Algorithmic Multi-Touch Attribution

The LeadsRx algorithmic attribution model uses customized “weights” for each touchpoint. Weights are determined each day as the system “learns” about the performance of each touchpoint, and the values are unique to your specific account.

 

Weights are based on conversion rates. For example, if the click-to-conversion rate of touchpoint “A" is 25% (for example, 25% of people who click on an ad end up converting at some point in the future), then the weight for the algorithmic model is 0.25. In another example, let’s say touchpoint “B” has a click-to-conversion rate of 50%, or an algorithmic weight of 0.50.  In this case, if both A and B are involved in the same conversion, then B will get twice as much attribution credit as A.

 

 

Notice how touchpoint B receives 2x the credit of touchpoint A because it has a conversion rate twice as good. In this way, the algorithmic model favors better-performing advertising programs and adjusts over time as performance changes.

 

 

For #2, the calculation of weights explained above is what we do.  For #1, we use the impression-to-conversion rate of the ad as a basis, and then give this a “bonus” value because it’s an impression and not a click.  For example, let’s say the touchpoint “A" above is a display ad with 100,000 impressions and 125 view-through conversions. In this case, our basis weight is 125/100000 = 0.00125.  Next, we make an adjustment to this weight based on an industry-average click-through-rate, which is the ratio of ad clicks to impressions.  Currently, we are using 2% for this value (see this article: https://www.wordstream.com/average-ctr).  What this means is, we give a 50x bonus to the basis weight computed above (dividing something by 0.02 is the same as multiplying by 50).  In this example, the resulting weight for our Touchpoint “A” impressions would be 0.00125 * 50 = 0.0625.  

 

When using this in attribution models, we combine the impression weight with the ad click weight. In the examples above, our final weight would be 0.25 + 0.0625 = 0.3125.

 

Lastly, the weights are applied to ALL touchpoints in a given attribution path. This tends to favor display ads with impressions since often a high number of impressions are served to the user.  Consider this example (below, “impression” refers to impressions of our touchpoint “A”):

 

 

Here you can see there are 4 occurrences of “Impression” and 1 occurrence of touchpoint A (the ad click on that same Impression touchpoint).  As a result, Touchpoint A will get 5x its weight value when using our algorithmic model.