Attribution Modelling

Attribution Modelling is a critical component of sophisticated and at-scale digital media plan that uses a large number of publishers. A well tuned attribution model reduces wastage in an advertiser’s digital media spends and closely aligns them with the desired end digital marketing objectives. Another big benefit of a good attribution model is that it allows the media plan to experiment with new publishers.

Here is a quick list of a few simple attribution models with their pros & cons –

1. Last Touch for Conversion: Attributes 100% conversion to a publisher based on who touched that user last either on a view through or a click through. The advantage of this is that it identifies the publisher that drives the last mile conversion and the disadvantage is that it creates incentive for some publishers to game the system and win a last touch by underhand means.
2. First Touch for Conversion: Attributes 100% conversion to a publisher based on who touched that user first either on a view through or a click through. The advantage of this is that it identifies the publisher that actually pushed the user into the advertiser’s marketing funnel first. However, the disadvantage is that it creates incentive for bad actors (publishers) to game the system and win a first touch by underhand means such as buying cheap first touches at scale rather than actually driving conversions. This also significantly disincentives good publishers that come into the marketing plan later.
3. Last Three Touches for Conversion: This is a variation of Last Touch Attribution and attributes 33.33% of a conversion to each of the last 3 publishers that touched the user. The advantage of this is that it attempts to use more data to give a more equitable attribution. The very obvious disadvantage of this method is that it doesn’t allow disproportionate influence by one of the 3 last touches.

Our advice to the digital marketer is to not use just one of the above (or other variations of these). Instead, all of the above should be used. If a publisher doesn’t add value against any of the models, then they’re an obvious candidate for replacement.

Case Studies & Conclusions

The importance of using a sophisticated model versus one of the simpler ones highlighted in the previous section is best illustrated through real life situations. This section of this article is a running repository of such examples –

Consumer Non-Durables

This Advertiser had Google Search, GDN, 2 DSPs (including Digital Infusion’s) and a few other publishers in the media plan. Select data from the ad server and publisher performance data –

• DSP #1’s Conversions: 500 (50 – Attributed Conversions on last touch)
• DSP #1’s Clicks: 6,000
• Google Search Conversions: 15,000

Upon digging some use cases, we found that after deciding to purchase, a significant number of consumers used Google Search to access the advertiser’s website and clicked the search ad instead of the first organic result. Our conclusion was that in case Google Search is used simply to access the advertiser’s website i.e. by searching for the brand keyword, attribution is best given to the previous publisher. In this situation, clearly rule based attribution would be far superior than a fixed model such as last touch.

Also, we’re not the only one having run into this situation. There are others.