Ampère has implemented an AI model for parcel planning. The logistics organisation, which originated from Bol, has been using the model systematically since September 2024 to better predict in advance how much space parcels will take up at collection partners. The model, which was developed and implemented in collaboration with KickstartAI, has now been applied to several million parcels.
The model, Quantile-Assured Volume Estimation (QAVE), predicts the expected collection volume per sales partner based on historical parcel measurements. In practice, the predictions have proven to be accurate: on average, they deviate by just 0.013 cubic metres from the actual volume collected. The model thus helps Ampère to make better use of capacity and to further automate its logistics planning.
Daily puzzle
Ampère handles the collection and delivery of parcels for Bol and its retail partners. In this operation, it is not only the number of parcels that is decisive, but above all the space these parcels take up. By better predicting parcel volume in advance, Ampère can determine more accurately how much capacity is required for its day-to-day operations.
“AI is only valuable to us if it makes our day-to-day work easier and better,” says Cas Beems, product owner and business analyst at Ampère. “This model helps us to plan more quickly and accurately. We can utilise our capacity more intelligently and avoid either planning too tightly or, conversely, transporting unnecessarily empty space.”
In-house data science team
The collaboration with KickstartAI began with the development and testing of data science models to address a specific logistics challenge. QAVE is now an integral part of day-to-day operations, and the model forms the basis for further AI applications within Ampère.
This project also served as the kick-start for setting up Ampère’s own data science team. That team is working on new predictive models, for example for parcel weight and stop times in last-mile delivery. In addition, it is investigating how to better predict how many extra parcels will enter the network later in the day.
“Many AI projects get stuck at the pilot stage. This project demonstrates what is possible when AI is directly linked to a specific operational problem,” says Erick Webbe of KickstartAI. “Together with Ampère, we have developed a model that isn’t just gathering dust on a shelf, but is used on a daily basis and demonstrably adds value.”
