Each year, airline and rail companies serve millions of cups of tea, biscuits and sandwiches to passengers. Transport executives know that getting customers their choice of food and beverages at the right time along the journey increases customer satisfaction and reduces the risk of staff abuse.
These operators must balance the upside of offering a wide selection of products against the risk of stocking items that don’t sell, leading to wastage and incurring an opportunity cost by taking up space that could have been used for other goods.
Anticipating what passengers will want to eat and drink is a task that many passenger services companies still conduct manually, as part of a time-consuming ordering and reconciliation operation. We set out to help one transport operator to use automated, data-driven insights to improve the stock ordering process.
We worked with the client through the entire project lifecycle, from proving the validity of a data-based approach and securing internal stakeholder buy-in, through to project rollout.
The project harnessed the power of machine learning algorithms to create a predictive tool that identifies cost efficiency opportunities, leverages the company’s data assets to improve the accuracy of catering sales forecasts with a user-friendly system, and delivers real benefits on each journey.
Every day, our public transport client must restock food, drink and tableware items in an extensive mobile catering operation spread out among many vehicles over hundreds of miles. Making sure its thousands of daily customers get the outstanding service they expect while minimising food and packaging waste is a logistically complex challenge. Achieving success depends on precision ordering backed up by accurate forecasts. Given the size of the operation and the many variables that drive customer purchasing from day to day, it is no surprise that attempting to meet this challenge manually had proven time-consuming and fraught with error. Levels of waste were unacceptably high, while lack of product availability meant lost revenue opportunities and reduced customer satisfaction.
A fresh approach to the forecasting and order process was necessary to rein in waste while making sure customers still had the best possible catering experience on their journey.
To arrive at the right solution for our client, we harnessed three key areas of expertise: tackling complex data at scale, determining the machine learning approach appropriate to the goal and packaging a complex model in a user-friendly, interactive application.
We examined the impact of daily life on the purchasing choices of customers on the move. The weather, the academic year and sporting events were just some of the factors we weighed up in a series of 50 feature-engineered models that analysed more than 20 million rows of data. After analysis, we developed the model formulation that would deliver the required results and integrate into their systems.
For non-technical staff, who will use their new data-driven capability to make daily stocking decisions, it was essential that they had a straightforward interface that hid the underlying complexity of the solution while harnessing its power. Our UI experts created a visually appealing web app based on modern design and user experience principles.
This transport operator is now equipped to make data-led stock decisions. They possess a powerful, machine learning model packaged in a modern web application with a responsive interface. This is hosted on a secure cloud infrastructure integrated into their existing systems. They control the trade-off between understocking and overstocking on any given product, enabling them to maximise sales while making inroads into their UN sustainability targets on minimal food and packaging waste.
The client is expecting increased revenue from a more efficient service. The risk of ‘deadwood’ products taking up space is significantly reduced meaning a lower order fulfilment time and more customers served per hour. Customers have a more targeted choice of on-journey beverages and snacks, which leads to an increased number of purchases and therefore additional profit.
Having more confidence that their stocking levels match levels of demand yields multiple benefits. Customer satisfaction will increase as the risk of an individual’s desired food or drink being unavailable has been significantly reduced. Employee well-being is likely to improve with happier customers less likely to be abusive towards staff.
Early results from testing on historical data indicate an average reduction in sales forecast error of 15% compared with the manual process. The increased accuracy feeds into the forecast confidence bands produced by the new tool, making it a powerful aid to making the right stocking decisions. Our analysis indicates that using these forecast confidence bands to stock enough to cover demand 95% of the time, should reduce the instances of products selling out by over 70%. The risk of overstocking will be reduced by over 50%, leading to a daily saving of around 5,000 units over 50 journeys.
If you would like support to make more effective, data-driven decisions, get in touch.