Algorithmic Attribution: Choosing the Attribution Model That’s Right for Your Company

Put simply, attribution is about identifying who receives the credit for conversions — a device, a person, an ad, a keyword, or a channel. And, it serves to reach several key goals within a company. First, there’s budgeting. Companies use attribution to help plan their budgets in a data-driven way, ensuring they invest money where the most profit is being generated. Second is bid optimization. Companies often use more granular data with attribution to determine how they want to emphasize their placements for certain search display ads or social apps. And, finally, companies will use attribution for general targeting refinement.

The Different Attribution Models
Companies have several attribution models to choose from. Let’s examine a few, including their strengths and weaknesses. Then, we’ll take a look at how to choose the best fit for your needs.

Last Click or Most Recent
The attribution model that first comes to mind for most marketers because of its ease of use is last click or most recent. We can liken this model to a relay race. The guy who runs the anchor leg and crosses the finish line is often the flashiest person and receives the most credit. Spectators often ignore the runners of the earlier legs and even the coach and sponsors — all of whom deserve credit for their contributions, too. Likewise, last click or most recent focuses solely on the last link in the customer journey toward conversion, without giving due credit for the contributions of other key touchpoints in reaching conversion.

First Click
First click is the other, less-popular extreme and is frequently checked as a backup to last click. Usually, this is where display and social channels really shine, as they do the heavy lifting to build initial awareness. Without a very complex option around attribution flexibility, many people often just choose last click as their primary attribution perspective and then double-check first click now and then when results seem too good to be true.

Linear Attribution and Participation
The linear attribution model attempts to be fair. We can liken this one to the political debate of equality versus equity. Many progressives believe in equality — everybody deserves the same opportunities — and that’s what linear is like. Linear starts with the final conversion total, divides it, and gives an average to each one of the touches. Linear’s strange cousin is called participation. Instead of dividing up the pie to give everybody a piece, participation tries to give 100 percent of the credit to everybody — even though this method tends to mess up the totals.

The custom-weighted perspective is an attribution model with many variations. For example, the popular U-shaped custom model heavily weighs the first and last clicks and gives less weight to the middle. Another popular custom perspective is decaying, in which much of the credit is given to the last click, only a little bit is given in some sort of predetermined recency curve, and the ones before that are “decayed out.”

The Inherent Problem With Attribution Models
The problem with these attribution models is that, no matter how smart you are — no matter how diligent you are in choosing last and double-checking first or trying to do some fancy data science to determine what the weight should be in decaying — you still must decide what you want the weights to be for each touch along the customer journey. No one wants to make that decision because it means a lot of heavy lifting. Users must review and revise every few minutes, hours, or days to keep it close to a version of the truth. This review is exhausting. There’s good news, though: there are now tools on the market that can help ease this heavy lifting.

Algorithmic Attribution as the Solution
Algorithmic attribution is about taking advantage of advanced statistics and machine learning to objectively determine the impact of marketing touches along a customer’s journey toward conversion, leading to a better understanding of marketing-campaign effectiveness.

Algorithmic attribution uses an econometric model of logistic regression, which estimates the true incremental number of purchases or conversions that can be attributed to or give credit to a given marketing channel. Going deeper, algorithmic attribution looks for differences in the ways customers who convert engage with marketing versus those who don’t convert to determine the credit that should be assigned. Basically, it looks for patterns between profitable audiences and unprofitable audiences. When it finds consistent patterns in the sequence and touches of people who do convert, it focuses more on those patterns than inefficient ones.

Benefits of an Algorithmic Attribution Model
Algorithmic attribution is objective instead of subjective. It’s automated instead of manual and is very actionable, which is important because the primary danger with attribution is becoming so engrossed in the search for perfection that you forget why you were searching to begin with. When you have algorithmic attribution, you can be confident that your budgeting, bidding, and targeting are going to result in more balanced and more sustainable profit for your company.

Computer Sciences Corporation (CSC) — a top IT service provider with a very long sales cycle — applied a combination of algorithmic and rules-based attribution. Notice that CSC’s solution wasn’t to just go full-blast algorithmic but to use both. In doing so, they were able to gain multiple perspectives and combine them to develop what they felt gave them a whole truth. As a result, they were able to budget better, driving a significant increase in leads year over year.

The Attribution Model That’s Right for You
Despite the power of algorithmic attribution, no single perspective on attribution will be 100 percent true for everybody and everything. It’s important not to obsess over finding the perfect one, but rather to choose the best one or ones that make the most sense to you, and then — before you make any rash decisions around staffing or advertising budgets or bidding or targeting or anything like that — validate against the farthest opposites.

Algorithmic attribution is a very powerful tool but only when used wisely. It is something that everyone should be asking about, but nobody should be blindly following. Adobe can help provide the tools for algorithmic attribution, but you still have to apply what some consider old-fashioned validation of the truth. Because, in attribution, one must always consider multiple perspectives to understand the holistic impact of multiple touches along the customer journey.