Guided by the team’s industry insights and expertise, T-DAB devised a machine learning strategy and solution that would meet their aims.

The answer was an API-based tool to analyse and predict growth and performance of “explainer videos” produced by digital marketing creatives and vloggers.

Using a range of attributes to quantify video characteristics, such as edginess and uniqueness, unsupervised machine learning models clustered videos to create an understanding of their performance in relationship to others

The production architecture equips YouTube content creators and brand content managers with predictive insights on performance alongside clear guidance on how to drive engagement.

This guidance was further strengthened through trend analysis for the extracted topics. This offers valuable insights into relevancy, trends identified and, most importantly, the remaining growth.

Using these details, together with video meta data, a supervised ensemble model was trained to predict audience engagement. The model features a heterogeneous mix of weak learners, each trained on a unique part of the overall input data. The model type for each weak learner was chosen automatically from several options, dependent on the test accuracy.

The user-friendly solution features an intuitive interactive dashboard offering video-specific insights. Integrated to the database, it provides data relating to a video’s predicted performance on YouTube, performance in relation to other creators, predictions of topic trendiness and much more.

Because content is always being updated, special care was given to its management and storage. The overall data architecture ensures collection and loading processes are automated, while a CI/CD pipeline allows the database to be quickly changed and updated. Video meta data and extracted features, for example, are securely stored on the Microsoft Azure SQL server and readily available for Cognitive Services tools.