Automating parcel volume predictions
Kickstart AI X Ampere - Distribution fitting
How do you improve sustainable package delivery? The logistics startup Ampère worked with us to automate their parcel volume prediction. We developed a data-driven solution called Quantile-Assured Volume Estimation (QAVE) to improve efficient route planning.
Optimizing sustainable delivery
Ampère, a delivery partner for Dutch consumer product giant bol., struggled to predict parcel volumes for pickup routes. Without knowing the package sizes from various partners in advance, it was difficult to plan van capacity efficiently. This means vans were often being underfilled or overloaded, leading to extra trips and higher costs.
Goals
Automate parcel volume prediction to eliminate manual inefficiency
Minimize unnecessary trips and associated costs.
Reduce fuel consumption and emissions by optimizing logistics.
Approach
We started by aligning on the use case. We then organised a structured 2 week challenge for our engineers and those of our partner companies to work together on a technical solution. After the challenge we collaborated to develop the infrastructure and implement change management within the company to ensure the successful integration of the solution. Following that, it was time to finalise and move into production.
We divided the challenge into three subtasks.
Evaluation procedure
Linking prediction models to route planning to measure real-world impact.
Volume prediction with dimension scanner data
Using historical parcel dimension data for prediction.
Volume prediction with bol. data
Exploring integrating bol's sales data for enhanced prediction.
We had to gain operational trust while improving efficiency. A phased approach, starting with conservative estimates and gradually optimizing, helped to build trust with the Ampere team. Close collaboration and on-site work made integration of QAVE smooth and seamless, with minimal disruption to their operations. The automated solution scales with Ampere's growing operations.
Key features
We found that as Ampères’ operations increased, the volume would restrict planning more and more. We estimated that with a better estimation, we’d land a 6-8% efficiency gain. With that in mind, a relatively simple data model would yield the most benefits. That is why we developed a method named Quantile-Assured Volume Estimation (QAVE). Here's how it works:
Historical data analysis
Analyzing the last 30 days of scanned parcel data for each partner.
Distribution modeling
Fitting a log-normal distribution to daily average volumes.
Predictive quantiles
Using the 95th percentile of the distribution for prediction.
Stacking inefficiency factor
Applying a factor of 1.5 to account for stacking inefficiencies.
Technology
The data pipeline runs entirely on Google Cloud Platform (GCP). SciPy fitsl distributions, storing parameters that a FastAPI endpoint uses to serve quantiles on request. This endpoint connects to the prediction service, an algorithm gateway in Ampères’ infrastructure, and integrates with their planning software. Dashboards in Grafana and Looker Studio track efficiency, business impact, latency, and coverage.
Smarter predictions and smooth integration
By automating parcel volume predictions, this project helps Ampère to boost efficiency and improve their sustainable logistics. Our solution, QAVE, proves the power of a simple, data-driven approach. With smart integration and automation, it delivers impressive results. We streamlined logistics, reduced waste, and landed tangible real-world benefits.
Building on this success, Ampere plans to integrate bol. product data and apply machine learning to refine packing strategies. Their ultimate goal is to scale the solution across all depots for even greater impact.
Results
Reduced planned volume
Achieved a 10% reduction in planned volume.
Improved operational efficiency
Minimized underutilization and overloading of vans.
Enhanced sustainability
Reduced unnecessary trips and optimized vehicle use.