WTF is attribution modeling?

In this edition of WTF, we demystify the entire thing.

I know I’m supposed to care about attribution, but I’m not sure why
Because, ultimately, it helps you figure out where to put future resources. Attribution essentially says a certain channel led to someone buying something. So if a certain channel is the one doing the job, it may need more of a budget next year.

OK, that’s great. I guess this solves all my problems.
Eh, sort of. Attribution is the answer, but nobody knows which model of attribution actually works. There are a few. Let’s break them down. Single-source is the most common model — it’s also the most deeply flawed. It’s got a few other names, like last-click or last-touch, or first-click. Essentially, it gives the channel that was the last one (or the first one) to present an advertisement to a converting customer the credit. 

I get why that won’t work.
Yep. But the other models also have issues. Multi-touch attribution recognizes that there are many touchpoints in the mix here. So a customer’s decision to buy something can be influenced by many things, including a search ad they clicked on last, or an ad they saw on Facebook but didn’t click on. Foursquare and Nielsen teamed up this week for a new attribution metric that uses phone location data to figure out how to track foot traffic in-store. Ultimately, multi-touchpoint collects all of the events related to a conversion, also known as the “buying journey.” That gives more importance to the sequence of the inputs. So a marketer may decide to target a customer with a video ad only after a that person has been primed by multiple search ads, for example. Sometimes this is known as a linear attribution model, which some people call a participation certificate, with every bit of the process getting equal credit. There are also time-decay models, which many people are fond of. These make a lot of sense: In time decay, the further back a channel is, the less credit it gets. 

You mentioned some other models.
Yes, the last bucket of attribution models is the most complicated: A fractional model will say that there are lots of touchpoints and it recognizes all of those factors. Ultimately, it eliminates “biases” that other methods have. It’s not perfect. Probabilistic models will, similar to fractional, weigh different touchpoints. But — and this is an IAB definition — the value of each event is determined in relation to all the other events. And the value calculation takes into account all the other events even if a conversion didn’t happen. (So even if a video ad didn’t prompt people to buy boots but got tons of Facebook comments, it’s still valuable.)

Why are these important, though?
Marketers are doing more and more across different channels, and they would like to know what’s working. For those working inside brands, it’s even more of an issue. As one digital executive told Digiday, “We have trouble convincing management that digital activities that drive average or below average ROI on our owned sites actually affect retail in a positive manner,” he said, explaining he has trouble attributing offline sales to digital marketing. “We are exploring match market studies for retail lift, as well as showing branded search uplift before and after digital buys to help show movement even if it’s not showing in our site analytics.” Everyone wants proof, which means that teams internally want credit for their own activities.

So it’s about transparency.
In a way. But it’s also about internal restructuring. People need to know what’s working. And those doing the right things want credit. As a report from Edelman puts it, brands are in a new position now to want both efficiency and accountability from digital but with transparency of traditional media.