One of UK’s largest retailers wanted to predict which of their thousands of brands customers are most likely to buy next. Following a Proof of Concept with Datatonic, they are able to use machine learning and advanced data engineering to achieve this, boosting operational efficiency and giving their Customer Insights Team on-demand intel.
Before working with Datatonic, the retailer’s Customer Insights team used dozens of manually crafted brand propensity models to measure affinity and anticipate which of their brands a customer was most likely to shop next. These models were slow to process and required many manual updates, making it difficult to scale beyond a handful of brands.
Each propensity model was updated monthly and took a few days to retrain, causing their Modelling Team to spend valuable time making manual updates, and less time making improvements.
Their Customer Insights Team’s top priorities were improving the company’s marketing activation programs, specifically in personalising product recommendations, and sharing insights with its merchandising, promotions, customer support, and finance departments.
Additionally, the Modelling Team needed these models to be scalable, automated, and able to be retrained quickly. They also needed to know the solution inside-out and be equipped to improve it over time.
After assessing several partners, Datatonic stood-out because of their familiarity with Google Cloud Platform (GCP), their expertise in machine learning, and the speed of their Technical Proof of Concept (PoC) service.
Several members of the Datatonic team had industry experience working for large retail and FMCG organisations, enabling them to not only prepare the data, design the scoring method, and engineer the model, but also work with the retailer to activate their scores in marketing and merchandising.
Throughout the process, a fully-documented knowledge transfer was carried-out, empowering the Modelling Team to make improvements once the Technical PoC was complete.
Datatonic rebuilt the retailer’s customer propensity models into a set of neural networks and tooling that prepares and trains over thousands of brands in just 4 hours using TensorFlow — Google’s open source machine learning framework.
The new solution enables the retailer’s Customer Insights team to leverage data from across thousands of online and offline touchpoints, and predict which brands customers will shop in the next 30 days. Their Modelling Team now have a single, highly-efficient propensity model for their entire brand portfolio. It can be retrained quickly when needed and allows for additional brands to be added without compromising performance.
To test its effectiveness, their Customer Insights team worked alongside their marketing department to deliver an email-based activation campaign promoting new arrivals, which A/B tested the existing and new frameworks with propensity scores simultaneously.