Delhivered by Data
Delhivery is India’s largest third-party eCommerce logistics company, carrying a third of India’s eCommerce on its backbone. Although the industry is growing at 2–3 times annually, India’s eCommerce woes are numerous. In this article, we give you a glimpse of how we mitigate these challenges using data and advanced technologies. [For a good read on India’s eCommerce, where delhivery and our data group is mentioned, read “The Great Race”in The Economist]
The data team at Delhivery is a mix of PhDs and Masters in Mathematics, Sciences and Operations Research (OR), as well as Engineers and MBAs from institutes as diverse as MIT, NASA, CERN, IIM and IITs. The team’s experience includes handling extremely large data and building Machine Learning Algorithms at Facebook, Amazon, AOL etc. and creating consumer behavior, digital marketing and OR models at GE, P&G, Coca-Cola, Microsoft, BMW etc.
Delhivery is a data goldmine: over a quarter of India’s e-Commerce passes through our network to 450+ cities with over 75 million packages annually, generating ~ 3 billion status scans.
The key areas the group focuses on are:
- Disruptive Algorithms: We have taken a few technology “bets” that enables us leapfrog the competition or make disproportionate efficiency gains. For example, our “AddFix” technology allows us to deliver 98%+ of the packages to the correct localities and sub-localities, despite poor addressing system in India.
- Predictive Models: Given we have data on 50 million shoppers and 100K+ sellers which keeps growing every day, we are able to build data-rich, robust statistical models. These models allow us to predict behaviors of marketplaces, sellers and consumers, enabling us to efficiently plan for the future demand, as well as to generate new product ideas.
- Descriptive Statistics: These post facto analysis make our operations more efficient. Example of such reports are: the demand and performance report by cities or the manpower efficiency reports at the delivery centers, and SLA metrics by cities/offerings. Our expertise with data has enabled us to drive efficiency every year since inception, making us the lowest cost provider in the industry.
The eCommerce industry grew 2.5–3 times in 2015 and will likely maintain its high growth rate in the coming years. A plethora of technologies including AI would help us resolve address ambiguity, traffic navigation and weather related issues to scale the logistics business, enhance efficiency significantly and reduce cost to enable eCommerce maintain its high growth rates in India.
Solving complex, real life problem requires deep focus on research. Some of the strategic research areas we are working on are:
- Addressing and Mapping: India’s addresses are poorly written: 30–40% addresses have incorrect postal codes, people use local abbreviations and colloquial locality names, sometimes manufacturing them as they see fit. Incorrect spellings are common and an there is an over-reliance on points-of-interest: for example: an address is said to be located “next to Domino’s Pizza”. A second generation machine-learning address disambiguation system enables us to correctly identify a locality with over 98% accuracy, significantly speeding up package deliveries. Currently, this technology provides us with better disambiguation of addresses than ANY off-the-shelf mapping technology available for India.
Our longer term goal includes building a “real”, searchable map of India comprising of POI data, weather and traffic, so as to provide “realistic” directions to people in a format familiar to them.
- Consumer and e-Commerce Models: The e-Commerce is a three-party model among Marketplace, Sellers and the Consumers. The biggest challenges the eCommerce industry faces relate to consumer behavior such as preference for cash-on-delivery services for products over pre-paying, over-reliance on discounts and cashback, different buying behavior across geography, age groups and income segments. With a strong in-house modeling team, we can build complex consumer, transactional and financial models to predict the future growths of the eCommerce market and uncover new product ideas and unmet needs.
- Network Modeling: Our network is complex: it spans across 450+ cities, with 1000+ facilities with over one million sq. ft. of fulfillment center and with over 15,000 vehicles/ delivery personnel. We offer multiple types of services for small packages, light white goods and heavy goods — to name a few. Our services are delivered at different timescales such as next day, same day, within an hour and are delivered to different types of locales such as hyper-local (<10 km), intracity, intercity and international. Simulating and optimizing this complex network with thousands of variables and constraints is an important area of research at Delhivery.
- Universal Product Catalog (UPC): Many small and medium sized businesses (SMBs) are now coming online in India, as the awareness about online as a new channel spreads. When some of these merchants list their products on the marketplaces, they tend to provide poor product descriptions and images, making categorization and discovery of their products difficult. Our UPC technology enables unique identification of products, making discovery and listing easier.
The measurable impact of these efforts has been remarkable. In the past two years the data team has:
- Increased the productivity of a biker from 25 deliveries per shift to 34 — a remarkable 36% improvement
- Reduced lost shipments by 67%
- Reduced “misroutes”, a shipment taken to a wrong address initially, by 68%
From predictive models, Operations Research to advanced Machine Learning algorithms, data@delhivery is there to enhance the experience of India’s e-Commerce consumers, speed up the delivery to their doorsteps, bringing smiles to over 50 millions families every year.
