One of the largest challenges facing small and medium restaurants is that of competing against big chains with better information infrastructure – an advantage that allows the big chains to plan resource allocation more precisely based on demand and other factors. iChef provides this platform for small restaurants “making enterprise level technologies affordable and understandable for small restaurants”. As Taiwan is a subtropical country, often affected by heavy rains and tropical storms, there is a question of if and how weather can affect sales; and how iChef can help its clients to mitigate or capitalize on this challenge.

By more accurately forecasting sales (considering externalities), managers are able to better
manage their resources to avoid losses and maximize revenue. By integrating data from iChef Taiwan and World Weather Online – and adding other external information such as holidays - into one database, we were able to propose a model that includes this information on weather and holidays and forecasts daily sales. We benchmark the target model against other simpler models like seasonal naïve, exponential smoothing and another linear regression without external information to compare its performance and consider if the cost of deploying a more complex predictive model is worthwhile for the business.

We find that at the level of daily sales per restaurant, there is little additional predictive accuracy gained by the inclusion of holiday and weather data. However, when we consider a single three hour meal time (such as 5 – 8pm dinner time), we find that the inclusion of weather and holiday data can provide us with higher predictive accuracy and improved forecasts over simpler methods. Since the collection and maintenance of data such as weather forecasts and annual holidays is extremely convenient and easy to conduct, we advise that this data be collected and considered for the prediction of daily sales per restaurant.