Datatonic designed and implemented a machine learning approach to improve TV4’s ad-targeting efficiency by enhanced prediction of viewers’ age and gender. For this purpose, a machine learning model was built on user features engineered from a comprehensive analysis of TV consumption patterns. This approach outperformed the accuracy of TV4’s existing age-gender predictions by over 2pp, with direct impact on reducing ad mistargeting - a key driver for advertisement revenue.
Improved ad-targeting efficiency (reduced advertising waste) through more precise matching to the industry “ground truth” of viewer demographics (i.e. MMS 3rd party measurement)
Enhanced performance of existing model by over 2pp in prediction accuracy of age and gender brackets while limiting the model to comparable data
Conducted in-depth analysis of user consumption patterns including a detailed set of features
Created a scalable and flexible cloud-based pipeline capable of generating updated age and gender brackets for millions of customers
TV4 targets its viewers for ads based on age and gender brackets. MMS is a 3rd party which measures consumption of a representative panel and uses these measurements to predict age and gender brackets for every viewer in Sweden. These predictions—in the form of reported actuals for a show or ad—serve as the media industry’s currency. Both broadcasters and advertisers use the reports generated by MMS to determine the extent to which advertising obligations have been fulfilled. MMS uses a sophisticated statistical modelling approach based on the most up-to-date viewership data across several major media outlets in Sweden. The challenge for TV4 is to match the age and gender brackets predicted by MMS as closely as possible, so that valuable inventory is not wasted on targeting the wrong viewers.
Objective: Our goal was to explore machine learning approaches to predict age and gender brackets that better match those output by the MMS model.
The solution was built with state-of-the-art tools such as GCP’s AI Platform, BigQuery and Tableau, with focus on performance while minimising the computational and maintenance costs. Datatonic developed a machine learning pipeline using BigQuery and XGBoost that allows TV4 to rapidly generate age and gender bracket predictions for all of their viewers. This pipeline first creates detailed user viewing logs. These logs are then aggregated to create a summary consumption profile for each user. Using parts of this profile as input features, a machine learning model is trained to predict age and gender brackets for each user. The final result is a table in BigQuery with updated segments for each user that can be directly inserted into TV4’s existing ad-serving processes. Our scalable solution was able to significantly reduce advertising wastage while providing TV4 with the flexibility to easily experiment with the pipeline to further optimize results.
“We highly appreciated Datatonic’s co-development approach and emphasis on knowledge sharing. We were also impressed by very professional presentations throughout the project that provided easy explanations of sometimes quite complex concepts that we were working on”
If there’s data, we can optimise, innovate with and automate it to drive business value. Find out how we’ve supported the biggest brands across industries.
Know exactly where and how to start your AI journey with Datatonic’s
three-week AI Innovation Jumpstart *.
* Duration dependent on data complexity and use case chosen for POC model
With your own data sets, convince your business of the value of migrating your data warehouse, data lake and/or streaming platform to the cloud in four weeks.
With your own data, see how Looker can modernise your BI needs
with Datatonic’s two-week Showcase.