Work
Forecasting online food orders for a virtual restaurant
We helped a virtual restaurant forecast sales at 17 of their sites; helping to reduce overstock and understock, food wastage and save costs.
Customer
A fast-growing startup in the innovative virtual restaurant sector industry.
Problem
Customers can buy food from one of the company’s innovative restaurant brands via three delivery platforms, including Deliveroo and Uber Eat. Overheads associated with running a restaurant (namely, space for tables and waiting staff) are drastically reduced. Supplying their kitchens with the right quantities of ingredients is crucial to their business. Understock and they miss out on potential sales; overstock and food is wasted – at large cost to the business. We were asked to provide a method of forecasting sales at all 17 restaurants.
Solution
We established data pipelines to collect all past orders and update with orders as they came in from all three platforms. This is a fundamental shift in how the company interacts with its data, using computation rather than manual analysis. Significant data-matching between platforms and languages was required. The result is that the company now possesses data that is a more accurate representation of sales than can be offered by any of the delivery platforms directly. It is stored securely in a series of SQL databases hosted on Amazon Web Services.
In addition to the forecasting work, Faculty implemented web-scraping of some of the delivery platforms to allow the company to keep track of when certain items are out of stock in their kitchens.
Impact
Despite the small amount of data (many restaurants had been in operation for less than a year), forecasts from the models are a significant improvement on the company’s previous system. As the company grows, it can scale the forecasting process up to any number of restaurants, with no increase in human supervision.