The company runs a large online auction platform and shopping website that connects buyers and sellers in over 190 countries. Their website has millions of sellers with 1.9 billion global listings and over 132 million active buyers.
THE CHALLENGE
Each seller has a virtual bank account where all payments from sales on the ecommerce platform are maintained, before they can be withdrawn. All sellers that are based out of the United States of America, have to pay federal taxes on their transactions and file their returns with the Internal Revenue Service (IRS). The company is responsible for calculating, collecting and submitting the taxes that are due to the IRS for each seller.
Based on the total amount available in their virtual bank account and how much had to be paid in taxes to the IRS, these sellers were separated into the different categories. For the categories in which the company had to pay the pending amount to the IRS, they then recovered it from future transactions of the corresponding sellers. The company found that there was a certain segment of these sellers that consistently indulged in fraudulent transactions and activities. It would be challenging to retrieve the unsettled IRS dues that were paid on their behalf. The tracking of these sellers was being done manually for each quarter.
The company needed an analytics partner to automate the identification and tracking of all sellers from the 2 categories that exhibited patterns of fraudulent behaviour.
THE SOLUTION
The team of business analysts from Prescience Decision Solutions worked with the company’s team that interfaced directly with their sellers, to understand the existing manual processes. There were different metrics based on which the company could determine if they were genuine sellers or needed to be reviewed for fraudulent behaviour. Our team explored the automation of various combinations of metrics, to predict the behaviour of all the US based sellers on the marketplace. The output of this seller behaviour prediction solution was compared against data from the preceding quarter, to test for accuracy.
For the selected time period, the solution was able to determine with 99% accuracy all the genuine sellers that were operating on the ecommerce platform. On detailed analysis, it was found that certain sellers that had been incorrectly tagged as for fraudulent behaviour in the manual process were actually genuine sellers. Since classifying fraudulent behaviour required a complex combination of metrics with some manual oversight, the automated solution was able to determine with over 80% accuracy all the sellers who were involved in suspected dishonest activities. The company’s team which interfaced with sellers used these findings to identify sellers that were genuine and sellers that had to be further reviewed.
Given the success percentage of establishing the genuine sellers on the ecommerce platform, the team incorporated this seller behaviour prediction solution into a self-serve tool for business users. The technology used for this engagement was,
- Microsoft SQL
THE IMPACT
With the new seller behaviour prediction solution, business users could separately or jointly view the details of the sellers from the different categories, based on which they could analyse whether or not they would recover the pending dues that were paid to the IRS. The automation of the solution eliminated the time-consuming and costly manual analysis for the seller interface team. It is also helping the company evolve its seller management strategy to efficiently manage the pending annual payments to the IRS.
Learn from the Success Stories of Business Transformations Fueled by Data, AI and Machine Learning
More to Explore
Prescience Team