Marketing Analytics Confidential: The Truth about MTA

For years, MTA (multi-touch attribution) was held up by many as THE gold standard for marketing analytics. The death of cookies is changing that, but even if cookies weren’t going away, MTA was never perfect.

It might look cooperative, but multi-touch attribution has some serious flaws.

In fact, in the wrong hands, MTA could be used for some pretty shady purposes if brand-side marketers weren’t being vigilant.

How? Glad you asked.

First, a primer on how MTA works: essentially, it takes a conversion and objectively assigns credit to a series of marketing touchpoints on the path to conversion. It does do a good job of giving a sense of a customer journey, but it is more directional info than hard truth.

But there are two big problems with MTA: 1. It does not account for incrementality. 2. It is easy for parties to cycle through models until they find numbers that tell the story they want.

MTA assumes that every conversion happened because of advertising, which is impossible - no campaign is 100% incremental. That means every MTA model is inherently flawed. Any party representing MTA results as 100% accurate is trying to take more credit than they deserve.

MTA will divide attribution among all of the touchpoints but cannot differentiate if those conversions would have happened without the touch. This is particularly common with retargeting and Meta campaigns. If you are not careful, you can see many touches from a channel set that didn't actually contribute to a conversion. (Check our blog on incrementality and halo effect to learn more.)

As far as vulnerability to manipulation, a shady actor can cycle through MTA models until a narrative that supports their goals emerges. For instance, retargeting or brand search campaigns would look great in a last-click model (the most flawed model of all), whereas upper-funnel MTA models would favor display, social, or even programmatic. All the "data-driven" models are a black box so they can't be scrutinized, and they’re owned by the same channels who sell the media.

So what should you use as an analytics model instead? Stay tuned to our Analytics Confidential series to learn about our recommendations.