A company providing a web and mobile application.
Each month the company processes over 80,000 orders in 20 countries through its app, which is the means of communication between a customer (typically a chef in a restaurant) and a supplier in the restaurant sector of goods such as dry and fresh ingredients.
When orders are confirmed and received on time, the app saves restaurateurs from spending hours contacting suppliers by phone or email. However, orders are not always confirmed by suppliers, and of these missed confirmations, some result in missed deliveries. The company asked Faculty to build a model that could identify orders that might be problematic.
We built a model to predict when a given order would be most likely to be confirmed by the supplier. That required including a number of variables, which had an impact on the confirmation process, including the time of day an order might be placed, the supplier’s typical average response time and the type of supplier (whether of dry goods or fresh goods, for example).
Fresh goods suppliers need to respond to urgent orders that come in after lunch service or dinner service, whereas dry goods suppliers can respond at a time more in keeping with a working day.
We adopted a random forest regressor model to predict the response time of 90% of orders correct to within 1.5 hours. This was a 50% improvement on the previous system, which was a simple mean response time for each supplier. The model was put into operation as an API in Faculty Platform, and runs four times a day.
As well as building the model, we integrated it with Slack. Any orders that have passed their expected response time are flagged to the operations team, who can then intervene. The improved results for orders benefits all parties: the company, restaurateurs and customers.