Flipkart said that its data scientists had developed an innovation that uses artificial intelligence (AI) to make sense of complex Indian addresses as well as detect and prevent address frauds. The home-grown online retail giant said the innovation correctly classifies and identifies Indian addresses and resolves inconsistencies with a 98% accuracy rate.
“There are people which we call as resellers, they exploit the online discounts that we provide and sell (products) offline for profit,” said T. Ravindra Babu, principal data scientist at Flipkart, in an interview on the sidelines of the company’s flagship event ‘slash n’ where it conducted ‘AI for India’ conference.
“They are like any other people and very difficult to identify. We have built machine learning models to identify them,” said Dr. Babu, who had worked as a scientist at Indian Space Research Organisation (ISRO) for more than two decades.
Flipkart, which has a registered customer base of more than 100 million, said “randomly typed alphanumeric characters as customer addresses” is one peculiar fraud. In order to avail price discounts or with possible fraudulent intent, there are instances where orders are placed with such addresses. These are referred as “monkey-typed addresses,” said Dr. Babu. Early identification of such addresses is necessary to reduce operations cost. Flipkart said when such addresses remain undetected, it would lead to activation of a number of stages in the operations chain. This includes order management system to supply chain leading to ship the corresponding packed item to the delivery hub for dispatch. This results in avoidable misutilisation of resources, according to the company.
Last-mile logistics
Indian postal addresses, which are intrinsically complicated, also pose a huge challenge for last-mile logistics. The incorrect addresses not only lead to delayed or failed delivery of shipments, but also severely affect the revenue of e-commerce companies as well as customer satisfaction. Dr. Babu said the AI-based solution developed by the firm not only rightly identifies such addresses, but is also able to save at least three hours of time per delivery hub.
The company, which deals with millions of addresses had found a huge number of spelling variations after analysing them. “For example, ‘apartments’ [can be] spelt in 700 different ways out of 30,000 addresses,” said Dr. Babu, who, before joining Flipkart in 2014, had also worked as a principal researcher at software-outsourcing giant Infosys. Typically an address would contain anywhere between eight to 12 words to identify it. “But there are addresses which [have as many as] 200 words,” he explained.
To strengthen location intelligence capabilities across its logistics network, Flipkart had also taken a minority stake in MapmyIndia in December 2015. It provides digital map data, Global Positioning System (GPS) tracking and geographic information system (GIS) solutions.
There have also been instances where e-commerce firms in the country have faced flak from consumers for delivering shipments which contained bricks, soap bars and pieces of wood instead of products such as iPhones, laptops, cameras and watches purchased online. In such instances, the online retailers have to compensate the vendor as well as the customer, as it is difficult to prove the case. An alumnus of IISc, Dr. Babu said that his team was also working on building machine learning models to capture signals which would determine whether there was any such issue with the product delivered.
