This is a project presented by Vivek Jodmote, Anish Karve & Arup Bansal, PGP DSBA students, in the AICTE Sponsored Online International Conference on Data science, Machine learning and its applications (ICDML-2020). A follow-up paper was published in the conference journal.
Cloud kitchens, also known as virtual kitchens, are take away food outlets having no dine-in facilities. These kitchens normally function as production units with sufficient space available for preparing different food cuisines, and the food is usually ordered solely online. In the Cloud Kitchen business model, the food order normally passes through three stages, namely,
- Order sourcing
- Order preparation
- Order delivery
Under the first step of order sourcing, the operator sources food orders either on their own or through other food aggregators such as Swiggy or Zomato. Compared with the dine-in restaurants, cafes, or food joints, this business model benefits from lower real estate costs, better expansion opportunities, and lower overhead costs. However, on the other side, this business model lacks visibility, has limited marketing avenues, and overdependence on food aggregators.
A one-year-old cloud kitchen start-up was facing declining, and localized sales in pockets with few repeat customers. In this study, our learners proposed to help the cloud kitchen by identifying a sustainable competitive advantage and creating stronger brand visibility in the market using cutting edge business development and marketing strategies by applying several machine learning techniques. The study sample contained one-year data taken from the cloud kitchen, which included Order data, Menu data and transactional data of ~3000 orders. The sample contained order-level details such as date of order, time, food dish, type, order status, order payment, order source, discounts, customer location etc.
The study suggested a strong relationship of Cloud Kitchen Order Sales with several variables such as discounts, location and customer demographics and highlighted a higher dependency on one food aggregator. Analysis of order frequency distribution was mined to understand the customer base, repeat orders, area-wise order distribution, and sales opportunities. Exploratory Data analysis suggested that the sales were concentrated to a few locations which were due to the serviceability of the cloud kitchen. Market Basket Analysis and Association Rules were used to recommend a few Combo Meals Options based on the items bought together to improve sales.
A combination of loyalty programs for repeat customers, order recommendations, Combo Meals and targeted discounts were introduced, which would further translate into optimal improvement in the sales of the Cloud Kitchen. The study also showcased the potential of opening a new kitchen in a different location to increase customer reach.0