Here’s how you get groceries in 10 minutes

Just before cooking, or sometimes in the middle of it, it’s common now to realise an ingredient is missing, and order it knowing it will arrive within 10 minutes. One user tells us that she ordered from her bedroom, and by the time she walked to the kitchen to start cooking, she was stunned to find the delivery guy knocking on the door. Given how busy our lives are, many think they can no longer do without this “instant” delivery.
But how are the delivery platforms doing it? This level of speed at scale is possible only because of AI/ ML. “The AI systems are making millions of decisions per second, and doing it concurrently at multiple levels – for the user, the store and the delivery partner,” says Goda Ramkumar, VP of data science at Swiggy, which offers the Instamart 10-minute service.
The moment you open the app, it locates you with location intelligence – it knows if you’re in a society apartment or an independent house. It then takes you through a personalised search experience, even as it sets expectations about the time of arrival (ETA). The search goes to the store/ restaurant, and rider that can deliver the fastest. Once you’ve ordered, and the products are packed and picked up, ML is used to find the fastest route. Till the time it reaches you, Ramkumar says, ML is proactively looking for delays or anomalies such as traffic jams or bike breakdowns to alert the consumer about delays.
ETA prediction is based on deep learning. There are multiple legs involved such as preparation or packaging time, delivery partner’s travel time to the outlet, and travel time to the customer’s home. Ramkumar, an IIT Madras alum who previously worked at Sabre and Ola, says complex optimisations have to run, and run fast, as orders keep flowing, in order to give the customer the ETA.

Data analytics is also key to forecasting demand in a particular locality, and ensuring right inventory in each dark store – where inventory is kept, packaged and delivered from. Rakshit Daga, CPTO and head of SaaS at BigBasket, which has the BBNow quick service, says AI-powered inventory forecasting models are particularly important for perishable goods, and helps in higher product availability and lower write-offs. “In the quick commerce model the inventory forecasting model is also slightly different from that for the regular slotted delivery model because the size of assortment you can support is different, as well as how you store this assortment,” he says.
Store location, layout
Data is used to decide the location of dark stores, to ensure rapid delivery. It is used to decide store layout – based on how frequently an item is ordered and which items are ordered together. This enables quick packaging. AI, Daga says, even helps decide the optimal placement and assortment of products based on factors like seasonality and upcoming festivals.
Analytics is used further in manpower planning in dark stores – forecasting peak and nonpeak hours, designing shifts so that right personnel are there at the right time. It is used to make sure the right number of riders are available at a point in time.
Assigning rider
Deep learning is used to assign the right rider. The rider needs to reach both the store and delivery address fast. Also, if the delivery address is nearby, a rider on a bicycle or e-bike may be fine, but if it’s some distance away, it might need a rider with a regular bike. So, Ramkumar says, there are parallel AI operations happening at multiple ends.
Daga says sophisticated algorithms factor in the existing traffic conditions, travel time, vehicle loading, and rider familiarity to determine the most efficient routes and rider allocations. “It’s such an exciting problem trying to figure out what the delivery time is and trying to deliver consistently an experience that is close to 10 minutes with all these variables, and at the same time keeping your riders safe by not putting undue pressure on them to drive fast,” he adds