By Anirban Sen & Jayadevan PK
Photography & Design by Vasu Agarwal
April 18, 2019
The sprawling ninth-floor office in a tech park in Koramangala, a southeast Bengaluru suburb, could easily be the digs of a large tech enterprise such as IBM or Infosys. Rows and rows of coders hunched over their laptops. Black workplace chairs, white tables with frosted glass partitions. Conference rooms with print outs of screen mock-ups jostling for space on glass walls with marker-pen scribbles and Post It notes.
This is a code factory with a difference — it is the engine that delivers thousands and thousands of food orders every day in over 100 Indian cities.
Welcome to the headquarters of Swiggy, the food delivery app busting the charts in India.
But Swiggy is not just any average food delivery venture. Its business is valued at over $3 billion – ranking it among the top five startups by valuation – and is active in over a hundred Indian cities. It counts the likes of Naspers, Tencent, DST Global, and Accel Partners among its top investors.
Back to the Swiggy corporate office. It is literally bursting at its seams, with hundreds of engineers, data scientists and product managers crammed close.
A large overhead LCD screen flashes names of new joinees every few seconds. Another screen above a huddle of engineers in the “RNG team” – short for revenue and growth – has a graphic highlighting spikes in traffic and other demand projections.
Accentuating the kinetic look-and-feel of the corporate quarters is the IPL cricket season, which has forced Swiggy’s top executives to create specific, designated “war rooms” where employees and executives strategise and brainstorm on newer ways to handle frequent spikes in user traffic. The IPL season is peak food delivery time, typically, and the business gets a boost with offers like discounts or cashbacks on orders made at the exact moment Andre Russell hits a six.
As if that were not enough, the top managers at Swiggy are scrambling to finish year-end appraisals of hundreds of employees. Swiggy employs about 5,000 odd people now — not including thousands of delivery agents ubiquitous in black t-shirts with the orange Swiggy logo.
Outside the Swiggy battle stations that vacuum terrabytes of data, India’s food-tech wars continue to scale new peaks, with Swiggy attempting to fend off and sweep aside its closest competitor, Zomato, and others such as Uber Eats and Ola’s Foodpanda.
Very soon, Swiggy will move its headquarters to a much larger space in Bengaluru. And it won’t be the first time it has done that – as Swiggy’s scale grows in leaps and bounds in the past four years. The company has been forced to relocate headquarters at least three times.
Seeing where it is today, it’s hard to fathom how Swiggy came out of the blue in 2014 and turned the business of food-delivery on its head canoeing its way through an Internet funding boom and the subsequent bust that had several food-tech ventures falling by the wayside.
So, what separates Swiggy from the rest?
Swiggy was not an early entrant in the food delivery business. Far from it.
In 2014, Zomato was the foremost venture in the business of food-tech and was attempting to desperately crack the market. As were a dozen others like TinyOwl and FoodPanda.
But none of them succeeded. From their failures emerged the Swiggy blueprint. But not before teething troubles.
Swiggy did not operate an app during its initial months. And even after it was released, the mobile app was a far cry from its present-day spiffy avatar – even though it was miles ahead of competitor apps then from a usability perspective.
What it lacked in tech sophistication at the time, it made up for other bold strategic bets at the time – most significantly, the decision of the founders to invest heavily on building its own in-house logistics. Which roughly translated to building their own delivery fleet.
While others such as Zomato were following the marketplace model and simply connecting users to restaurants and outsourced deliveries to third-parties, Swiggy figured out quickly that the model was unsustainable — simply because it compromised customer experience, as delivery executives from third-party operators provided inconsistent quality of deliveries.
For founders Sriharsha Majety and Nandan Reddy, in particular, building their own logistics operations for Swiggy came naturally — prior to starting Swiggy, the duo had attempted to build a technology product called Bundl that focused on logistics and helped ship products across the country.
Next, they realised that their technology needed to stand out from the rest of the pack. To that end, they swiftly brought on former Myntra tech executive Rahul Jaimini as their third co-founder. Myntra is a fashion retailer part of the Flipkart group.
Over the years, Swiggy has scored well on three counts over rivals on the tech front.
Firstly, the Swiggy app has the best user interface among all food-delivery services. Unquestionably. What is also not very widely known is that Swiggy’s tech teams operate and monitor fours apps — one each for users, restaurant partners, delivery executives, and an in-house team app. All these apps talk to each other at any given point of team, ensuring seamless coordination between all parties involved for smoother deliveries.
Secondly, Swiggy took a punt on data sciences earlier than others — it crunched the vast troves of data it collected from users and bet heavily on personalisation, a feature that made its app far more easier to navigate than others.
Thirdly, Swiggy invested big in “cloud kitchens” — the practice of setting up kitchen spaces in areas where restaurant partners don’t operate, with the idea of being accessible everywhere. Swiggy wasn’t the first in the game to talk about cloud kitchens, but it perfected the model. No restaurant would be out of reach for users in any part of the city. Well, almost.
And the rest is history. Such was the gulf between Swiggy’s user experience and the rest, that older rivals such as Zomato were eventually forced to copy the Swiggy playbook. The two even briefly held merger talks, although those conversations did not lead anywhere.
Unlike most consumer internet startups that connect the user to a service provider (think Uber), Swiggy is not just a two-way market. Uber, for instance, is a classic two-way market where passengers are connected to drivers on the move. Swiggy is a three-way market where there’s a customer, the restaurant (service provider), and the delivery fleet run by Swiggy. Which means there are more moving parts: a user trying to decide what to eat, a delivery executive on the move, and a restaurant readying the next order.
About three years ago, the company had a basic looking app which helped users order food online. Around this time, the company started asking itself how to rank restaurants better? How can it personalize the experience for a user and his tastes? Is there a supply demand mismatch? How long does it take to deliver?
Talking about restaurant choices, Anuj Rathi, VP of product management at Swiggy, says: “It’s a non trivial problem and it takes a lot of machine learning to figure out how do we balance a user’s experience with the mass user experience. All this needs to work in real time.” The company also optimises a user’s experience based on various other parameters such as the time of day, previous orders, payment methods, and so on.
Once an order is placed, questions are about assigning the right delivery executive, the best route to take, and so on. In the meantime, the restaurant needs to start preparing the food ordered so that it is, ideally, ready by the time the delivery executive reaches the location. “The idea is to get the food to the user hottest, freshest and in time,” says Rathi, who was one of the earliest product managers at Flipkart.
Before the launch of every new feature, many different versions of a user’s journey are tested out.
“When you are hungry, you are also probably ‘hangry’. That’s why we need it to be a really simple user experience,” Srinath Rangamani, who heads design at Swiggy. Users very often start searching for food when they are hungry. So for the company, it’s not only enough to reduce delivery time but also to reduce the discovery time.
“It takes a lot of machine learning to balance a user’s experience with the mass user experience”, says Anuj Rathi, VP of product management at Swiggy. He was one of the earliest product managers at Flipkart.
Swiggy Pop was born out of some of these insights. It was taking more than 10 minutes and several steps at the time to choose and discover the dishes they wanted to order. “How can we build a product that can make ordering a 3-step process,” says Rathi. At present, most users are shown similar items within Swiggy Pop but as the company personalizes its product more, the choices will be different, Rathi added.
Swiggy broadly started applying machine learning techniques in two areas around two-and-a-half years ago. One, in discovery and personalisation. The big question here is: how to get the best food to a particular user in time for that occasion depending on the kind of persona. Then there’s the post-order journey which involves delivery executives and restaurants.
Two, bringing down the time to deliver is another area where Swiggy has spent a lot of technology resources. Optimising the route a delivery executive needs to take, factoring in the time taken by the restaurant, traffic and weather conditions, and also the overall availability of delivery executives in an area.
Even though metros still form a core part of Swiggy’s overall business, the company is starting to train it sights on the next 100 million potential customers, lying deep in the Tier-2 and Tier-3 heartlands of India.
Artificial intelligence is at the core of all things future at Swiggy as it aims to expand rapidly from beyond 100 cities, says Dale Vaz, head of engineering and AI at Swiggy. “Just on scale and complexity of the business perspective, we just will not be able to manage the business the way we were doing it the old way. We have to enable our teams with smart systems — part of what we’re doing is looking at AI as an existential need for Swiggy. It is not something that we’re looking at as just a cool thing to do,” says Vaz, an Amazon veteran. He rates AI as about the only thing Swiggy needs to do to scale in the long term — “if we have to go from 100 cities to 300 or 400 or 500 cities”.
“We’re preparing for that future,” he adds.
“We’re looking at (personalisation) from a broader lens of 100 million customers. We can add language as another dimension of what a customer needs. Maybe some customers like to place an order by voice, while using the app. Some people want to chat more. So, we’re trying to understand our customers better, their needs and building those products for them. That’s the journey we are on,” sums up Vaz.
AI is an existential need for Swiggy, says Dale Vaz, head of engineering and AI at Swiggy. Vaz is an Amazon veteran.
As part of this future, Swiggy is making a series of bold, futuristic bets that will give it an edge in the next phase of the food wars.
To that end, it has set up a new applied research team comprising of leading data scientists and AI experts that will focus on futuristic moonshots. There are four people on that team, at present. That number will grow.
This team will work on three key areas — voice, natural language and vision.
“Even as we solve the current problems, there are some of these deep, hard, unsolved problems that require a much more experimental and long-term research-like activities…..We also recently acqui-hired a start-up called the Kint.io that is now part of Applied Research. We’re continuing to look for more talent that will help us with that,” says Vaz.
He insists “voice, vision and natural language… will give us leverage over the business.”
Most of these moonshots are still at an early stage and Swiggy declined to share details on the kind of projects it is working on. But there are other new initiatives that Swiggy is working on right now that will be rolled out soon. One of them is around creating a Netflix-like experience for food delivery.
The concept of a “food graph” is relatively new in the Indian food-tech business. Swiggy, however, is attempting to make that part of its daily experience for users.
“Catalogue intelligence is one big area that we’re looking at. One of the things we’re looking at is how do we identify understand food better — which means, on a conceptual level, there is a food graph, which is the equivalent of a social graph,” explains Vaz.
Suppose A knows B, C knows A. It is possible C will know B though the link may be lighter. Vaz draws a similar analogy for food with ingredients, recipes, cooking style, calorie value etc. “What we’re trying to do is give customers a Netflix-like experience for food discovery — on Netflix, you get suggestions based on movies you’ve already watched. On Swiggy, we are trying to bring a really nuanced understanding of what the food is and recommend accordingly. So, we want to get to that level of detail where we then understand food better and then start linking them together on what we call the food graph,” he explains.
Swiggy is also putting in place a new machine-learning model that classifies and separates vegetarian and non-vegetarian food. All in, it’s a relentless pursuit of crunching and bringing down delivery times — Swiggy now delivers an average dish at roughly 32 minutes per dish, which is down from the average of 42 to 45 minutes it used to take when it first started out.
Already Swiggy has automated classification of vegetarian and non-vegetarian dishes by inspecting the contents of a dish. “The next thing that we’re doing is launching a model that does cuisine based classification…what that’s helping us do is improve our delivery times. When we are estimating a time to prepare an order, we are using this as an input. Previously, we used to say, this is a restaurant, this is an order, it’ll take 20 mins. Now, we’re actually able to say that this is a restaurant, but the order is going to be for a pasta versus a burger. So we know that pasta will take more time to prepare, so we add a little more time to our estimates. So the estimation accuracy on the order prep time has gone up by bringing in the dimension of what kind of food it is,” says Vaz.
Other inputs into dish classification are also in the works. “We’re actually working with a company to get recipe content into our database, so that we can start to mine that and build that into the next level of data,” says the Swiggy head of engineering.
Clearly, Swiggy is in the swing of things with data and machine intelligence. And that edge looks set to help it power even faster ahead of its peers as it steps into new adjacent businesses like Swiggy Stores.