The company

Focused on retail and the distribution of car components and spare parts in a B2B (business to business) and B2C(business to customer) model, this company is part of a multinational group with a global presence. In Portugal, they have two main warehouses (one in the north and another in centre/south) to cover deliveries in almost any part of the country with challenging delivery times, same day deliveries and up to four windows per day. Final customers or local workshops can be served with distribution, or they can visit one of the stores.

With more than 4.000 active clients and almost 10,000 units dispatched every day, delivery times can be as low as 1.5hours.

The challenge

With almost 600 delivery points per day(always changing from one day to another) and having to negotiate and work with14 different transporters, the operation is complex.

Identifying the clients’ profile is a difficult challenge but represents a huge opportunity to adjust service level to each client.

Understanding and categorising the client's profile is the first step to design a well-structured and flexible programme that aims to maximise costumers’ satisfaction while reducing costs by adjusting the service level (reduction of transportation costs).

In order to find clusters of clients, key variables should be taken into consideration: how recently they bought something, their geographical location, frequency, product mix and overall volume.

The final challenge is to define a distribution network that fulfils the customers' needs at a minimum cost. It is crucial to efficiently design both reactive and predictive routes, encompassing different delivery points every day, always taking into consideration existent routes and different restrictions from service providers.

The Approach

Customer segmentation

In order to fit the right service level to each client, clusters of clients should be identified. Only after this classification, business decisions can be made to define the service of each one of the clusters. In fact, each one of the groups identified have different behaviours, different expectations and represent different levels of profitability.

Data collection and data preparation of relevant business data was the first step for the customer segmentation process. Understanding where to get the data and translating that into business knowledge is as important as identifying which data should be gathered.

Data mining techniques were used for analytical customer segmentation and a density-based clustering algorithm to cluster customers.

This customer segmentation permitted the allocation of clients to the right transportation route, applying a mathematical model. Routes were previously optimised and each one represented a different combination of transporter, number of delivery points, capacity, and frequency of delivery.

Distribution network

With the objective of minimising transportation costs, an optimisation of the distribution network was needed. It is clear that frequent adjustments to the distribution network should be made not only because active customers’ geographical distribution is always changing, but also because of the overlapping of transport services providers which can be found and can result in great savings opportunities.

Using density-based clustering algorithms to cluster customers based on their geographical location permitted grouping clients into macro regions. For each one of the regions, an estimation of the last mile delivery time was made. It is important to notice that inside of each cluster, not every client is visited every day.

Achievements

The tools developed not only enable flexible scenario analysis that will allow fast and easy adjustments in the future, but also a decrease in transportation costs of almost 20% without negatively impact customer service level and satisfaction.

This reduction of transports cost came from the adjustment of previous errors in the definition of service levels of certain customers and the negation with transporters by redefining routes and service areas (overlapping was identified and eliminated).

‍