Comparing Incrementality and Last Touch
Comparing incrementality metrics to last touch metrics is an important type of analysis. In many cases, both strategic and operational decisions are still being made based on last touch attribution, so any discrepancy between last touch and incrementality represents a potential inefficiency, sometimes with drastic financial impacts. It is critical to identify these discrepancies and confirm those with the largest impacts through incrementality experiments.
To begin, there will very likely be discrepancies between the two and those discrepancies are unlikely to be uniform across channels and campaigns. Incrementality measurement and last touch attribution are designed to measure completely different realities. Incrementality measurement believes that interactions exist amongst media and between media and organic demand and therefore considers the entire media mix including interactions amongst media and between media and organic demand.
In contrast, last touch attribution believes in a simpler reality with no overlapping interactions at all and therefore only considers the ad that happened to have the last touch within an arbitrary attribution window. We have confirmed that the version of reality championed by Polaris is much closer to objective reality than last touch based on a significant number of ground truth incrementality experiments clients have conducted on our platform. While last touch attribution can be correct at times, it is mostly random and cannot learn or adapt based on ground truth experimentation. That said, we believe it still has a place as an alternative signal to incrementality measurement.
Below, you can see the Analytics screen which provides a pivot table view that allows you to drill down into multiple configurable dimensions like platform and channel. The metric columns are also configurable and include both incrementality and last touch metrics. In the above example, the user is interested in comparing day 3 and day 7 ROAS and Polaris is showing that, in general, the incremental returns are lower than last touch is reporting.
In our experience, there are a few common discrepancies between incrementality and last touch. First, self attributing networks like Facebook and Google no longer have the benefit of self-attribution on iOS due to the deprecation of IDFA and the emergence of SKAdNetwork. Often, you will see some decrease in incremental installs allocated to those networks when compared with last touch. Sometimes, downstream metrics like incremental revenue are similar or even greater than last touch due to strong targeting and optimization algorithms on those platforms.
Also, display and other low engagement non-addressable media like TV and influencer can be underappreciated in the last touch model. High engagement ad units like playable ads can capture a lot of demand at the bottom of the funnel that was actually mostly generated organically or by other media.
Finally, organic demand is often underestimated by last touch especially for popular apps or apps that license strong IP. Whether last touch under or overestimates organic demand is often based on the attribution windows used. Long windows can actually cause last touch to overestimate organic. Furthermore, the composition of the media mix is also a significant factor since mixes that greatly cannibalize organic demand can cause last touch to underestimate organic and vice versa.
Basics
To begin, there will very likely be discrepancies between the two and those discrepancies are unlikely to be uniform across channels and campaigns. Incrementality measurement and last touch attribution are designed to measure completely different realities. Incrementality measurement believes that interactions exist amongst media and between media and organic demand and therefore considers the entire media mix including interactions amongst media and between media and organic demand.
In contrast, last touch attribution believes in a simpler reality with no overlapping interactions at all and therefore only considers the ad that happened to have the last touch within an arbitrary attribution window. We have confirmed that the version of reality championed by Polaris is much closer to objective reality than last touch based on a significant number of ground truth incrementality experiments clients have conducted on our platform. While last touch attribution can be correct at times, it is mostly random and cannot learn or adapt based on ground truth experimentation. That said, we believe it still has a place as an alternative signal to incrementality measurement.
Polaris Analytics
Below, you can see the Analytics screen which provides a pivot table view that allows you to drill down into multiple configurable dimensions like platform and channel. The metric columns are also configurable and include both incrementality and last touch metrics. In the above example, the user is interested in comparing day 3 and day 7 ROAS and Polaris is showing that, in general, the incremental returns are lower than last touch is reporting.
Common Discrepancies
In our experience, there are a few common discrepancies between incrementality and last touch. First, self attributing networks like Facebook and Google no longer have the benefit of self-attribution on iOS due to the deprecation of IDFA and the emergence of SKAdNetwork. Often, you will see some decrease in incremental installs allocated to those networks when compared with last touch. Sometimes, downstream metrics like incremental revenue are similar or even greater than last touch due to strong targeting and optimization algorithms on those platforms.
Also, display and other low engagement non-addressable media like TV and influencer can be underappreciated in the last touch model. High engagement ad units like playable ads can capture a lot of demand at the bottom of the funnel that was actually mostly generated organically or by other media.
Finally, organic demand is often underestimated by last touch especially for popular apps or apps that license strong IP. Whether last touch under or overestimates organic demand is often based on the attribution windows used. Long windows can actually cause last touch to overestimate organic. Furthermore, the composition of the media mix is also a significant factor since mixes that greatly cannibalize organic demand can cause last touch to underestimate organic and vice versa.
Updated on: 28/07/2022
Thank you!