United Arab EmiratesLogistics and transport2021 – ongoingAramex
Machine-learning prediction of shipment transit times
Aramex cut call-centre volume 40% after a machine-learning model improved its delivery time estimates
Problem
Aramex wanted customers to get more realistic delivery windows in the last mile. Poor estimates led customers to call to check on their parcels.
Approach
Aramex trained an XGBoost model on Amazon SageMaker using 6.3 million rows of shipment data and 12 features. AWS partner Inawisdom built the production solution in eight weeks, starting from models Aramex had developed in notebooks. The system served up to 4 million predictions a day at peaks such as Eid and Black Friday.
Results
Takeaway
Better delivery estimates reduced the calls that customers made to ask where a parcel was. The results come from an AWS blog and the article does not explain how they were measured.
Categories
- Size
- Large enterprise
- Industry
- Logistics and transport
- Business function
- Production and qualityCustomer service
- Innovation type
- Process innovation
- AI technology
- Predictive analytics and ML
- Adoption
- Partner co-development
- Evidence
- Vendor case study
Sources
Source summaries
An AWS blog post describes how Aramex built a transit-time prediction model on SageMaker with partner Inawisdom. It gives the training data size, the peak prediction volume and the build time of eight weeks. It reports a 74% rise in prediction accuracy and a 40% fall in call-centre volume.
Full details are in the original sources.
