Flipkart has come up with a novel way to enhance customer experience and moderate profane user reviews, in preparation for its biggest annual sale, Big Billion Days, in October.
The e-tailer has crowdsourced its quality function to college students, wherein they are made to answer multiple choice questions through a third-party app for which they are paid, while they have no idea that the questions are from Flipkart.
“We started with crowdsourcing the quality function two years ago and have tried to experiment with it in areas such as catalogue quality and user reviews, to raise the bar for our seller and customer experiences, respectively,” Ram Papatla, Vice-President, Product Management, Flipkart, told
Pointing out that there are two dimensions of quality, objective and subjective, he said: “Examples of objective quality assessment handled internally are: ensuring that our systems scale to the highest level of shopping traffic spikes without crashing during the festival season, and predictable listing workflow for sellers on our platform, where we derive information from supplied images and verify those images with the descriptions sent by the seller, using AI (artificial intelligence).”
For subjective quality assessment, Flipkart resorts to crowdsourcing to involve a larger community of shoppers such as college students, to provide feedback on products and select initiatives. One of the major reasons for this, is to improve its Net Promoter Score among its primary target audience — youth, said Vidhya Shankar, Executive Director, Grant Thornton India.
Cost savingsCrowdsourcing has also helped the e-tailer build a massive repository of training data for its AI models, which moderates profanities in user reviews. Good reviews/ratings are a key factor in driving sales in e-commerce.
“We have built the largest dictionary of profane words in Hindi at Flipkart because there is no way a Microsoft AI dictionary will understand and reject a profane user review without human intervention. We are able to determine the sentiment of a ‘Hinglish’ text through our Neuralnet for Sentiment tracking, by providing millions of samples of annotated text from a crowdsourced project,” said Papatla. He added that it has saved Flipkart several hours of automation costs. However, experts warn that the veracity of the information received via crowdsourcing is highly doubtful.
“Flipkart will only get authentic results if crowdsourcing is done in a controlled environment with a closed group of users that represent the e-tailer’s core target audience, either in its own testing centres or in the user location,” said Srikanth Sundararajan, General Partner, Ventureast. Intuit pioneered this very successfully in the US with its ‘Follow Me Home’ programme, where it observed customers to determine exactly how they use the company’s financial software products, he said.