In optimising the group’s supply chain, inovex has worked particularly intensively with REWE’s IT subsidiary, REWE Digital. This arm of the company is responsible for all the REWE Group’s strategic online activities and aims to become the leading provider of online solutions in all REWE’s associated supermarkets and supply warehouses. These include those distribution warehouses that REWE set up specifically for their home delivery service, as well as the group’s in-store delivery and collection points all over Germany.
Fulfilment: Extended Supply Chain vs. Retail
Traditionally, logistics chains end at the supermarket. The hallmark of a delivery service is the additional step in the process: delivery right to the customer’s door. This ‘last mile’ plays a decisive role in ensuring customers’ satisfaction with the service, making it a key factor in the success of the business model as a whole. In the course of inovex’s collaboration with REWE Digital, we have implemented multiple data-driven use cases aimed at improving KPIs such as REWE’s Net Promoter Score or the availability of items in its online stores.
In the ‘Estimated Arrival Time’ use case, an intelligent information service both provides REWE’s end customers directly with automated information and gives customer service staff at REWE Digital tools they can use to proactively inform customers of delays or to provide specific order information.
The Prognosis Model and Interval Prediction
When a customer places an order with REWE’s delivery service, he or she fills their virtual basket online in the web store or in the app. At checkout, he or she can then select one of several time slots over the next few days for the order to be delivered. The delivery window selected serves as input for REWE’s route planning system, which plans the delivery routes and distributes the orders among the group’s delivery vehicles. Providing the most accurate estimates and communication of projected delivery times is a key factor in ensuring that the customer feels well informed and satisfied. In particular, communicating intervals substantially narrower than those originally selected is a way of offering customers considerable added value. If, for example, a customer chooses to have their items delivered between 2 and 4pm, he or she will welcome the information that their delivery will arrive between 2:35 and 2:55 pm.
The next step involves determining when to update the customer. In this use case, determining a blanket time is impossible, because it depends on the reliability (or unreliability) of the prediction. It is easy to imagine that, the further away the delivery time, the more difficult it is to provide an accurate prediction. And, in fact, this is true; the statistically valid intervals predicted automatically get smaller as the planned delivery time approaches. This feature is used to give specific customers information at the exact moment when the prediction is sufficiently reliable to be accurate AND the information provides the customer with added value. If everything goes punctually and according to plan, the customer will receive a message as soon as the delivery time can be predicted to within 20 minutes.
As soon as a delay is foreseen, the second user of the information generated steps in – REWE’s customer service department. Drivers’ estimated arrival times change over the course of the day, due to events such as traffic jams, for example. This can push back, or delay, customer deliveries. In such cases, the customer service department is informed of the situation. The staff can then contact customers to explain the reasons for the delay and to arrange compensation for the late deliveries.
Agile Data Science – Implementation with Scrum
The project was developed in an agile manner using the Scrum framework as an organisation model, with the current solution evolving gradually over several iterations. We believe that the Agile method was decisive in ensuring the successful implementation of this use case. When it comes to Data Science solutions, in particular, it is difficult to assess the project’s complexity at the outset. The power – and thus, the benefit – of prediction models cannot be measured until they have been developed.
The project focused on providing customer benefits through easily implementable solutions. Not only were the delivery predictions constantly improved during the course of the project, but new features were successively implemented.
An iterative approach is even more critical to the success of Data Science projects than it is to traditional Software Development. During each cycle (see graphic) the result is evaluated to determine whether it works well and is stable and beneficial. Only once all three requirements are fulfilled is the solution deployed.
- Agilely developed Data Science use case
- Enhanced prediction of statistically valid intervals
- Two user groups: REWE Digital’s customer service department and end customers
- Fully automated, customised text messages to customers
- Enables customer service staff to proactively inform customers of delivery delays
- Increase in customer satisfaction as measured by Net Promoter Score