How would you describe your experience at Swiggy over the past 6 months? How is it different from you past stints in the space of engineering?
It’s been fairly action-packed — from learning a new domain space and working with the team to create a two-year vision, to engaging with stakeholders to put together the 2019 roadmap, and debriefing and planning the charter with the team. The team here has been amazing, and having truly passionate and helpful peers have really helped me ramp up very quickly.
The unique difference at Swiggy is the on-demand nature of our offering and tight delivery promises that we adhere to, combined with the hyperlocal nature of the business. Our systems also need to deal with the reality that different localities within a city will have very different characteristics from an operational point-of-view. For instance, the model we use to optimise delivery from a restaurant serving Biryani at Indiranagar (Bengaluru), may be very different from the model needed to deliver a Biryani in a city like Surat (Gujarat), and so on.
Our Delivery Partners also depend on us for their livelihood. A very humbling thought is that when our systems have issues, it impacts the earnings of our Delivery Partners, which has a repercussion on their families. All this in turn affects our Customers who get hungrier with every passing minute of delay. Meeting the high expectations of our Customers and Delivery Partners requires putting a lot of thought into ensuring the robustness of the core logistics system, and that’s what we are scaling, and continuously improving, every day.
We know that behind the success of Swiggy’s delivery model, there’s a robust engineering team that keeps the wheels turning. Can you shed some light on what the team does to satisfy the hunger of millions of Indians across the country?
AI and Data Science models are at the heart of all the systems that we build at Swiggy. Having an on-demand fleet with tight delivery times (30-45min) poses unique challenges. When customers open the application, we process hundreds of restaurants to surface and then determine the predicted delivery times (promise to the customer) of each restaurant. In order to perform this prediction, we build models using historical data for travel time, and also to estimate the food preparation time. Once the customer places the order, we compute over a million combinations to find the right set of Delivery Partners who will be assigned the orders. This is a multi-objective function that is designed to maximise the efficiency of the fleet, while also honouring the promise made to the customer. During the assignment process, we determine which orders can be batched together; whether we need to delay assigning an order to a driver; make assessments on orders that have a high preparation time; and multiple other factors. To complicate matters further, during this time the drivers might be in constant motion, and may also be logging in or out of the system. All these factors make it imperative that we perform these calculations in less than a minute.
There are several other complexities the team deals with. Take order batching for instance. This activity has a single driver assigned to deliver two or more orders from a restaurant to different customers. Batching requires us to deal with some unique set of challenges. For example, it may seem intuitive that you would batch orders of two customers (C1, C2) who are in close proximity in order to minimise time to deliver the orders, as compared to two customers (C2, C3) who are further apart. However if C2 and C3 are part of the same apartment complex, and C1 is in a different apartment complex, then batching orders of C2 and C3 is actually a more efficient approach. At Swiggy, our location services team is chartered to leverage the Delivery Partner’s location ping data (gathered over the last 4 years) to derive this depth of information. To be able to do this at scale, requires the team to rely heavily on Data Science models that try to optimise multiple objectives simultaneously.
To provide a great customer experience, we have chatbots to automate responses and exception processing flows for customers, technologies to enable our customer service team and provide live tracking of an order for the customer. We are also investing heavily to detect and automate interventions, such as in cases where an order might breach the delivery promise. And finally, the technology that the team builds also powers the systems that sustains the Delivery Partner’s career-cycle. From sourcing and on-boarding, to video training and payouts are all enabled through the tool, and that too at the scale of over a 100 cities, and expanding rapidly every day..
The problems the team deals with are of extremely high levels of complexity. We have seen a lot of interest from PhDs in universities across the globe who wish to partner with Swiggy in researching these particular problems in the space of hyperlocal logistics.
Tell us a bit about the Mayank Talati that Swiggsters are not too familiar with. What keeps you ticking outside of work?
I love reading science fiction and fantasy novels, watching movies and travelling to new places. My favorite sport was sailing. While in the US, I would head out at least every month to hit the waves. The rougher the sea, the better! Unfortunately, I haven’t been able to indulge in this hobby after coming back to India.
Another interesting tid-bit: what got me interested in computers initially wasn’t software, but assembling my own computers. To this day, I still have a desktop assembled from parts bought online.
You are stranded on an island and are allowed to eat only one dish for the rest of your time there, what would that be, and why?
That’s a hard question. I like to have a variety of food. Eating the same thing would be depressing. While doing my Masters, pasta was the easiest dish to make when pressed for time (which of course was always). I ate pasta regularly for 18 months; as a result I still detest pasta 20 years later! If I had to choose only one dish though, I would choose pizza; that way I can at least vary the toppings (laughs)
Any message that you would like to share with the world about what it takes to be a Swiggster?
With a hyperlocal, on-demand fleet and a business that is rapidly growing, Swiggsters are working on some of the most challenging problems in e-commerce. To be successful at Swiggy, you need to be a great problem solver, who is comfortable with experimenting, learning, delving into data and exploring uncharted territories. At Swiggy, we encourage unconstrained thinking in order to sustainably scale our systems, while maintaining a razor focus on continuing to improve the customer experience. Maybe not cut out for everyone, but not everyone makes the cut.