution models can track customers as they make a purchase much closer to real time. Marketers can use Google Analytics or tools from Adobe for attribution modeling, while MMM has until recently involved setting up a bespoke framework.

“If you’re launching a digital campaign and you want to understand how it’s performing in real time, touch based attribution can give you the answer, or at least an answer,” Jackson said.

In any case, plenty of marketers use both frameworks, per the needs of their business. “They should always be used in complement with each other,” he said.

At Vinted, a European e-commerce reselling platform for fashion and electronics, that’s very much the case. The brand, which formerly emphasized a performance marketing approach, recently launched its first major brand-building campaign. 

As such, Andrew Smith, senior director, brand, told Digiday: “Efficiency metrics and the performance metrics are always critical for us to follow so that we know we’re spending wisely. But we’re also really interested to see where some of the longer-term or more top-line metrics are pointing.”

But why is its use climbing?

Partially because it’s become the default way of reviewing digital marketing investment which, surprise surprise, continues to rise. GroupM recently projected that global digital advertising spend would account for 72.9% of all ad investment in 2025.

For marketers skipping on TV investment, impact modeling that takes into account brand awareness or other upper-funnel measures won’t be as much use. “In a fairly short consideration cycle for purchases, you know you don’t need as complicated of an attribution model,” said Walters.

Sure, the system’s open to misinterpretation. But every model has bias. “It may be underestimating the impact of some media,” said Jackson. At least we know in which direction attribution modeling can tilt the board.

In any case, despite the turn back towards brand among advertisers like Nike and Airbnb, marketers are still under enormous pressure to prove their marketing activity is worth the cost. The 2024 edition of Gartner’s Annual CMO spend survey suggested marketing budgets had fallen on average to 7.7% of a company’s overall revenue, down from 9.1% a year prior.

In short, marketers need to measure every penny of their investment lest it be reduced further. It might not provide the answer, but attribution modeling can provide some insight and justification.