By Adarsh Jain, Rahul Agarwalla and Vinish Kathuria
The Indian e-commerce ecosystem is quite unique. Unlike mature markets in US /Europe the most popular payment method is an Indian ‘Jugaad’ — cash-on-delivery (COD). COD suits the Indian consumer and thus makes up 70% of all Indian e-commerce. However, it comes with a unique Problem — Return to Origin (RTO).
RTO is when orders cannot be delivered and have to be shipped back to the warehouse. This puts a significant cost burden on e-commerce firms as they lose money on:
- Forward & Reverse Logistics.
- Blocked Inventory (Items stuck in transit)
- Physical Quality check and re-packaging of returned items
- Increased probability of damage to fragile items.
- Operations cost in processing this order.
In case of COD orders, RTO can be as high as 40%. When one in three orders has the potential to damage your bottom line, instead of adding value to it, the situation is alarming! To find a solution we analysed RTO data and some common patterns emerged:
- Order without Intent (for Fun)
- tomer Error (Intent is there but incomplete address etc.)
- Orders from transitory addresses (hotels, friends place etc.)
- Price sensitive intent (drop in price — reorder)
- Impulse buy but without paying. (there is no downside to refusing delivery).
- Intent to Fraud (Habitual fraudsters)
Companies have little choice and fewer tools to prevent RTO — they just take it as a ‘cost of doing business’. Most firms resort to blunt static rules like:
- Block all International credit cards
- Do not deliver to certain pin codes and cities
- Cap the order size
These macro-level rules do more harm than good as many genuine orders are lost and customer relationships damaged. Solving the RTO problem by manually scanning every order does not work either due to the sheer scale of the problem and evolving nature of fraud techniques. With the Indian e-commerce market becoming hyper competitive, firms need better solutions as they cannot afford to lose customers and orders. Machine Learning technology offers an attractive solution as it addresses all the challenges in preventing fraud — scale, complexity and changing patterns.
Catching digital frauds requires us to first gather the ‘Forensic Evidence’. Every user interaction leaves behind a subtle digital forensic trail like proxy IP, device ID, email address, time to order etc. Machine learning models combine hundreds of such innocuous parameters, which are seemingly unrelated, to identify the patterns that indicate fraud.
Machine learning and natural language processing are used to differentiate between real and fake address. This is just the beginning. Transaction and User data is enriched by adding context to it. For example by adding the price of the user’s phone device or categorizing an address as five stars or one star we turn meaningless data (Phone model) into actionable information which increases the accuracy of the Red or Green flag that the machine learning models generate for every transaction.
As we augment and enrich the data we can create risk profiles for each user. We can understand common email and physical address patterns as well as the digital fingerprint of the device used. Even when fraudsters reset their phone, to wipe the traces of their previous frauds, the digital fingerprint enables the algorithms to identify the fraudster. By leveraging the augmented data and machine learning algorithms, we saw a 81% reduction in RTO within 3 months at a large e-commerce firm.
Fraudsters are habitual creatures. They leave similar footprints on multiple sites. Network effects can be harnessed by pooling in anonymized data to predict and prevent fraudulent behaviour. This de-incentivises and penalises fraudulent behaviour across the ecosystem.
By combining network effects with augmented data and machine learning, RTO can be reduced by over 80%. RTO reduction has a cascading effect as it make the entire ecosystem more robust, accountable and profitable. It will help the Indian e-commerce ecosystem to mature and become available pan-India — where no location is a “no-delivery” zone. Moreover, e-commerce firms will truly know their customers so that goods are delivered to a person not merely to an address. Most importantly, RTO will no longer be just a “cost of doing business”.