Chatbots harness the power of AI to improve customer experience


The very basis of artificial intelligence is machine learning. And as machines learn, AI-enhanced chatbots continue to grow in what they can do — and how retailers are using them to perform.

Fung Global Research has identified three main applications of AI in retail: personalization, such as recommendations; customer service; and data analytics-enabled inventory management. Fung says AI is not some futuristic technology — it is already widely used, though “the technology tends to have a low profile.”

That may be changing as customers become more comfortable communicating with a robot thanks to Siri and Alexa. The sheer volume of data that AI can produce makes it ideal for chatbots, though it goes beyond questions about ship dates or return policies.

The bot makes suggestions for style and fit, while the ProFlowers chatbot helps customers find the right floral arrangement for the occasion.

“What we are seeing and hearing from the partners that we’re working with — small, medium-sized and large businesses — is that there is a desire to offer a more personalized experience and conversational experience,” says Karen Ouk, chief business officer for, which builds AI-powered virtual shopping assistants for companies including Levi Strauss, ProFlowers and Louis Vuitton.

“We are seeing this particularly with the Millennial customer and in trends that we’re seeing the Far East, where AI shopping has been happening for a long time.” she says.

Meet the stylebot

Jeans aren’t the easiest product to shop for, but they are one where a chatbot can help. Mavi Jeans shoppers can filter and search by styles they are seeking, says Ha-Kyung Choi, Mavi’s director of ecommerce. “Depending on whether someone is looking for skinny jeans or a more relaxed fit, distressed or clean, the stylebot has helped us customize the shopper’s journey to find exactly what they’re looking for,” she says. “Because we’re able to seamlessly leverage product attributes — including color, size, fabric composition — into the experience, the stylebot helps our customer navigate our products to their exact specifications.”







Mavi’s stylebot is currently available on Facebook Messenger and is most helpful with new customers since existing customers are already familiar with the product and fit, Choi says. “We’re really seeing the stylebot as a way to help new customers,” Choi says. “A feature of the bot that we see great potential with is visual search, which enables shoppers to snap a picture or grab a style inspiration off the internet and upload it as part of the conversation. In response, the bot combs through our catalog and will make suggestions based on the image, which improves discoverability of our products and gives a customized experience for the user.”

Working with to integrate the stylebot onto Facebook Messenger was “fast and simple,” Choi says. “It was really a matter of entering our Facebook credentials and giving it permissions.” has built a bot to work with Shopify Plus; Mavi is integrating with that site as well. Despite the seamless integration with those sites, Mavi continues to take leaps and test new advances.

“We are all about testing as a company,” Choi says. “Buying denim online can be complicated and frustrating. Because denim buying is highly personal, we constantly challenge ourselves to test new solutions that eliminate friction for our customers, especially in product selection. A solution like this presented a lot of possibilities.”

Mavi continues to test other AI and machine learning solutions to improve the customer journey and conversions. “AI is in its early days, but we see huge potential,” Choi says. “We’ve seen less success replacing customer service functions with chatbots, but more as providing an interactive and fun way for shoppers — especially younger users — to interact and learn about our products in a very hands-on way.”

Each chatbot is built according to the brand’s unique proposition: The bot makes suggestions for style and fit, while the ProFlowers chatbot helps customers find the right floral arrangement for the occasion. The service is not just for large corporations; the company also works with smaller retailers, offering a turnkey chatbot solution for companies powered by Shopify.

Chatbots are equipped to offer unprecedented levels of personalization, which has become critical for companies looking to meet the demands of today’s consumer, says Eitan Sharon,’s founder and CEO. “Customers are demanding personalization and visual analysis — they want companies to know what they want, before they ask for it.”

Continued growth in customer service

As AI-enabled bots learn, even what they’re doing with customer service-related functions has grown, says Luke Starbuck, vice president of marketing for customer care automation platform Linc Global.

“We’ve had quite a few retailers looking at what’s outside the boundaries of customer care and customer service,” Starbuck says, such as using voice platforms, answering common questions and helping with returns.
Brands also are exploring creating voice experiences in ways that are an “opportunity for that brand to be more part of their daily life,” Starbuck says. “How could you offer content or expertise that goes far beyond that expertise? In Messenger, you start to use chat in different ways to send a coupon out to their customers, only if the customer has had a positive experience, that they can share with friends and family. It can be easily shared just by long-pressing on that image. That will allow customers to leverage some of the nuances about the platform.”

There are challenges — mostly related to data, Starbuck says. “The ROI can be as smart as the data it has access to. A lot of those data points that you want AI to use are housed in different systems. If the AI doesn’t have access to the data it needs, you’re lost in creating what you have in mind.”

That can be particularly challenging in an ecommerce world, more so in omnichannel, Starbuck says. “It’s quite a fractured landscape, as projects have been done independent of each other. We recognize that retail can’t wait to rectify all those back-end challenges or they’ll be waiting forever. There’s a lot of real-time decision making involved there; you just can’t do that unless you have all that data on hand.”

Building a collection

“Every partner we work with has their own experience and strategy regarding how they welcome new shoppers, which we try to replicate in each chatbot,” Sharon says. “Just as brand employees are trained on how to treat customers when they enter a bricks-and-mortar-store, we train each chatbot to communicate with users in a very natural way that is in line with our customers’ brand.”

It is not an area without risk. In addition to the challenges with integrating disparate data — key for helping the bot learn — there continue to be challenges with natural language understanding. Starbuck says that is most common with off-the-shelf solutions.

“They aim to be a general platform that could be applicable for all kinds of uses. When you try to do something like natural language understanding product names, that’s not something they can handle. If you ask about order status of socks or sweaters, perhaps your product names don’t include that specific word. The AI can’t match up the fact is that this is what a human would call a sweater. There’s no simple fix to that.”

There also is no simple fix to getting customers to use the chatbot. “The challenge there is, really, how do you best explode that functionality and channel to a customer?” Starbuck says. “That’s something that we coach our brands on. The ideal scenario is to make that available on the order confirmation page, in shipping confirmation and order confirmation emails.”

Order confirmation can be a powerful draw, Starbuck says. “We see more than 10 percent of people opt in to get order status. We think being part of that purchase journey immediately after they’ve checked out is a great position to be in.”

Speed continues to be important, as with any aspect of web and mobile. “We’ve learned that milliseconds matter,” Sharon says. “A hundredth of a millisecond matters. Speed matters. When we go to build things, we make it a requirement that the system will be fast. Visual interactions are much more demanding. If a customer asks the bot to find something in an image, the challenge of speed is vastly bigger.”

Where AI is headed

AI-enabled bots have come a long way in a few short months, and brands that have gotten a head start can extend the lead — especially given that the intelligence grows as the machine learns. “If you are a brand that has established online in both the voice and chat platforms, you are now positioned to think beyond what you’re already able to do and think about what you want to do next,” Starbuck says.

Chatbot developers continue to push what the tool can do. “One of the advances I believe is coming is AI-learning technology that is not necessarily visual per se, but allows you to engage wherever you are,” Sharon says. “Imagine where you are not actively shopping, but in a conversation. The [chatbot]could tap on my shoulder and say, ‘Would you like to know more about that?’ in a way that is relevant. If you agree, you can engage. We are seeing a lot of interest in that.”

The level of personalization available to shoppers will also become more advanced, especially because AI can gather information from a variety of outside sources. For example, AI could analyze photos posted in social media feeds to make smart suggestions on what clothes will look good based on body shape and size.
The devil is in the details — and for AI to reach its full potential, it comes down to data.

“When there is data in the system, AI can learn tremendously,” Sharon says. “In the absence of data, there is only so much we can do. That is subject to getting enough real data. In a year, more customers will have used the product and we will acquire much more mileage in seeing the visuals. If you don’t see enough people wearing these items, the machines will be restricted by the level of exposure.”

Sandy Smith grew up working in her family’s grocery store, where the only handheld was a pricemarker with labels.


Comments are closed.