Get startedGet started for free

Attribution modeling

1. Attribution modeling

Another component to marketing impact is attribution modeling. Changes in attribution can impact current performance and historical trends!

2. What is attribution modeling?

Over time, multiple channels can help a customer move from awareness to consideration to decision. In the simplest scenario, a single channel leads to purchase, and that shows up in reporting. In reality, customers can be exposed to TV, display, and paid search before making a decision. Determining which channel gets credit in that customer journey is called "attribution modeling." Marketing Analysts typically make an attribution model recommendation to the business. Marketers care deeply about attribution; it impacts optimization decisions, revenue influence, and future budget decisions if a channel drives many (or few) purchases.

3. Last touch attribution

There are many different attribution methods, but the most commonly used is called "last touch attribution" or LTA. It works the way it sounds, where even if multiple channels play a part in purchasing, the last channel before a purchase gets 100% of credit. Last touch is popular because it is a method that is easy to explain and interpret trends. Channels that naturally occur later in the funnel, and have Direct impact, will get more credit with last touch. By contrast, Indirect impact channels receive less credit in this model because they are driving awareness and consideration goals. In some cases, Indirect channels never receive credit for last touch because someone cannot purchase through that channel.

4. Multi-touch attribution

As marketing programs become more sophisticated, analysts can explore options beyond last touch attribution (even if they decide to keep last touch). The other main option is called "multi-touch attribution" or MTA. This attribution method splits credit across multiple channels that contributed to a customer's purchase. There are a variety of MTA modeling options, but most of them attempt to account for impact throughout the entire marketing funnel. This means that both Direct and Indirect impact channels are treated the same in an MTA model. The trade-off with MTA is that dividing credit makes it hard to understand why overall marketing trends are changing and requires extensive deep-dive analysis.

5. Heuristic and algorithmic MTA

When we test different MTA models, they divide into two main categories: heuristic, a rules-based attribution, versus algorithmic, an attribution based on statistical models. Both types divide credit fractionally, meaning it all adds up to 100%. Heuristic MTA can be a good middle ground between LTA and algorithmic MTA since analysts control the rules and complexity. Linear (splitting credit evenly between all steps) is a good heuristic model to baseline channel volume. Algorithmic MTA uses various statistical models, but usually, we start with regression. In this approach, channels with a stronger relationship to purchase receive more credit.

6. Time decay attribution

One popular heuristic model is called "time decay". Time decay attribution assumes that the closer a channel is to purchase, the stronger influence it had on the purchase decision. Time decay assigns more credit based on recency to purchase (or other outcomes).

7. Choosing an attribution model

There is a saying that "all attribution models are wrong." It can feel daunting knowing which one to choose, but there are no perfect attribution models. We can weigh the trade-offs between different models and decide. Some things to consider are: how precise does the model need to be, how much do people need to understand the model logic, does customer behavior align better with certain channels, does our channel mix skew toward Direct impact channels, and what access to journey data will we have? A final area to consider is that with privacy laws reducing access to individual data, algorithmic solutions can mitigate data loss.

8. Let's practice!

Now let's practice identifying differences between LTA and MTA.