Exclusive: Read how Swiggy used AI and Data Science to grow its order value by over 200%!
Swiggy generates terabytes of data every week, which is then leveraged for delivery efficiency and to connect customers to the right restaurant. The order volume has also grown over 200%. The Swiggy app has metamorphosed from being an app that would just get the job done to an app that delights the user every time they come to place an order. Real-time, micro-optimization of dynamic demand-supply, done millions of times every day, says Dale Vaz, Head of Engineering and Data Science, Swiggy to EC’s Abhishek Raval
Please discuss the benefits from AI at Swiggy
Swiggy’s mission has always been to elevate the lives of urban consumers by delivering unparalleled convenience. We do this by operating a 3-way hyper-local marketplace where we match consumer demand with supply from vendors (e.g. restaurants, stores) and delivery executives (largely gig workers). We use AI / ML across this 3-way marketplace to deliver a wow customer experience, unlock business growth and drive operational efficiency.
AI and Data Science models are at the heart of all the systems that we build at Swiggy. At our scale and complexity, it is humanly impossible to run this machine efficiently without an AI intervention. We have gone past the stage where scenarios can be managed by just smart people. Now, we need systems in place to enable them to be smarter and scale faster. So, for Swiggy, moving from a human-intelligence based model to an AI-led platform was only natural. Without that, we won’t be able to keep up with our pace of growth.
We generate terabytes of data every week and leverage this data for delivery efficiency and to connect customers to the right restaurant. With our expansion across cities, our order volume has also grown over 200%. We have gone from being an app that would just get the job done to an app that delights the user every time they come to place an order. Real-time, micro-optimization of dynamic demand-supply, done millions of times every day.
How do you use AI on the vendor and consumer side ?
On the vendor front, we use AI for time-series based demand prediction models that help our partners plan for demand. We use ML models to reduce financial risk by detecting and preventing abuse (e.g. automatic suppression of Cash on Delivery payment option for high risk users).
On the consumer side, we use AI to deliver a personalized discovery experience to customers. This experience is powered by ML driven investments in:
- Catalog intelligence: using ML models to enrich our catalog with meta-data (e.g. classifying products as veg/egg/non-veg) and using ML models to build associations between products in the catalog (e.g. building a food knowledge graph based on probabilistic similarities between products)
- Customer intelligence: using ML models to segment customers (e.g. Affordability conscious customer) and track customer lifecycle changes (e.g. customer churn prediction)
- Relevance and Personalized CX: using ML models that leverage catalog intelligence, customer intelligence, user context information (e.g. time of day, user location, previous order history, points of intelligence) and real-time signals (e.g. last mile distance between the restaurant and customer location) to deliver a personalized listing of restaurants in the Listing page and Search
- Dynamic Driver Capacity: ML based models for demand and supply forecasting and Real-time capacity estimation at a zone level.
What kind of AI roadmap have you set for AI in Swiggy?
Our in-house Labs team is working towards innovative solutions for various consumer segments such as adding Intelligent layers: Catalogue intelligence, Location Intelligence, Dynamic driver capacity and Dynamic restaurant serviceability in real-time. Some of the initiatives currently underway are:
- Logistic Optimization: Optimizing the cost of delivery while ensuring that customers get their product within the promised delivery times. This involves a combination of ML based models and a mathematical multi-objective function for optimizing across multiple business and customer experience constraints.
- Easier Restaurant Discovery: The “neighborhood snapshot” on the home banner gives an overview of the deals of the day, as well as new and trending restaurants, so that users can discover twice as many new guides on the app. It also has persistent filters to help users find the most popular restaurants in the neighborhood, the best offers, and the shortest delivery time
- Greater Personalisation: Aided by machine learning, the app personalizes the list of restaurants that users can view, based on their past orders, searches, and interactions with the app. This reduces the time to arrive at the choice of restaurants and dishes by half
- Quicker Decision-making: Tags for restaurants (“newly opened”, “daily changing menus”, “repeat”) and dishes (“bestsellers”, “must try”) aid quicker decision-making. The app also provides useful information such as restaurant ratings, average cost for two and restaurant-specific charges, which helps users in making an informed decision
- Faster Reordering: “Repeat” tags for preferred restaurants and dishes, and a complete order history helps users reorder their favorites with just a tap
- Simpler Order Tracking: Live order tracking is one of the most loved features of the Swiggy app. The app shares the expected time of arrival (ETA) of the order in real-time
- Smart Kitchen: Optimizing in-kitchen operations through AI driven initiatives such as forecasting demand, inventory optimization, dynamic order prioritization, intelligent kitchen capacity management, food quality management through computer vision and more.
Looking ahead, our goal is to leverage ML to build innovative solutions for a Billion customers. This includes solving for unique needs of vernacular (local language) voice and text interactions, guided discovery through voice bots/chatbots/AR, hyper-personalization, faster and more efficient delivery optimizations, better food quality at lower prep time through Smart Kitchens etc. We will do this by doubling down on our investment in deep technology such as Neural Networks, Reinforcement Learning, Optimization Research, Econometrics, Edge computing, Network Simulators etc.