Contact

Why counterfactuals matter to DDA

According to "Gartner", attribution is getting closer to the phase of real productivity. What does that mean for advertisers?

The "Gartner Hype Cycle for Digital Marketing"

According to the “Gartner Hype Cycle for Digital Marketing,” attribution has moved from the consolidation phase and is entering the phase of enlightenment and productivity. Based on recent developments, this attribution enlightenment must reflect the flexible data collection now available and the combined application of appropriate algorithms. This best-of-breed approach makes attribution an essential part of any marketer’s toolbox for performance optimization.

Attribution is dead. Long live attribution.

Advertisers have long known that half of their money is wasted. In recognizing this, they have sought ways to identify their most valuable activities and divert more budget into them from less efficient activities.

For performance marketers, a valuable activity creates incremental conversions (engagement, purchases, etc.) that wouldn’t have happened without marketing activity. In today’s marketing landscape, we see many analytical approaches that promise advertisers these insights, the most common being:

  • data-driven attribution (DDA)
  • controlled experimentation
  • marketing mix modeling

Data availability has changed hugely over the last few years, leading to a high degree of uncertainty among advertisers regarding the best methods to support their digital marketing.

This has resulted in some analytical influencers and commentators openly criticizing attribution in the marketing press and on social media. This mainly focuses on the shrinking data volumes and lack of counterfactual thinking. Although these points are valid for old rule-based attribution approaches, they are irrelevant to modern methods of data-driven attribution.

How incrementality attribution works

It isn’t easy to evaluate data-driven attribution and compare it to competing methods like controlled experiments. The variety of algorithms in DDA is more comprehensive than in the field of controlled experiments or media mix models. Years ago, the market was dominated by Shapley or Markow approaches. But today, market leaders like Exactag have developed more sophisticated approaches to overcome the drawbacks of earlier approaches.

Incrementality is at the core of the attribution as it conducts countless comparisons. The process includes:

  1. Tracking every touchpoint from successful and unsuccessful journeys.
  2. Describing every touchpoint through a series of custom characteristics (e.g., tactic, channel, etc.)
  3. Breaking down journeys into various components (i.e., a set of touchpoints).
  4. Comparing relevant situations against their conversion probability, all things being equal.

This approach allows us to identify those touchpoint patterns that truly drive a conversion probability. Compared to controlled experimentation and MMM, this approach benefits from high scalability and up-to-dateness as the underlying data set provides all relevant marketing dimensions (channel, campaign, tactic, publisher, placement, creative and custom parameters) and allows daily modeling.

How to measure incrementality despite consent and browser restrictions

Today’s DDA can integrate new data types, e.g. non-consented from anonymous tracking or view data from a data clean room. This new and granular data empowers advertisers to continue with data-driven attribution without the fear of biased results.

Obviously, the underlying data for attribution is more diverse than before, and suitable approaches to quantify incrementality must be defined by data type. As a result, there’s no good “one size fits all” model.

How to extend incrementality measurement to offline

Following the focus on incrementality, it’s also possible to extend attribution to other channels like direct marketing or TV (addressable/connected/classical) by using modules. Even brand impacts and exogenous factors can be integrated into the attribution, but again, selecting a method based on the given data structure is essential.