Case Study
Swiggy splits every order into stages (restaurant prep, rider assignment, travel to the restaurant, pickup and last mile) and predicts each with machine-learning models. It assigns the rider who can reach the restaurant just as the food is ready, batches orders going the same way, and for 10-minute Bolt it only uses nearby restaurants and quick-to-prepare dishes.
You order a biryani, and about half an hour later a delivery partner is at your door with it still hot. In some areas Swiggy even promises food in 10 minutes. Making that happen thousands of times a minute is not just about fast bikes. It is a prediction problem, a matching problem and a timing problem, solved with data. This case study explains how Swiggy's app works, how a rider gets your order, and the technology behind fast delivery.
To promise an arrival time, the app splits every order into stages and predicts each one:
| Stage | What happens | What the system must predict |
|---|---|---|
| 1. Order placed | You pay; the restaurant gets the order | Will the restaurant accept quickly? |
| 2. Food preparation | The kitchen cooks and packs | How long this dish takes at this restaurant, at this hour |
| 3. Rider assignment | A delivery partner is chosen | Which free rider can reach the restaurant just as food is ready |
| 4. First mile | Rider travels to the restaurant | Travel time in current traffic |
| 5. Pickup | Rider collects the parcel | Waiting time at the counter |
| 6. Last mile | Rider travels to you | Travel time, plus time to find your gate and floor |
The ETA you see is roughly the sum of these stages. Machine-learning models predict each one from past orders, time of day, weather and live conditions, and the ETA keeps updating as the order moves.
The key idea is timing. If a rider reaches the restaurant too early, he waits and the system loses his time. If he reaches too late, your food goes cold. Public talks and engineering write-ups about Swiggy describe an assignment system that looks at:
Assignment is a continuous calculation, re-run as new orders come in and riders become free. A Swiggy data scientist's talk at The Fifth Elephant 2018 listed the building blocks: demand forecasting, real-time pairing of orders and riders, delivery-stage predictions, order batching and controls for the supply side.
When two orders come from the same restaurant or area and go in the same direction, one rider can carry both. Batching saves rider time and lowers cost, but it can add a few minutes for one of the customers, so it has to be used carefully. Swiggy offers an opt-in Eco Saver option where customers agree to batched delivery.
Swiggy launched Bolt in October 2024. The speed comes from design choices, not from riders driving faster:
Swiggy reported that Bolt contributed over 12% of food delivery orders within six months of launch.
Grocery in 10–15 minutes works because the products are already close to you, in small warehouses called dark stores. Swiggy had 1,021 of them at the end of FY25. Swiggy's VP of Data Science, Goda Ramkumar, told Business Today that deep-learning models predict delivery times and that the system even picks the right vehicle: bicycles for close deliveries and motorbikes for longer ones. Demand forecasting decides what each store stocks.
Peaks are the hardest part. When it rains or a big match is on, orders jump and riders slow down at the same time. A 2018 Swiggy talk named IPL weekends, rain and New Year's Eve as examples, and described combining time-series forecasting, operations research, machine learning and simulation to keep delivery promises. In practice you see this as longer ETAs, surge or rain fees, or some restaurants showing as unavailable for a while: the app limits how much it accepts so it can still deliver what it promised.
| Job | Technology idea |
|---|---|
| Predict delivery time | Machine-learning and deep-learning models trained on past orders |
| Assign riders | Real-time matching that balances rider distance and food-ready time |
| Combine orders | Batching logic for orders going the same way |
| Plan for peaks | Demand forecasting, simulation and capacity controls |
| Live tracking | Rider app sends GPS location; your app shows it on a map |
| Show you the right food | Personalised recommendations from your order history |
| Help restaurants sell | In-app advertising and analytics for partners |
Want this kind of speed and tracking in your own lead funnel? Book a free strategy call or explore our services.
It predicts each stage of the order (food preparation, rider travel to the restaurant, pickup wait and the last mile to you) using machine-learning models trained on past orders, then updates the ETA live as the order moves.
The system looks at where free riders are, how long each would take to reach the restaurant and when the food will be ready, and assigns the order so the rider arrives close to the time the food is packed.
Bolt only covers restaurants within 2 km and dishes that need little or no preparation. Swiggy says riders are not told which orders are Bolt orders, so they are not pushed to ride faster.
A small warehouse that is not open to walk-in customers and only packs online orders. Swiggy Instamart had 1,021 dark stores at the end of FY25, which is how groceries can arrive in about 13 minutes.
Keep reading
When you tap Book, Ola finds online drivers near your pickup point, ranks them by real road ETA (not straight-line distance), prices the ride for current demand and offers it to the best driver. Its own Ola Maps platform (built on OpenStreetMap, OSRM routing and live vehicle GPS data) powers pickup points, routes and ETAs.
How much does digital marketing cost in India? A clear 2026 breakdown of agency fees, ad spend and DIY costs for small businesses, with sample monthly budgets.
A practical 12-step Google Business Profile (GMB) checklist to rank higher on Google Maps, get more calls and win the local 3-pack in your city.
Book a free 30-minute strategy call. We audit your online presence and suggest a plan, with no obligation.