Name of Founder (s): Adarsh Jain and Shashank Agarwal
Industry: Artificial Intelligence
Investors Details & Amount raised: Undisclosed
November 2017: Flipkart cheated off of Rs 17 lakh by a gang
October 2017: Amazon conned to the tune of Rs 50 lakh in New Delhi
May 2017: Amazon India cheated off of Rs 69.91 lakh in Bengaluru.
Dec 2016: Flipkart duped of Rs 1.05 crore, 152 expensive mobile phones in Kota
The above instances of customers’ duping ecommerce companies made the headlines, but there are thousands of cases every year that go unreported in the media. If you think that etailers are the sole victims of ecommerce fraud, you would be mistaken. “Fraud” says Adarsh Jain, the creator of a fraud-detection AI engine, ThirdWatch, “affects the entire ecosystem of ecommerce – etailers, merchants, and also customers”.
“Fraud is a big problem for ecommerce companies and these companies end up blocking complete zipcodes of states to deal with it. There was a time when one of the biggest ecommerce companies in India was not shipping anything over Rs 7000 to 20% of the population. The whole of Madhya Pradesh, Bihar, Uttar Pradesh, Jharkhand, and Uttarakhand were blocked,” adds Jain.
But, it is not like these ecommerce companies do not have systems in place to mitigate, if not, eliminate fraud. But, despite having multiple checks and balances in place, etailers continue to regularly fall prey to the machinations of fraudsters. So, then, why does Jain think his solution would work better than the methods employed by the ecommerce companies? “Because we are not building a better mousetrap, but a cat. While a mousetrap confuses any similar looking thing with a mouse, the cat knows better,” says Jain.
Most of the systems used by ecommerce companies are rule-based, which are not adaptive and therefore fail to identify new fraudulent patterns and these broad-based guidelines are not capable of differentiating between genuine customers and fraudsters. For instance, if someone returns more than three packets the etailer may block that user from placing any more orders. “What that does is that it impacts even genuine customers. Even if someone has returned the last three packets that customer may have successfully placed 50 past orders. So, when you employ general rules you end up having a high rate of false positives. More importantly, it is not adaptive which means if there are new fraud patterns, these rules would not be able to help,” explains Jain.
Time to hunt down the mice
The metaphorical cat built by ThirdWatch is an artificial intelligence-based solution which catches a possible fraudulent transaction before it reaches the ecommerce company’s doors. Once any ecommerce company integrates the engine developed by ThirdWatch, the company’s AI system generates a flag for every transaction – green for genuine transactions and red to signal a fraudulent concern. Those marked red goes to a review dashboard instead of being directed to shipping like the green ones.
“Our system springs into action once the order is placed and generates either of these flags in less than 200 milliseconds. Also, we do not just raise a red flag, but also provide the reason behind that flag, along with a risk score. Our AI-based engine is adaptive so even if it has not seen a particular kind of fraudulent method, it is still capable of flagging it,” says Jain.
The clients, that is, the ecommerce companies then review the flagged transactions with some of them even opting to automatically cancel a transaction above a certain risk score to reduce their dependence on manual review.
Fraud – the albatross
Fraud has been the albatross around ecommerce companies’ necks for the longest time, but the burden is becoming noticeably unbearable now because the succour in the form of funding now comes with questions of profitability. The very same profitability whose topline gets hit the hardest by fraud.
“Estimates suggest that ecommerce companies lose 5-7% of their topline to fraud. Out of all kinds of frauds perpetrated in India, RTO (return to origin) is the highest at 30%. We know of a website which used to music labels which had to shut down their online selling services because their RTO was as high as 35%,” says Jain.
There are multiple use cases of RTO fraud – with sometimes even competitors indulging in unethical practices like placing orders to make their competition lose money in forward and reverse logistics. Another kind of fraud from a buyer’s side is duplicate item fraud. “A lady defrauded Amazon for Rs 1 crore over a course of two years by returning products after replacing items with cheap replicas. If Amazon can fall victim to such frauds, you can only imagine what may be happening to smaller companies,” says Jain.
Another kind of attack comes from hackers. Jain narrates an incident: “Recently, some hackers froze the request going from an ecommerce company to their payment gateway and changed the amount of the order from Rs 10000 to Re 1. Once, the payment got through, these hackers changed the amount back to Rs 10000. This made the ecommerce company think that they received Rs 10000 but in reality they only received that one rupee. By the time the fraud was discovered, the item was already shipped and delivered”.
A library of frauds
The team behind ThirdWatch employs multiple layers and methods of fraud detection, but Jain says it is prudent to not divulge much of what goes on behind the scenes. “The challenge in revealing too much of how our algorithm works is that at some point hackers will read it and may think of exploiting the information,” explains Jain.
The company also makes sure that once a fraudster comes into their net, they do not make a fool of any other company. “We use our network effects for all those who have signed up with us. For instance, if a customer has done some fraud on one website and leverage the user’s information across our customers without telling them where the information is coming from; we raise network flags,” says Jain.
Jain insists that the startup is cognizant of data security of its clients and in no instance compromises on that while sharing information on fraudsters. “We encrypt each byte of data with a different key, so even if someone makes away with our servers, they will not be able to decrypt it,” assures Jain.
Jain claims that their clients have been able to reduce their RTO by 80% and even effect an increase of 5% in order delivered, made possible by the system differentiating between incomplete addresses provided by genuine customers and fraudsters.
Started in April 2016, the company has five clients and is also working on a trial period for two big names in the ecommerce business. “We offer a 60-day trial period because we want to prove to these companies that we are not going to be a cost centre for them, but a profit centre. If we save them Rs 100, we take Rs 15 out of that,” says Jain.
Funded by IAN, the company is also receiving interest from investors, claims Jain and is looking at turning operationally profitable by 2018.