Kaizen Gaming builds real-time betting personalization with Azure Databricks

Overview

Kaizen Gaming, a European game-tech company handling up to 4,000 transactions per second, used Azure Databricks with Delta Lake and MLflow to move ML model training off local PCs onto a scalable cloud platform, cutting data set creation time from 240 hours to 20, and building a sportsbook recommendation engine that achieves 75%-85% precision across its top 100 events.

The challenge

Kaizen's model training ran on local PCs that couldn't cope with data volumes, limiting personalization training to just six days of data, and the lack of collaborative functionality held back innovation; manual bonus and reward processing via CRM could take up to two weeks, by which time customers had disengaged.

The solution

Kaizen adopted Azure Databricks with Delta Lake to move machine learning off local PCs onto a scalable cloud platform with a common workspace for models, notebooks and data, and used MLflow for model training, experiment tracking and deployment to build a sportsbook recommendation engine that proposes a daily top-10 list of events.

Reported business value

Kaizen cut data set creation time from 240 hours to just 20 (and in some cases to a few hours), expanded training data from 6 days to 365 days, moved its sportsbook personalization use case to Databricks in just four days, and its recommendation engine hit 75%-85% precision across the top 100 events.