Arvato Systems - BMS AdStore
Ratings forecasts are critical for broadcasters in defining their prices and sales models; models that are becoming increasingly complex as broadcasters look to offer cross platform campaigns to compete with newcomers disrupting the market. Advertisers want to know how and when they will hit their target GRPs or impressions. At the same time, the accurate prediction of ratings, by human beings, takes years of experience and practice and is very much increasingly a dying art.
This is a crucial business. In Germany, for instance, linear television still dominates advertising expenditure with a market share close to 50%.
There are a large number of factors that may affect ratings and the relative importance of these factors varies significantly between peak and off-peak viewing. Factors included in our historical analysis and the resulting prediction models we developed include related data such as genre, time information like weekday, week and month and also contextual data.
Many factors were considered with even just the top 15 varying significantly as models were developed. Four different machine learning methods were also evaluated with significantly different error — or accuracy — rates in the results.
It's vital for linear channel programmers to have data available to make informed scheduling decisions. Arvato is bringing new AI-technology to solve a real-world problem. An innovative product using Artificial Intelligece and Machine Learning, providing predictive information essential for broadcasters ratings.
From the start of the project, we very quickly reached a point where the AI was matching the accuracy of the human predictions. This included areas (particular in terms of weeks or time slots) where the error rate was higher both for machine and human. This became a focus point for the AI model to analyse what factors were driving the higher error rate which led to further models delivering even higher levels of accuracy.
As a result, we are not only now providing more accurate forecasts with BMS AdStore, we are able to do so faster and to iterate forecasts as critical external factors change.
There has been a lot of ‘buzz’ around AI in recent years, but the vast majority of demonstrations have involved augmenting or adding metadata – speech to text, object recognition etc. While this makes a good demonstration, our application of AI in ratings forecasts has a much more direct relationship to the commercial success of our customers. By providing more accurate forecasts more often, broadcasters are able to better optimize the placement of commercials and ensure they hit the KPIs of their advertisers.
What’s more, BMS AdStore can also be combined with other innovations in advertising sales and delivery models that are proved to maintain or even grow the value of TV advertising in an increasingly competitive space