fbpx

The core part of Conversion Logic’s cross-channel attribution product is incorporating all user-level addressable channels – digital and direct mail – with TV and radio. This innovation is unique, and until recently, not offered by any marketing measurement solutions.


This industry has traditionally divorced the “top down” (marketing/media mix modeling), “bottom up” (user level attribution) and TV attribution approaches. We have built a streamlined advanced statistical approach based in time series to bring all sides together. TV attribution consists of 2 steps: estimation of TV lift using the cross-channel model, and spot-level lift allocation.


1. TV lift estimation At Conversion Logic, we build an ensemble of statistical cross-channel time series models to predict conversions over time. This is based on all digital and offline marketing channels, including TV airings at the spot level, TV ad-stocks, and seasonality-related variables (time of day, day of week, week of year, holiday dummies, etc.) Other variables such as promotions, macro economic variables, and competitive airings can also be included based on availability and model fit.


Once the ensemble conversion model is built, TV’s total and station-level lift over time is estimated by running multiple simulation scenarios with and without TV-related variables, and in different combinations of marketing stimulus. This tells us, in short, how much TV as an overall channel is contributing to each conversion.


2. Spot-level lift allocation The lift due to TV’s direct response is then distributed into individual spots using exponentially decaying lift function within the post window, impressions and station’s performance as estimated by the conversion model. This tells us how much individual spots are contributing to each conversion.


This two-step model is based on the Ensemble Method, mentioned above, which trains multiple state-of-the-art machine learning algorithms and combines their predictions together to deliver most accurate, unique and flexible model for each client. These are the most innovative, best-of-breed algorithms in academia, practice, or potentially even developed by client data science teams. The Ensemble framework allows us to measure TV attribution more accurately than ever before.

by Mert Bay, Principal Data Scientist @baymertbay