Marketers, We Need More Accurate Attribution Modeling

Share this:

Imagine you’re at an amusement park. For the price of three hats in a normal store, you buy one silly hat that you probably won’t wear again. 

Why did you buy it? 

Maybe because the sales clerk brought it to your attention. Or perhaps it was because you saw people wearing them and taking selfies around the park. Or was it because the ad you saw weeks ago made the hat and the park experience inseparable in your mind? 

The answer could be all of these reasons and more. How can marketers track and understand all the steps over time, both online and offline, that led you to buy the hat? 

Attribution modeling is the science of “why did they buy?” And marketers, if we’re going to have any hope of influencing customer behavior, we have got to get better at it.

The Current State of Attribution Modeling

There are multiple types of attribution modeling with varying degrees of accuracy and nuance. The most straight-forward models are first-touch and last-touch attribution. However, with all of the ways a customer can shop and buy now, it is clear that customer journeys no longer follow such simple paths (if they ever did).

Current attribution models, such as multi-touch attribution, attempt to account for all of the steps along the way in a customer journey. The process of multi-touch attribution injects customer data, analyzes thousands of customer journeys, scores the effectiveness of each customer touchpoint, and then uses aggregate data to determine the most essential components of a sale.

 Multi-touch attribution models generally fall into four categories:

  1. Linear: All touchpoints are valued the same.
  2. Time-decay: Assigns the highest value to recent touchpoints.
  3. Full-path: Four key touchpoints are weighted at 22.5% each—first touch, lead creation, opportunity creation, and close. The rest is distributed among remaining touchpoints.
  4. Custom: Using historical data, every touchpoint has its own attribution percentage. A full-path model can provide the baseline to build a custom model.

A similar concept, omnichannel attribution, scores the effectiveness of each channel along a customer journey. Omnichannel attribution takes into account all touches on all marketing channels—including online, offline, paid, owned, and earned.

Attribution models that take into account as many touch points as possible, online and off, give far more accurate views of the customer journey. Multi-touch and omnichannel attribution allocate credit to the big draws — the multi-million dollar advertising campaign or Black Friday sale — and also account for less-visible factors such as social media engagement that grabbed consumer attention.

Let the Data Lead: Intelligent Attribution

Though manual multi-touch and omnichannel modeling provide more detail, they still can’t account for a dynamic, evolving marketplace. Conversions are happening everywhere all at once, and no predetermined model can keep up. Marketers need AI-driven modeling to truly chart customer journeys when and where they are happening. 

Traditional multi-touch models make upfront assumptions about value that may distort what is really driving a conversion. While they are more inclusive and flexible than a first- or last-touch model, they can’t update and evolve swiftly. They fit the data to the model.

AI-driven modeling uses all data sources available to analyze conversions and create an ever-evolving predictive algorithm. In short, intelligent attribution fits the model to the data. Because it’s constantly taking in new data, AI-driven modeling is scalable and continues to improve accuracy over time.

Elevate Your Customer Data

In order to produce accurate attribution models, data must be combined, centralized, clean, valid, and recent. Brands that compile customer data from all channels and assemble the tech that produces multi-faceted views of customer journeys will have a competitive advantage. AI-driven modeling is possible with the right data tools in place.

Tom Treanor is CMO at Treasure Data.