How Uniqlo developed its ‘digital concierge’ voice service

Last month in Japan, Uniqlo launched Uniqlo IQ, a predictive service built on Google’s Dialogflow platform and designed to make it easier for customers to find things they want to buy. The developers of Uniqlo IQ acknowledge the limitations of AI but see the service as the start of a shift from reactive to proactive customer service that has implications beyond the retail sector.

A first for the fast-fashion giant, the initiative was developed through a collaboration between Party and Inamoto & Co. Campaign asked the people behind the project to explain how they did it.

Rei Inamoto, founder of Inamoto & Co, said his company first presented the idea of an “AI-powered customer service engagement platform” two years ago. He said the main aim was to help Uniqlo manage inventory more efficiently, which is deceptively difficult in the retail business.

“It’s a question of how you manage expectations and predict what kinds of products will be popular and sell more,” he said.

IQ sits within Uniqlo’s mobile application and is also integrated into Google Assistant. It is connected to real-time store inventory data and uses text and voice interaction to help would-be customers find products to buy via the app or in the outlets closest to them. It is also designed to be used during the physical shopping process, and recommends new products based on individual searches, hourly product rankings, occasions and personal specifications such as daily horoscopes.

Although the service only operates in Japan at present, the developers started out by testing conversational scenarios in English. “AI is more advanced in English than Japanese so the concept had to be proven in English first,” Inamoto said. The partners tested it for about nine months, then switched to a Japanese version, upgrading it on a weekly basis up until mid-July.

Tatsuhiko Akutsu, information architect at Party, said the biggest challenge after creating a prototype incorporating all imaginable scenarios was developing the ‘voice’ to sound natural and friendly. This was especially difficult given the many nuances in the Japanese language, said Norikuni Takamiya, Party’s chief strategist. It required extensive testing and reworking, meaning that IQ “took a lot longer than expected to come into reality”.

In a parallel field, the usefulness of smart speakers as a sales channel is still unproven, and the low volume of people currently using them to shop suggests they may still be better suited to promotional activities. But Inamoto said Uniqlo made a point of treating the IQ project “not as a marketing campaign or a toy to create buzz, but as a tool that can provide service to customers”.

This attitude meant the company “dedicated enough resources to keep the lights on for it to grow”. In its current form, IQ is not seen as a finished product, but the start of something that will evolve.

“We’ve had a meeting once a week for the past two years and for the client not to say ‘this is a marketing thing’ and ‘this is over’ [is commendable]. They believed in it and committed to keep the team that worked on it. Now they’ve opened it, we really have to have the commitment to keep making this better and provide a better service over time.”

Inamoto said the service would become available in English at a later time. As for Chinese, he said there is no defined roadmap, but launching would be “a no-brainer” given the importance of the market for Uniqlo’s business.

AIs: Akin to toddlers

Inamoto acknowledged that in general terms, users and companies have inflated expectations of what so-called AI systems can do.

“It’s not magic yet,” he said. “They try to make it do everything. It barely has the capabilities of a 2-year-old when it comes to general AI. Specific AI can surpass human beings, but in general people try to do too much. It’s severely limited at this point.”

He advised that it’s better to build functionality slowly. “Some of the things you would expect to be useful to users may not be as useful as you think,” he said, adding that this holds a lesson for more straightforward marketing too.

“You can put as many promotions as you like on a home page, but people tend to go specifically to the product they’re looking for,” he said. “What’s on the home page matters less than marketers think.”

Another thing that surprised him was that people sometimes resent being told what might look good on them. That’s not to say they don’t find recommendations useful, but they need to be made subtly, with a deep understanding of the individual’s tastes. Considering how difficult that is, and how much data is required, IQ is certain to make mistakes, but the developers hope that its usefulness will prevail.

Inamoto gave Spotify as an example of a company doing this well. “They choose words carefully, don’t use recommendations too much but use the word ‘discovery’. It’s not being pushy. You discover as opposed to, ‘we know what you like’. The same principal can be applied to fashion. It’s our job to surface things they might like and let users discover what they really like. IQ opens that possibility.”

Takamiya said he thinks IQ “has the potential to become the foundation of Uniqlo’s business” as a means of data collection and customer support that enables the company to become more proactive.

“We’re seeing 1% of what it could be,” said Inamoto. “In general, customer service up till now has been a very reactive thing. Outside retail, customer service is just painful. It needs to evolve to being proactive so it anticipates needs and problems before they become problems.”