Serving Up Data

Mastering the Menu with Machine Learning and Market Trends

Thus far we’ve addressed two ingredients of Rebel Foods’ recipe for creating an internet restaurant empire. The first is their ownership of the means of production - they operate hundreds of ghost kitchens. The second is what we called the “virtual dining hall” model in which Rebel deploys multiple restaurants from one kitchen. Third and potentially most impactful is the company’s use of software to rapidly iterate new food concepts. Jaydeep Barman, Rebel’s CEO, frames Rebel to be to restaurants what Tesla is to the automotive industry. That is to say, Barman is running his food business like a tech startup. 

During Google’s early days, the product team faced a narrow decision: what shade of blue should the links be? To identify the optimal color the team took a hyper-methodical approach by putting users into buckets representing 2% or so of the total and displaying one of about forty unique shades to them. #1A0DAB produced the highest click yield and so #1A0DAB it was. Over two hundred million dollars of revenue was produced from this exercise in what’s become common practice across the tech industry: A/B testing. That is, deploying numerous iterations of a product in test markets to optimize the approach. 

Rebel Foods’ culinary innovation center might as well have a Google sign across the entryway. Their process of originating and testing new restaurant concepts mirrors what the Silicon Valley behemoth would likely come up with. First step is data. The innovation center leverages machine learning algorithms to detect trends, missed opportunities, and new markets in the food sphere. Once an opportunity is identified, Rebel comes up with a provisional menu optimized to take advantage. The menu along with cooking instructions are then handed off to five or so Chief Delight Officers (general managers) to deploy in their kitchens - five out of hundreds. 

That’s where the process starts. What follows is a period of testing and measuring. Testing and measuring. Rebel will continue to iterate on a new restaurant until one of two outcomes is reached: the menu reaches product market fit or it’s discarded as a null concept. Stipulations for PMF are codified: a delivery app rating above 4.2 and an net promoter score of above 40. Rebel’s hit rate on these metrics is impressive: seven out of twenty restaurants tested during 2018 surpassed them and were deployed to the entire kitchen network. Regional tailoring is leveraged in this model as well, Rebel can code their menus to certain taste profiles depending on where the production kitchen is located. They can A/B test their audience.

As Tesla deploys updates to a car, Rebel deploys software updates to their restaurants. When holidays come around, Rebel’s data scientists identify the special items that will play best. Leveraging their massive presence in the pan-Indian supply chain, they can deploy these concepts within six weeks. It tracks, then, that Rebel’s job board has listings for data engineers, UX designers, and AI researchers. Rebel Foods is a software company first.