Marketing measurement is experiencing a renaissance - new innovative approaches emerge at an accelerating pace. One of these novel approaches is Causal Attribution, which attempts to solve the fundamental challenge in click-based attribution: the inability to measure the true incremental sales impact of each channel and campaign. As an example, large majority of conversions attributed to retargeting campaigns by Ad Platform attribution are current customers who would have purchased even without seeing advertising.
But what is Causal Attribution and how does it work? In this article, we’ll take an in-depth look at Causal Attribution and how it can help marketers improve their measurement.
Causal Attribution is a method for estimating the Incremental ROI of each channel by adjusting attribution-based ROAS with Incrementality Factors. This means that Causal Attribution is not really a marketing measurement method itself – it’s an approach that combines existing approaches to make attribution measurement more accurate.
The formula for estimating Incremental ROI with Causal Attribution is simple:
Incremental ROI = ROAS x Incrementality Factor
This formula captures the essence of Causal Attribution: you start with a familiar metric, attributed ROAS, and adjust it with Incrementality Factor to arrive at Incremental ROI, which is an estimate for media effectiveness that is far superior compared to the raw ROAS. The approach is simple and intuitive, making it attractive to many marketers. Incremental ROI is also sometimes referred to iROAS (Incremental Return On Ad Spend).
Let’s next look at the different elements in the formula one-by-one.
ROAS (Return On Ad Spend) in Causal Attribution is provided by an attribution tool. Attribution tools measure ROAS by dividing the value of attributed conversions with the amount of spend on the channel.
The most common attribution method used in Causal Attribution is Ad Platform attribution. However, other attribution methods can also be used, such as Google Analytics 4 Last-Click Attribution (or Data-driven attribution), or a Multi-Touch Attribution (MTA), as long as the Incrementality Factor is calculated specifically for the used attribution method. While the attribution logic is slightly different in each tool, they are all currently based on how purchase events can be tracked to ads that the buyers clicked during the purchase journey.
Incrementality Factor is a multiplier that is applied to the attributed ROAS, to arrive at Incremental ROI. Incrementality Factor, also sometimes referred to as Calibration Multiplier, is specific to each attribution method. The intuitive explanation for Ad Platform attribution is that Incrementality Factor represents the share of attributed conversion value that was truly incremental due to marketing.
Share of ad platform attributed conversions that are incremental can vary significantly depending on the type of marketing, but also by the type of company. As an example, prospecting campaigns to new customers tend to have high share incremental conversions. In contrast, attributed conversions for retargeting campaigns to existing customers tend to include lots of conversions that would have occurred even without advertising. As another example, mature brands with high brand awareness tend to see a higher share of non-incremental conversions in Ad Platform attribution compared to young high-growth brands.
To illustrate the math in Causal Attribution let’s take two examples, one based on Ad Platform attribution, and one based on Google Analytics 4 (GA4) last-click.
Let’s first look at an example, based on GA4 Last-click attribution. The calculations are summarized in the picture below.
Causal Attribution example using Google Analytics 4 Last-click attribution
In this example, GA4 last-click is reporting that prospecting campaigns have a ROAS of 1.50. Based on a Marketing Mix Model, you have estimated that Facebook Prospecting’s Incrementality Factor for GA4 last-click attribution is 2.5, meaning that the Incremental ROI is 3.74.
Similarly, GA4 Last-click in this example reports that Branded Search has a ROAS of 10.53. Based on a Marketing Mix Model, you have estimated that Branded Search’s Incrementality Factor for GA4 last-click attribution is 0.29, meaning that Incremental ROI is 3.16.
In this example, Causal Attribution was able to show that while Google Analytics 4 was showing a very low ROAS for Facebook’s Prospecting campaigns, it was actually more effective in driving incremental revenue than Branded Search.
Let’s now take the same example as before, but use Ad Platform attribution instead.
Causal Attribution example using Ad Platform attribution
In this illustrative example, Facebook Ads is reporting that prospecting campaigns have a ROAS of 9.84. Based on a Marketing Mix Model, you have estimated that the Facebook Prospecting’s Incrementality Factor for Ad Platform-reported ROAS is 0.38, meaning that the Incremental ROI is 3.74.
Similarly, Google Ads is reporting that Branded Search campaigns have a ROAS of 15.09. Based on a Marketing Mix Model, you have estimated that Branded Search’s Incrementality Factor for Ad Platform attribution is 0.21, meaning that Incremental ROI is 3.16.
The obvious question has probably already lurked into your mind: Where does the Incrementality Factor come from? There’s three main methods for estimating
Marketing Mix Modeling (MMM) is a time-series method for estimating the ROI (Return on Investment) of different channels and campaigns. In Marketing Mix Modeling, the model is given time-series data on sales, media activities, promotions, and other factors influencing sales. MMM model then learns each channel's sales impact by analysing whether increases in sales coincide with media activities, as illustrated below.
Illustration on the basic concept behind Marketing Mix Modeling
Next Gen Marketing Mix Models, such as Sellforte's MMM, are calibrated with experiments and attribution data, which makes them radically more robust and granular compared to traditional MMMs. As a standard output, Next Gen MMMs provide Ad Platform ROAS, GA4 last-click ROAS, and MMM-estimated ROI next to each other, making Incrementality Factor analysis easy. Check Sellforte’s Marketing Mix Modeling Guide for a thorough overview of MMM.
Marketing Mix Modeling is the recommended approach for Incrementality Factor estimation because of following benefits:
Experiments, sometimes also referred to as Incrementality Tests, are one-off efforts that provide an estimate for a channel’s incrementality. There’s two primary types of tests that marketers use.
Conversion Lift Tests are provided by Ad platforms, and include for example Meta’s Conversion Lift test, Google’s Conversion Lift test, and TikTok’s Split test. The main idea in conversion lift tests is to divide the target audience into a test group, which can see the ads, and a control group, which sees a blank space where the ad should be. Based on comparing these groups, one can derive insights about marketing’s incrementality.
Geography-based Lift experiment (or Geo tests) use a similar concept of comparing control and test groups, but in this case we are talking about geographies, such as U.S. states or U.S. Designated Market Areas (DMAs).
Best-practice in modern measurement is to use the information generated by Experiments as an input to Marketing Mix Models, where its used as one data point among many to calibrate the MMM model. However, using Experiments as a stand-alone method for Incrementality Factors is not optimal, because of three disadvantages:
Since Google popularized the idea of Incrementality Factors (or Calibration Multipliers in their terminology) in Google’s Modern Measurement Playbook, some public benchmark data has emerged on Incrementality Factors. As an example, below is a study done by Sellforte on typical multipliers for Meta channels that need to be applied Google Analytics 4 Last-click ROAS to measure Incremental ROI. Read the full study here.
Typical Incrementality Factors for Google Analytics 4 Last-click
While using benchmark data is useful for research purposes, it is no recommended for measurement use as a stand-alone input. It has one main disadvantage, which can be seen in the chart above. Even though all companies in the sample were from eCommerce, the range for the benchmark numbers can be quite significant. This is because Incrementality Factors are company-specific. If the sample had included companies from completely different industries, the min-max range would have been even wider.
Causal attribution has two main limitations:
Limited support for optimization due to lack of response curves: Causal Attribution is not able to provide response curves that would enable optimization of media budgets. For example, Causal Attribution might estimate high Incremental ROI for a channel, but additionally investments might not make sense because the channel is saturated.
Does not work if attribution data is sparse or non-existent: Causal attribution works best in channels with high volume of clicks and conversions that the ad platform can track. Causal Attribution breaks down for channel where there's a low amount (or even zero) clicks and conversions. These channels include offline channels, and several types of digital awareness channels.
Both of these limitations can be overcome with Marketing Mix Modeling -based measurement.
To summarize main points of this article:
Are you read to see Sellforte in action? Start your free trial.