Fast Data Science Ltd
Apr 06, 2023
No image
Predicting customer purchases

Predicting customer purchases

7-12 months
United Kingdom, London
view project
Service categories
Service Lines
Artificial Intelligence
Big Data
Domain focus
Retail and Restaurants
Programming language
Artificial Intelligence
Deep Learning
Machine Learning
Big Data
Predictive Analytics


Tesco needed to understand and predict when customers are likely to add large heavy items to their online shopping basket. Tesco’s delivery vans are typically small vans, which are able to turn in residential streets, and have a maximum load capacity of 800 kg. Customers’ orders are often in the range 20-50 kg, and one driver typically visits up to 30 customers’ homes in a trip. Customers are typically allowed to modify their baskets until midnight the day before their delivery. This can cause problems with logistics, since a large last-minute change to an order can push a driver’s load over 800 kg. For this reason, Tesco wanted to better understand and predict customers’ shopping preferences, in particular customers’ order weights.


Fast Data Science developed a predictive model using Gradient Boosted Trees in PySpark which predicts the likely shopping weight in kilograms of every customer with a Clubcard, varying by date. For example, Christmas is a peak time when customers order large quantities of alcohol, which is heavy and expensive to transport. The model takes account of the customer’s purchase history, location, demographic data, and the date of the delivery, to produce a prediction for all 15 million customer and address combinations in Tesco’s order database, daily.


This means that every time you place an order on, Tesco is now able to anticipate how much that order will weigh in kilos, and will have allocated you to a vehicle… before you have even decided what to buy! The model provided a 3% increase in accuracy of Tesco’s prediction of shopping basket weights, allowing Tesco to make better use of its van fleet, save on fuel, and pack its vehicles more efficiently.