How GenAI is Being Used to Reduce Ecommerce Returns

ecommerce returns feature

With generative AI adoption being so widespread across industries, one key application and business process of the technology near and dear to the hearts of retailers is finding ways to reduce the expense, hassle and poor customer experience of ecommerce returns that grow year after year.

From making website descriptions more accurate to improving product quality and delivering personalized offers and deals in real time, there are any number of ways that retailers are tapping into GenAI to reduce returns.

“Ecommerce returns remain one of the biggest challenges for retailers and one of the top reasons why many of them are not profitable,” said Sudip Mazumder, SVP and retail industry lead, North America for digital consultancy Publicis Sapient.

An uptick in returns is especially common during the holiday season. According to Salesforce data from earlier this year, there was a 63% increase in holiday returns in 2022 compared to 2021.

Here are some of the ways that Publicis Sapient sees leading retailers leveraging GenAI and machine learning tools to reduce ecommerce returns, especially as they look ahead to Q4 when volumes spike.

Browsing as a Conversation

Mazumder said AI-driven chatbots create personalized responses and recommendations in a conversational format, reducing the likelihood of a return. For instances, Asos and H&M use them to scour the catalog and provide recommendations based on a shopper’s price and style queries. At Sephora, AI chatbots serve up makeup styles based on variables like skin tone which can be tried on virtually, or help schedule store appointments with a stylist.

By 2027, Gartner predicts, chatbots will be the primary customer service channel for about 25% of organizations. And an April survey from Capterra found 67% of ChatGPT users feel understood often or always, while only 25% of shoppers feel understood by current chatbot technology.

“Once shoppers see these personalized recommendations, the chance of a return are reduced significantly,” he said. “It moves the needle quite a bit, having a conversation that gets to what they really need, leading to a desired outcome.”

Content Creation

GenAI tools can update product descriptions by automatically scanning through customer feedback and reason codes, such as unclear sizing charts, thus reducing the incidence of returns.

It can also generate new video content from existing media files, providing product views from all perspectives, including virtual try-on. Aggregated content from customer reviews can be summarized to directly address issues or questions raised by a shopper.

“Walmart for instance not only creates product recommendations but also generates blogs and social posts about products based on their descriptions, which are served up to shoppers,” Mazumder said. “Target is developing in-store signage copy that is all created by GenAI.”

Personalized recommendations are built based on search and purchase history. “The holy grail is matching what the customer is looking for without asking too many questions in order to get to the desired end outcome,” he said.

Nike does a great job of using GenAI to offer up a specific pair of shoes based on things like brand preference, previous purchases and browsing history. “It helps them choose exactly what they’re looking for, reducing returns in the process,” Mazumder said.

Netherlands-based global luxury retailer Mytheresa has seen a 15% reduction in ecommerce returns this year by using GenAI to make product recommendations based on parameters like preferences and recent reviews.

Improving Product Quality

Mazumder said Walmart is using GenAI and machine learning to proactively detect potential food safety issues in its supply chain. “The large language model is very good at predicting them,” he said.

At fast-fashion retailer Zara, AI-driven computer vision can identify product defects like a loose thread or missing buttons, catching them before an item is put into stock. Adidas is making product quality improvements by scanning customer feedback with GenAI.

“The LLMs are constantly learning and saying, here are the types of issues we’re seeing, and based on statistically significant data, this a problem,” Mazumder said. “Now everything is automated, there’s no manual intervention, creating continuous learning and a virtuous cycle.”

Using these AI tools, Amazon can quickly identify problematic suppliers and boot them off its marketplace. “There are measures in place driven by AI to help with that process,” he said. “I expect more and more marketplaces to adopt this technology going forward.”

Reimagining Inventory Optimization

GenAI and machine learning tools can optimize decision-making with automated, predictive forecasting. It can simulate scenarios and help retailers make informed choices in real time to ensure consumers consistently get what they want when they want it, improving satisfaction and reducing returns.

Stitchfix is using GenAI to reimagine its inventory management, demand planning and forecasting by analyzing customer preferences. “They’re proactively driving inventory decisions, replenishing and forecasting based on what customers are going to buy,” Mazumder said.

While some retailer CIOs and CEOs are rightly concerned about customer privacy and data security using GenAI tools, he said retailers can address that issue by limiting LLM inputs to their own data and avoiding external sources. Strict safeguards and policies need to be put in place to ensure that remains the case, but it’s early days in terms of how successful those efforts will be.

“They’re sitting on a trove of their own content from the call center and chat logs that can be fed into the LLM, and creating their own sandbox,” he said. “But companies need to ensure that the data is not available publicly, and that they have complete control over what has been entered into the system.”

Mazumder stressed that retailers needed to implement more than one of these strategies in order to have a significant impact on reducing returns. “Implementing just one or two of these will not cut it,” he said.