First generation attribution solutions refer to the following providers: Visual IQ, Adometry and Convertro among others. They are characterized by an attribution model that is built typically using statistical analysis techniques, such as data regression . This approach collects data for 3-6 months and then uses the collected data to compute the coefficients of a regression model that estimates the attributed value of an ad. While this approach can be better than an ad-hoc rules based approaches there are several drawbacks:
- The model is built on historical data and even if it is applied in "real-time" it provides the attributed value from data estimated in the past/historical setting. This is unlikely to capture at least two major kinds of events: (i) changes to the media mix performed by the advertiser after the data used in the model was collected. Clearly the model will not capture these changes (such as a new social media campaign) and will have to be recomputed (ii) external changes such as a competitor introducing a new product which changes the response to advertising.
- The model based approach generally has SLA's of up to 45 days for a model revision and this is a consulting service and is generally unscalable and expensive
- Regression models are hard to interpret and generally require an analyst or "expert" to interpret the results for a marketer.
In contrast Abakus provides:
- A game theoretic formulation that computes the attribution in real-time while accounting for on-going and continuous changes to internal and external factors of the advertising
- Results are available daily for a marketer to directly access and get prescriptive actions on how to change their campaigns to improve efficiency
- All results are provided with white-box transparency that explain to a marketer why something is working or not and what they should do about it to further improve efficiency.
Abakus is a software only solution versus first generation solutions that provide consulting services wrapped around a software system for regression modeling.
Comments
0 comments
Please sign in to leave a comment.