eBay Head of UK Analytics Amit Agnihotri explains how eBay is improving customer experiences by using NLP to tap into a vast trove of previously unused data



Selling items on eBay is supposed to be convenient for both buyers and sellers – but should a dispute arise, eBay needs to know when to step in, and how to resolve it.



In the past, users would have to manually create a ticket and wait for a response. Now, eBay is using natural language processing (NLP) technology to analyze member to member (M2M) messages to predict outcomes and modify their remediation.



In this week’s episode of the Business of Data Podcast, eBay Head of UK Analytics Amit Agnihotri explains how they are doing this, and the big plans they have for the future of NLP at eBay UK.



Tapping into unstructured data



Over the past 20 years, eBay has radically improved its analytics capabilities, however, this has been primarily based on numeric data. Using advances in AI technology, eBay is now targeting the vast amount of unstructured data its platform generates – specifically, M2M messages.



“When we do analytics, we are mostly dealing with numbers. How many users are there? What is the length of the time? So, it's very numeric.”,” says Agnihotri. “What we have ignored that almost a hundred times bigger is data available on natural language.”



By tapping into this vast repository of unstructured data, Agnihotri’s team can provide analytics to enable eBay to act more effectively to resolve disputes, make inferences about customer satisfaction, and reduce the rates of customer churn.



Taking account of cultural differences



Language and communication are highly culturally specific, and as a result, any approach to NLP must consider cultural differences in communication styles before inferring meaning.



One of the challenges that I see is the culture-specific challenge,” says Agnihotri. “For example, the same sentence from a German customer and a UK customer could mean something very different.”



Understanding the nuances in how people communicate dissatisfaction is essential to preventing customer churn, a process Agnihotri compares to ‘death by a thousand cuts’.



“We call them paper cuts. The smallest paper cuts occur, and they add up,” Agnihotri says. “And then one day, [the customer] decides to search for some other platform.”



Getting started with NLP



The path to NLP involves a certain amount of trial and error. For those who want to experiment with NLP, Agnihotri recommends an iterative process focusing on step-by-step improvement.



“If anyone is trying to get into NLP and using that as a learning tool, don't expect that you will get a very quick answer,” Agnihotri advises. “Be patient on this. This is a huge tool, like 10 times, or a hundred times more data than we have on natural language compared to the numbers. It will take time.”



One strategy that Agnihotri suggests as a starting point for NLP is monitoring social media to ‘listen’ for feedback about your company.



“On Facebook, Twitter, a