Supercharging Rakuten’s merchants with AI: CEO Mikitani

“When I started Rakuten over 25 years ago, artificial intelligence was the realm of The Matrix. Today, that reality is closer than ever.”

At Rakuten’s annual New Year Conference in Tokyo, Chairman and CEO Mickey Mikitani delivered this message to over 50,000 Rakuten Ichiba merchants around Japan. One major theme of the event was No AI, No Future – a call to action for all businesses, large and small, to embrace the potential of the newest AI tools.

“All of our merchants need to be leveraging AI to its fullest potential,” Mikitani urged his audience. “I want to make this a win-win-win situation. We’re running this race together.”

Simpler synapses for scalable AI

The conference came on the heels of fresh commotion on the global AI scene: Chinese company DeepSeek had succeeded in training an open-source AI with performance rivaling that of the tech giants’ – at a fraction of the cost. As AI companies struggle with cost-efficient scaling, Mikitani compared the technology with the human brain.

“The brain is said to have 86 billion neurons. Connecting those neurons are synapses – almost like circuits,” he explained. “It’s said there are around 100 trillion of these synapses.”

Within this complex structure is the foundation of human emotions, ideas and creativity. But when it comes to synapses, more complexity isn’t necessarily better. Mikitani highlighted a fascinating discovery from research on Albert Einstein’s preserved brain.

“Einstein’s brain wasn’t just big. An analysis of his synapses revealed that they were a lot simpler than the average person. Instead of a complex web of synapses, he had simple connections. This means not needing to do wasteful calculations can lead to all sorts of new things and creativity.”

This insight has profound implications for the scalability of neural networks. Should models be trained on as much data as possible, forcing them to reference it all for every inference? Or should they be finely tuned to specific tasks?

“Large AI models have so much information that connecting it all to make inferences becomes not only extremely expensive, but less accurate,” Mikitani reasoned. “The key lies in how much effective information you can connect – that’s why we needed to do it ourselves.”

Embracing AI from within

Rakuten has established itself as a significant player in the open-source AI community, last year releasing Rakuten AI 7B, which topped charts in Japanese language performance on release.

“Compared to OpenAI’s ChatGPT, Rakuten AI 7B is a smaller, more focused AI,” Mikitani noted. “It’s efficient, and more specialized. Because we developed it ourselves, we can tune it specifically towards shopping.”

Rakuten’s AI engineers have since launched numerous open AI models, including Rakuten AI 2.0, a Mixture of Experts model optimized for the Japanese language, and Rakuten AI 2.0 mini, a small language model designed to be run locally with high efficiency on edge devices.

This research, combined with the use of various third-party AI models, has formed the bedrock for the Rakuten Group’s ambitious Triple 20 goals to leverage AI to boost marketing, operational and client efficiency by 20%.

“There are already 30,000 Rakuten employees using Rakuten AI for Rakutenians – 8,000 of them every day,” Mikitani revealed. “90% of our EC consultants are leveraging it… One in four have built their own AI tools with simple programming. I think this will grow to four in four within the year.”

Rakuten’s internal embrace of AI tools has provided powerful insights into potential business use cases. These insights are set to reach a new audience with the recent launch of Rakuten AI for Business – a generative AI service geared to empower Rakuten’s clients with powerful and flexible tools.

Supercharging search on Rakuten Ichiba

“I stood here last year and said to you: AI, AI, AI,” Mikitani told the conference, before going on to highlight the myriad ways AI is being leveraged around Rakuten Ichiba: “First of all, semantic search.”

Rolled out on Rakuten Ichiba and other services in 2024, this AI-powered approach to search allows users to find products based on the meaning of their queries rather than the exact terms.

“Previously, we had used word-matching, so if the words didn’t match, you would get zero search results. With semantic search, we’ve reduced such cases by nearly 98.5%, improving GMS by a significant 5.3%.”

The search experience on Rakuten Ichiba is also evolving to tailor results to specific users, regions, seasons and times.

“In the future, we could potentially separate search results shown in Okinawa versus Hokkaido. Or we could show search results that draw from a user’s purchase history. This is more than just marketing – it’s the shopping experience.”

Getting AI into merchants’ hands

AI is also being used to drive more appropriate product recommendations on Rakuten Ichiba, improving efficacy by around 60%. Promotional ads on the platform are also benefiting from better cost-efficiency.

On the backend, too, AI tools are helping merchants streamline their operations: “Currently, out of more than 50,000 merchants, over 30,000 of you are using RMS AI Assistant (Beta).”

This comprehensive suite of AI-powered e-commerce tools enables merchants to write product descriptions, automatically find product codes, support answering customer inquiries, or check and edit messages. Rakuten Ichiba’s merchant analytics system has also been supercharged with AI, giving insightful reports with just a few clicks.

Some merchants are leveraging Rakuten’s AI tools for powerful insights about which products to highlight during major sales, while others are tapping into shop analytics to have AI automatically choose promotional materials and provide feedback on ad performance.

“Which product stood out the most? Which product was clicked?” Mikitani posed. “You could even tell it to take the best-selling products and make them into a banner. Rather than just have a human make it, you can have AI choose the products and decide how to display them for greater efficiency. This is what we’re making, and ad performance has already improved by 50%.”

Entering the agent age

“It started around last year that large language models – all-purpose generative AI – really hit the scene. Next, we’ll see AI agents,” Mikitani predicted. “Instead of just recommending and researching… these agents could reach the point where they handle real transactions. A secretary AI, for example, might make appointments for you, or shop for you. These personal assistant-like functions could start gaining traction even this year.”

While the ultimate destination of AI tech remains uncertain, Mikitani encouraged merchants to make the most of the AI tools that are available here and now. With the support of Rakuten’s diverse suite of AI offerings, their businesses are well-placed to stay ahead in the competitive e-commerce market – both now, and in the years to come.