AI model helps Ampère predict parcel volume for millions of bol shipments

DELFT, 16 September 2026: Logistics provider Ampère has successfully moved an AI package-planning model from a pilot project into daily operations. Spun out from bol, the logistics company has used the model continuously since September 2024 to predict how much physical space packages will take up at pickup locations. Developed and deployed alongside KickstartAI, the model has already processed millions of packages, the two organizations announced today.

The model – Quantile-Assured Volume Estimation (QAVE) – forecasts expected pickup volumes for each merchant partner based on historical package measurements. In practice, the predictions have proven accurate, missing actual collected volumes by an average of just 13 liters. This precision helps Ampère make better use of truck capacity and further automate its logistics planning.

Daily Logistics Puzzle

Ampère handles package collection and delivery for bol and its partner sellers. In daily operations, total physical volume matters just as much as item count. By accurately forecasting package volume ahead of time, Ampère can determine its daily capacity needs with far greater precision.

"AI is only valuable to us if it makes daily work easier and better," says [xxxx] at Ampère. "This model helps us plan faster and more accurately. We can use capacity smarter, avoiding tight scheduling bottlenecks without hauling around unnecessary empty space."

The partnership with KickstartAI began by building and testing data science models for a specific logistical problem. Today, QAVE runs live in daily operations and serves as the foundation for broader AI initiatives at Ampère. It also served as the catalyst for establishing Ampère’s own internal data science team. That team is now developing new predictive models for package weight and last-mile stop times, while also researching ways to better forecast late-day package entries into the network.

"Too many AI projects get stuck in pilots. This project shows what happens when AI targets a real operational challenge," says Erick Webbe of KickstartAI. "Together with Ampère, we built a model that didn't just end up sitting on a shelf. It runs daily and delivers measurable value."

More information about the project can be found at: [https://kickstart.ai/projects/automating-parcel-volume-predictions]