Attribution is not a new problem. The desire to quantify the influence of marketing in terms of sales can likely be traced to early market days of the Roman Empire.
Yet over the last several years, attribution providers have stoked the notion that more definitive credit might be allocated to marketing using “algorithms” and “machine learning” on user/touch-level data.
In practice, however, reality frequently falls short of aspiration. Many of the reasons attribution projects fail can be rolled up under five current categories of challenges:
The People Factor
Data unification, the process of aggregating data from across sources and mapping it to a cross-channel user ID, is typically where an attribution project starts, and for good reason—incomplete data creates a problem that attribution can’t solve. But this process isn’t trivial. Before committing to an investment in an attribution platform, assess your existing data and tracking infrastructure and consider whether you will be able to connect enough of your data sources for the platform to work as intended.
Inability to Effectively Incorporate Offline Marketing into Models
Unlike many online systems and direct mail, where it is possible to collect data at the user level, offline marketing analysis typically requires looking for relationships on more aggregate data—testing factors like spend, location, and timing.
There is no perfect solution for these kinds of problems, but better solutions involve more opt-in data and a combination of modelling approaches. Consider evaluating user-level and non-user-level factors and how the methodology has been validated.
Failure to Account for Seasonal/Environmental Factors
Sometimes rain sells umbrellas, not marketing, but marketing attribution platforms aren’t always built to account for the influence of weather or other (less literal) environmental factors. Well-known brands also face the challenge of modelling the influence of momentum built-up over decades and the probable slow decline if they go dark on their marketing. Generally some experimentation and creative thinking will be necessary to account for the effects of seasonal factors and ambient brand awareness.
Outputs Don’t Translate Easily into Optimization Levers
Ad impressions are multidimensional events and it is often difficult to know which attribute(s) of an impression is most responsible for an outcome. From an attribution modeling perspective, there is a forced trade-off between abstraction and granularity when answering the question, “to what do we attribute this conversion?” Higher levels of abstraction (at the marketing channel level, for instance) allow analysts a lot of flexibility with the choice of the analysis mechanism and result in interpretable models, but they provide less visibility into the relative influence of the different attributes of the advertising impressions. This forces an analyst to provide a measure of effectiveness for the entire channel, when different tactics within that channel may range in effectiveness from ineffective to extremely effective. The operative question when developing a modeling approach is what you feel you’re going to need to get out of it to know which levers to pull.
Inability to Validate the Model
Many attribution solutions come pre-fitted with a model or algorithm built to sort out fractional credit for sales across marketing tactics. From a utility perspective, this is conceptually attractive, but the idea that general-purpose math can work well “out-of-the-box” is debatable. The reality is that there are often significant differences in predictive performance when applying the same model to different data and questions. Basing big decisions on an invalid model is potentially worse than simply leveraging experience and intuition. It is worthwhile to spend some time testing and experimenting after the implementation is complete.
Even if we ask the right questions, we sometimes receive answers that downplay the potential complexity and gloss over the kinds of challenges marketers face.
Reasons Attribution Fails And How To Avoid The Trap
ð Don’t underestimate the demands on your in-house marketing teams and vendor partners
ð Drive to incorporate offline marketing and paid social into models
ð Account for seasonal and environmental factors
ð Think through the process of translating outputs to marketing optimizations
ð Plan to validate the model