Scheduling content is in itself a complex and time-consuming task. It takes long experience and deep knowledge of how to attract and retain viewers by optimally scheduling the available content in line with the company mission.

Recent years have seen this complexity increase considerably as media companies try to reach and engage viewers with ever more content on an ever widening array of channels and platforms. At the same time, both viewers and advertisers have come to expect ever higher levels of, respectively, addressability and personalization. As a result, the list of dimensions that make or break a schedule has never been so wide and varied.

Combatting augmented complexity with augmented intelligence

Amid this growing complexity, we aim to help scheduling experts make those key decisions with ease, accuracy and speed. We have chosen to do so by automating time-consuming manual processes so that the experts can concentrate on what they do best, and by leveraging augmented intelligence on the experts’ input to make optimum use of the available content.

That is why we have developed and will continue to develop Scheduling ArtIst, our suite of AI tools that seamlessly integrate into the scheduling workflows. These AI products can either help with small individual tasks, or be combined into essential tools that help lift the entire content supply chain to new levels of operational and strategic excellence.

The purpose of these tools is not to replace cognitive judgement. The programme planners stay in control, and determine the level of abstraction that the algorithm is allowed to handle. This is demonstrated by the following three Scheduling ArtIst tools: Ratings Prediction, Scheduling Recommendation, and Automated Off-peak Scheduling.

Predicting ratings on linear channels with greatly increased accuracy

In order to help planners capture the target audiences and meet the expectations of the advertisers, we have developed a machine learning tool that accurately predicts the TV ratings of scheduled content for every relevant target segment. The predicted ratings are updated whenever new information (e.g. the schedule of competitors) becomes available. This information can be used to make changes to the schedule where required, and dramatically reduces the costs incurred due to mismatches between expected and actual ratings.

We have developed this tool for MTV Oy, the largest Finnish commercial broadcaster. MTV Oy has three channels, MTV3, Sub and AVA. The programmes on the MTV channels can be viewed online via mtv.fi streaming service. Paid content (films and series) can be watched through the C More service. In the competitive environment they are operating in, market share is very volatile and choosing the right content can make a huge difference.

While the movies they broadcast during primetime are crucial for their commercial success, the ratings those prime-time movies receive depend on a vast amount of variables. First of all there is the movie itself: what genre is it, who are the leading actors, what is the storyline? And also: how long is it, how recent, what kind of budget was it made on, … ? There is the effect of the surrounding schedule: what was the previous transmission, is it the first time this film is scheduled? Then there is the competition. Is Yle targeting a similar or complementary demographic, and if so, is their content more appealing? A lot more variables come into play. How many hours of daylight do we have this time of year? What time is it? What kind of weather can we expect, …? Is it a holiday?

By training a machine learning model to take all of these variables and more into account, we are able to accurately predict the ratings for these movies. Better still: our predictions are 25% more accurate than the current method where human experts enter the predictions.

Getting smart content recommendations when scheduling

The next step at MTV3 to support planners is to combine these ratings prediction data with other value criteria and filters in order to automatically provide the planner with a list of recommended content.

Filters can be query based (genre, parental guidance, duration, cost, …) or ML based (similarity, ratings, …), and so can the value criteria. Query based value criteria include cost, stock optimization, and frequency of scheduling. Examples of ML based value criteria are ratings, competition intelligence and similarity.

Generating off-peak schedules in a matter of seconds

Recommending content for scheduling seems one step away from automated scheduling. Primetime scheduling will basically remain a human craft, whether or not assisted by augmented intelligence. The context is different for off-peak scheduling, however. In WHATS’ON, programme planners can have entire off-peak schedules automatically generated.

They just need to apply filters and specify which value criteria are important. Based on these criteria, the algorithm calculates a weighted value score for all of the available content. Taking into account the structural requirements (such as already planned content, fixed starting time for episodes of series, no gaps or overlaps, …) it creates the schedule in a matter of seconds. While the planners can always finetune it, the algorithm will keep learning which choices result in the best schedule.

 

Media companies are collecting an ever increasing amount of data about viewer behaviour and content. Understanding this data, generating actionable insights from this data and responding to these insights to improve impact and monetization, requires the kind of combination of art and science MEDIAGENIX is advancing through augmented intelligence solutions that seamlessly integrate into the workflows.

 

Watch how Lucas Denys, Product Manager Data, demoed Automated Off-peak Scheduling and Ratings Prediction at our customer event (UAB) in March this year.  In this video he starts by explaining the MEDIAGENIX approach for building valuable products guided by the ‘self-reinforcing circle of data’. The actual Scheduling Artist demo starts at 05:22. 



Agree