Surendra Kumar, 30, has an eye for detail and an instinct to match. A few weeks ago, Kumar, tasked with spotting fake Apple products at the New Delhi airport for Delhivery, a logistics company specialising in ecommerce deliveries, was betting on gut feel when he pointed out a package contained a fake iPhone.
When confronted, the seller indignantly unboxed the phone. Everything looked perfect — the phone had the trademark iPhone all-metallic body with the plastic wrap cover, and the box had its usual placements. But Kumar, who until a year ago used to scan Air India travelers and their bags for explosives, remained unconvinced and asked the vendor to switch on the iPhone.
“It was hilarious watching an iPhone booting Android software,” recalls Suvayu Ali, Data Scientist, Delhivery.
If you’re buying an iPhone online during the festive season, you want your delivery to be cleared by the likes of Kumar. Fake iPhones are a Rs 100-crore problem annually for India’s ecommerce companies, accounting for nearly half the overall losses incurred on fake phones, as per Delhivery. The Gurugram-headquartered company takes care of close to a quarter of the total number of ecommerce shipments in India, delivering about six million packages every month — second only to Flipkart’s logistics arm, Ekart.
Every day, Kumar and his team of 79 colleagues across Mumbai, Delhi, Bangalore, Chennai and Hyderabad, spot over 100 fake iPhones using X-ray machines, weighing scales and other tools. But nothing works like the human eye and brain in spotting a fake: how the box is wrapped in plastic, how the charger or cables are placed inside, the placement of the phone itself, the quality of the packing, a piece of cellotape holding the box together.
Kumar’s and his team’s brains hold answers that can shape Delhivery’s future — one ecommerce packet at a time. But humans can scale only so much and Delhivery is doubling down on a software algorithm mimicking the specialist brains of the fakes-spotting team.
The algorithm is just one of the projects that a team of data scientists at Delhivery, led by former entrepreneur and Facebook data scientist Santanu Bhattacharya is working on. The others include mapping addresses and reconciling inaccurate ones (a hugely complex task in India, as will become clear later in this story), warehouse stocking, prioritizing deliveries based on individual consumer behaviour, drawing out delivery routes in real time based on weather or riots, and, the biggest of them all — predicting products that will be in demand during festival sales.
As Sahil Barua, CEO and cofounder of Delhivery, puts it, the ecommerce logistics company is soon going to become a data firm, led by a bunch of data scientists hired from Facebook, Google and even CERN (the European Organisation for Nuclear Research that is known as the birthplace of the Internet and is home to the largest particle physics laboratory in the world).
“What we are doing with tech and data science is giving us a sustained competitive advantage,” says Barua, who in an earlier avatar worked with consultants Bain & Co. He doesn’t specifically refer to it but Delhivery’s bet on data sciences in its operations will likely be highlighted in its initial public offer tentatively planned for 2017 or 2018. The IPO will give an exit to its investors such as Tiger Global, Times Internet, Nexus Venture Partners and Multiples Alternate Asset Management who have so far invested $127 million in the company.
Delhivery’s annual revenues are somewhere between Rs 1,100 crore and Rs 1,300 crore, according to company insiders. “We’re in investment mode — last year, we spent $25 million on operations — but are really close to profitability, just a month or two away,” a person from the company says, asking not to be identified as being a privately held company, Delhivery does not disclose revenues.
For the company’s data warriors, the mission is two-fold: help the company manage the growth in scale and complexity while halving delivery costs. Its data bet will critically help Delhivery ramp up operations from the current six million packages a month (approximately one million every five days) to its aim of delivering over a million shipments a day in two years’ time. (Ekart, billed India’s largest logistics and supply chain company, delivers nearly 10 million shipments a month to over 3,800 pincodes).
The mission is to sift through massive chunks of ecommerce data from its experience of delivering over 100 million packages till date, combine them with demographic information about locations, online buyers and seasonal inputs, and help Delhivery halve the cost of delivering a package — to less than a dollar (~Rs 66) from around $2 today.
In the business of ecommerce, last-mile delivery is the most important frontier. Delhivery believes it is already the most efficient among its peers. While it costs Flipkart’s eKart Rs 190 to deliver each package, FedEx and Amazon spend Rs 250 and around Rs 300 respectively, according to executives at Delhivery (FactorDaily couldn’t independently verify these claims).
“We go after every penny, ruthlessly using data insights, and that’s why we are at least 40-50% cheaper,” says Bhattacharya. (Disclosure: Bhattacharya contributes to FactorDaily occasionally; read ‘The leaky bucket of Indian food delivery startups’ that he and his team wrote in June).
To be sure, Delhivery’s costs are lower also because it’s a third party service provider and it does not really own an ecommerce marketplace where it has to spend extra on serving the customers.
There are several components to cost of delivery, and data insights are helping the company bring them down, one by one. These costs include staff salaries, transportation (air, road), physical and technology infrastructure (warehouses to software), apart from several non-operating expenses such as city managers manning the workflow and so on.
Using software and data insights, Delhivery is beginning to optimise these costs and even getting to a situation where it can predict demand in the near future.
The software determines the sequence of loading, unloading shipments on one hand, and also allocates deliveries to individual agents based on past experiences.
“I want to be in a state where human interventions are almost negligible,” says Bhattacharya.
Delhivery has already started making sense of “returned goods” data for instance. “If you are returning ecommerce purchases too often, then maybe we should not be rushing to deliver you the packages,” says Bhattacharya. By mapping the quality of consumers and how valuable they are, Delhivery can prioritise speed and experience.
Man versus machine
Over the past few months, Kumar, Delhivery’s fakes-spotter at the Delhi airport, has been sifting through Google image search results to study iPhone hardware designs. He lists a dozen attributes based on which he can tell if a phone is fake or not. These include the colour of the battery, vibration engine when looked through X-ray machines, and several nuances related to the way various components and accessories are placed in a box.
“The counterfeiters are getting smarter and smarter, and it will be impossible to detect them purely using software and X-ray,” says Kumar. Even Alibaba founder Jack Ma admitted in June that Chinese-made counterfeits are now superior to the real ones by using exact raw materials and parts coming from the same factories.
Indeed, sometimes fake products manage to pass through all the checkboxes. Which is why the man-versus-machine debate starts looking like the Holy Grail for Delhivery’s artificial intelligence (AI) pursuits.
Delhivery’s data scientist Ali, a former CERN researcher who hunted for the God Particle, says algorithms find it tough to keep pace with Kumar’s ability to pick newer counterfeit products. Nonetheless, he and his team have built a software solution that runs image recognition checks and analyses each shipment using AI to spot fake products.
“The algorithm is based on Surender (Kumar); we are trying to mimic his expertise and his brain,” says Ali. It’s easier said than done, especially because intelligent software will always find it tough to keep pace with new masking techniques used by counterfeiters.
The data team has insights that make sure that each of the ecommerce packages Delhivery handles reaches the right destination across 70 million unique addresses from Jaipur to Gangtok, Srinagar to Rameshwaram.
Last year, Delhivery shipped 55 million ecommerce packages across 400 towns in the country. This year, it expects to deliver over 70 million of them.
Algorithms, based on insights gathered from such a massive chunk of data, are becoming the life blood at Delhivery. Lines of codes advise the most efficient route for each packet and even offer real-time analysis of weather or any other unforeseen disturbances such as a riot.
“Analytics often becomes an ivory tower in large organisations,” says Bhattacharya. “By ensuring we pick the problems on the ground and expose ourselves to real world issues, we are doing better.” His team has grown to 40 today from the six he helmed upon joining Delhivery in August last year.
Meet Delhivery’s data queen
What will India buy this festive season — running from Diwali (end of October this year) to Christmas?
Depending who you ask, it’s a question worth anywhere between $3 billion and $5 billion for India’s ecommerce industry. That is how big festival sales in India are. That number will only grow as more buying shifts online. Securities firm UBS estimates the ecommerce market in India will be between $48 billion and $60 billion by 2020.
For Snigdha Gupta, a data scientist at Delhivery and the mother of a five-year old, the ‘What will India buy this festive season’ question is almost like solving the Da Vinci Code. For a week now, Gupta has been clocking early-in and late-out hours at Delhivery’s Gurugram offices. A few days ago, she left her daughter with her parents in Jaipur to be able to crack the code.
Day and night, Gupta is sifting through hundreds of thousands of data sets from 418 Indian cities and towns, diving deep into different shopping categories, to figure out what exactly they will buy.
“Thinking what people will buy in Jhunjhunu in Rajasthan keeps me awake,” she says.
The complexities are many and unpredictable. “I am dealing with a three-headed beast: dynamics of the location, consumer buying behaviour, and understanding which category will get the lion’s share of a household’s wallet,” says Gupta.
She looks at the deadline mentioned on a whiteboard right across the hall, where all projects are mentioned, and points to the August 15 deadline for answering the big question.
“It’s actually a three to six months project if you ask me, but we have to answer this fast to be able to apply the learnings when it really matters,” says Bhattacharya.
So what will Delhivery do after finding answers to what India will buy this season?
Gupta’s answer, says Bhattacharya, will help Delhivery prepare its big clients (including Flipkart, Amazon, and Snapdeal), its army of ecommerce delivery agents, and its 17 warehouses to meet the festive demand.
“For example, if we find that NCR area will buy more white goods such as refrigerators, then we need to make provisions for handling that capacity, e.g., extra spaces for storage, bigger trucks for delivery, longer delivery times for carrying a refrigerator from our truck to an apartment and more people in the truck to carry a refrigerator,” he says.
Another way to apply the data insights will be to cope with any demand surge.
“Within the Diwali time period, the volumes will surge in a period of two weeks. During that period, there will be queues likely in our sortation facilities and hubs, which could eventually cause delays in delivery to customers. We’d need to model these queues mathematically against the sorting capacity to ensure that the delays are minimized,” says Bhattacharya.
Believing in data
Much of the current push on deriving insights from data comes from Delhivery’s success in this direction in the past. Advising field executives on the most efficient routes for delivery is a prime example. An algorithm designed for that has Delhivery executives now delivering 35 packages each a day, up from about 20 a few months ago.
This may sound simple but it’s a mammoth and complex task in a country of nearly 40,000 pincodes. “Almost 30-40% of pincodes are wrong or missing,” says Bhattacharya.
Using software algorithms, Delhivery has been able to fix the pincode problem, and it delivers over 95% of its packages in the first attempt. The company has mapped multiple landmarks within same localities across over 400 Indian towns.
The software also helps with advice on sequencing deliveries.
“The order in which the shipments need to go in a given locality can make or break service level agreements,” says Bhattacharya.
And the order of shipments is not just based on route mapping, but also on size and weight of the packages, all of which is factored in by the software.
Over the past few months, Delhivery has been able to accurately map addresses in top metros to less than half a kilometre.
Delhivery has around 6,000 delivery agents, and they all go through a few days training on etiquette, how to use their devices, etc. In an industry where attrition levels for delivery agents are in high double digits, Delhivery uses data insights to make their incentives and promotions transparent.
“We do not have to go by how many packages were delivered all the time; sometimes the last mile could be tough because of localities and even weather conditions,” says Bhattacharya. He declines to reveal the exact attrition rates, but says it’s down thanks to the use of data-linked incentive schemes.
“In India, the last mile costs account for nearly a third of total costs. In other developed markets with better infrastructure, such costs are anywhere between 8-10% of the total cost of delivering a package,” says Bhattacharya.
The last mile delivery costs are higher in India due to multiple reasons. Unlike in the U.S., where addresses are short and mapped out well, in India, there are several variables attached to most of the addresses.
“Many people mention nearby landmarks to help delivery guys locate the address. In small towns, sometimes landmarks are the only recognisable feature in an address,” says Bhattacharya.
Moreover, in markets such as the US, a FedEx agent will leave the package at the doorstep if the addressee is unavailable. In India, the delivery agents call before dropping by and, even then, sometimes the recipients are unavailable.
For Delhivery to achieve its daily target of one million shipments in two years time, technology and data science are going to be crucial. So far, the plans seem to be working. And perhaps that’s the reason why it has caught the attention of Alibaba’s Jack Ma who is attempting to build its ecommerce operations in India.
For Alibaba, a strong logistics arm in India will help complete its “iron triangle” strategy, as we wrote in June. But executives at Delhivery deny any such talks with Alibaba.
Apart from the legacy courier companies such as BlueDart and FedEx, Delhivery has to compete with Flipkart’s eKart, which is not just bigger in size (about 40% marketshare) but also counts the country’s biggest ecommerce company as its top customer.
To Delhivery’s advantage, rival ecommerce companies may shy away from giving away too much of work to eKart because it’s backed by Flipkart.
The most important takeaway from Delhivery’s use of data science is perhaps the fact that despite being dwarfed in size, India’s ecommerce market offers enormous, complex challenges to solve for the brightest technology professionals in the world.
“If you’re looking for the hardest problems to solve, India is the place,” says Bhattacharya.
Ali, the former CERN scientist, cannot agree more.
“For an algorithm to decipher different ways in which addresses are marked, is in itself a mammoth challenge,” he says.
Disclosure: FactorDaily is owned by SourceCode Media, which counts Accel Partners, Blume Ventures and Vijay Shekhar Sharma among its investors. Accel Partners is an early investor in Flipkart. Vijay Shekhar Sharma is the founder of Paytm. None of FactorDaily’s investors have any influence on its reporting about India’s technology and startup ecosystem.