Adding AI and ML to a Rich Data Pool
Optimizing product assortments for each location required us to aggregate multiple data sources. At one level, these could be broken down into transaction data, kiosk data, and title data. But a closer look revealed just how rich and varied this data pool was. In reality, we needed to process:
- Kiosk behavior for each location
- Average cost per title
- Average cost per DVD/Blu-Ray disc
- Average profit amounts per title and kiosk
- Historical rental patterns
- Time and volume of transaction
- Member demographics
- Web/app activity (i.e. renting movies via app or website)
To forecast how new movie titles would perform in a given location, we added third-party data (i.e. reviews, customer sentiment) to the mix. We also used historical data from similar movie titles to predict how each location would respond to a given new title.
Over a four-week period, advanced AI and ML techniques (specifically, Gradient Boosted Machines, Random Forests, and K Nearest Neighbors) were applied to the above data. The client was able to drill down to the title level and see the suggested assortment for each kiosk – not just which titles to stock, but how many to stock and how to split stock between DVD and Blu-Ray formats.
From Guesswork to Data-Driven Assortment
With ML-based predictive analytics, much of the guesswork has been removed from this rental company’s per-kiosk assortments. By comparing how similar titles have fared in the past, they can predict which new releases will do well in each location. And by employing AI-assisted number crunching, they can quickly see the most profitable choices for DVD and Blu-Ray purchase.