Using optimisation to drive efficiency and improve the customer experience

The problem

As the demand for home delivery continues to increase, so too does the need for retailers to implement delivery operations that maximise efficiency and meet customer expectations. DFS’s existing process of scheduling orders was a complicated system of non-automated business functions. The schedules were far from optimal, manually intensive to produce and under utilised expensive resources.

We developed a last-mile delivery solution which combines cutting-edge algorithms and machine learning to optimise the routes and schedules of thousands of orders every day.

Dynamic optimisation and slots

In the previous method, operators would call customers to arrange deliveries, and only after all orders were arranged would a schedule be created. Operators could not offer customers’ specific time slots, as they did not know when during the day the delivery could be made.

In our solution, the schedule is live, and is re-optimised continuously. The newly generated schedule from one order informs the system of which slots the operator can offer the next customer. This helps operators to arrange only the orders they have the capacity to deliver, and maximises the number of slots they can offer their customers.

Better models, better algorithms

We built a custom algorithm that accounted for the various business constraints of DFS, including vehicle restrictions, loading times and driver shifts. We fine-tuned it to run a full optimisation in under 500 milliseconds, which was necessary for it to continuously offer the most optimal time windows to customers.

Machine learning and time at door

The time is takes to deliver furniture depends on factors such as it’s size, the proximity of parking and the number of staircases en route to the customer. These variables made it previously impossible to predict time at door and these ‘unknowns’ resulted in inaccurate schedules. We built a machine learning model that took product, telematics and locational data and predicted, based on previous orders, how long time at door would be. These time at door predictions were then fed back into our algorithm — improving the accuracy of future schedules.

Building a scalable, accessible solution

Building breakthrough technology is hard, getting people to use it is harder. We regularly tested our system, it’s features and usability with the technical teams, operators and drivers of DFS to ensure it met their operational needs. We worked together to build business intelligence dashboards that enabled operators to make more informed decisions. Time slots for instance, were coloured green, orange and red; and operator’s encouraged customers to choose ‘green’ slots — which if selected, would result in a more eco-friendly delivery schedule for DFS. The solution was hosted on the Cloud — this ensured the solution would scale with demand, and our algorithms had the computational power to run at the speed and frequency that the system required.

Efficiency gains

Against leading vendors, our solution increased capacity by around 5% and reduced idle time by 10% — this meant DFS could use less vans, and fewer drivers to make more deliveries. These marginal gains result in significant savings across thousands of orders every day. It has also enabled DFS to offer time slots, which improved customer experience and loyalty.

“The speed and effectiveness of the rollout has been outstanding. I’ve overseen many system implementations in the last 10 years — and I’ve never seen anything as stable and as user friendly from day one,” says Daniel Wallace, Head of Supply Chain at DFS.