The third session of Mercari’s internal talk session “Leaders Talk! AI/LLM Behind the Scenes Open Door” welcomed CXO Mayumi Narisawa (@narico) as guest speaker.

As AI reshapes the way creators work, what kind of transformation is Mercari aiming for? Could a future with 23 million uniquely optimized user experiences actually be within reach? Moderated by Gomi Hayakawa (@gomichan) from the AI/LLM Office, this passionate discussion explored what it will take to deliver the best possible UX for every user, and what the future of product creation at Mercari could look like.

This article was created and written using AI tools based on an audio recording of the session held on June 13, 2025. This English version was translated from Japanese using generative AI.

The dramatic evolution of GPT-4—Japanese support and a turning point for design

@gomichan: Thanks for joining me today. To start off, how are you using generative AI tools in your day-to-day work?

@narico: I try pretty much everything that comes out, like ChatGPT and Midjourney. Since I have a design background, I also make heavy use of Figma AI and newer tools like Figma Make.

*Recorded as of June 13, 2025.

@gomichan: I often see on Slack that whenever something new comes out, you try it the very same day.

@narico: Exactly. I just dive in right away. I want to be the first to say, “Look, this is amazing! You can do something completely new with it!” (laughs)

@gomichan: Being an early adopter really sets the tone. When leaders jump in like that, it encourages others to think, “It’s good to try things early” and “Let’s just give it a shot.” It creates such a positive atmosphere.

@narico: I originally joined Mercari as a team member, so I still use tools with that mindset. When GPT-4 got its big update at the end of March, it felt like a real turning point for design.

@gomichan: Right, image generation was all over social media then. People were making illustrations that looked like certain animations, and suddenly the models became great at handling Japanese text too. It also became much easier to generate thumbnails with text in them.

@narico: That was definitely a turning point. I’d been using them before, but I found that with more back-and-forth, they could do things I had assumed they couldn’t.

From global roadmaps to 23 million UX possibilities

@narico: As VP of UX and Product, I feel a responsibility to look at the bigger picture and think about how to upgrade Mercari. I often tell my team that we need to rethink assumptions—should we stick with a single app or split into multiple apps?
To explore this, I looked at product roadmaps of global companies that have scaled massively. Normally, this would take ages to research. But with ChatGPT, I could get structured insights quickly—almost like playing while learning.

@gomichan: That kind of wide-angle thinking seems crucial for breaking away from precedent.

@narico: Exactly. Our team members are doing amazing work focused on tomorrow’s or next week’s growth, but it’s easy to get caught up in immediate goals. Looking at global product roadmaps helps me see how loyalty programs are positioned, how they’re customized to fit business models, and what makes them unique. From there, I can come back to our own products with fresh ideas. To me, that’s creativity—having conversations with ChatGPT that spark new perspectives.

@gomichan: Because the team is handling the day-to-day, you’re able to focus on what Mercari should look like 5 or 10 years from now. AI is clearly helping support that kind of long-term vision.

@narico: That’s right. One thing I keep in mind is not letting ChatGPT give me the final answer. What matters is how we generate new ideas in the first place. I often ask for as many perspectives as possible, then sharpen the direction from there.

@gomichan: Sometimes hallucinations, where AI confidently gives wrong answers, can actually spark creativity. Having more options just makes your choices clearer.

@narico: That’s true. When I first joined Mercari, an engineer told me that, “Mercari is a company with so many ideas, but we’ve only accomplished about 5% of them.” When I asked if they were working toward 100%, he laughed and said, “No, because new ideas keep popping up, so there’s never really a finish line.” (laughs)

@gomichan: I know the feeling. (laughs)

@narico: Recently, I’ve also been talking with Marketplace CPO @unryu-in about systematically optimizing the UX of safety and trust at Mercari. There are countless improvements we could make—not because we’ve neglected things, but because what’s “best” for one user can be a “bad UX” for another. With 23 million users, there are 23 million versions of the “best UX.”

If we really want to solve things at that level, we need to rethink the structure of product creation itself. Instead of narrowing everything down to one idea and serving it to 23 million people, in theory we could deliver up to 23 million different ideas—23 million different screens—each tailored to the individual.

@gomichan: Ideas that once seemed unrealistic now feel surprisingly possible with how quickly the technology is evolving.

@narico: Exactly. For example, we recently worked on classifying what users value based on their profile text. In the past, we used keyword-based pattern matching, which often led to misclassifications. Doing it accurately meant humans had to check everything, and it was estimated to take more than three months. Instead, we used AI to analyze and categorize the text, then had humans cross-check samples. Compared to traditional pattern matching, the accuracy was clearly higher. The match rate was about 70–80%, not perfect, but many of the disagreements were cases where even human reviewers interpreted things differently. In other words, AI was handling content at the level where humans themselves weren’t aligned.

In the end, we had humans review and finalize the results, but the overall workload dropped dramatically. What would have taken months was completed in just two weeks.

@gomichan: Being able to say with confidence, “This is reliable enough to release initiatives based on real expectations,” is pretty amazing.

Rethinking UX—A company-wide transformation in how we create

@gomichan: I heard you recently launched a project to rethink UX from the ground up. Can you share what that’s about?

@narico: The mission is simple: shift Mercari’s approach from plan-heavy to experiment-driven creation. Instead of narrowing down to one idea because of resource limits, we want to use AI to generate multiple design options, evaluate them through simulation and testing, and make decisions that way. This isn’t just for designers—it’s about spreading this mindset to PMs, engineers, QA, and everyone involved in building products.

We’re focusing on two main areas right now:

• Enabling anyone to create a UI: Building tools that allow anyone with an idea to generate and deliver UI
  
• UI simulation: Using virtual personas to simulate how users would evaluate and interact with generated UIs, solving review bottlenecks.
  

At first, we planned to build everything in-house, from planning to delivery. Now, though, we’re also making flexible use of external tools as we explore what systems we really need to create. This initiative will continue through the end of June. Rather than waiting for a perfect plan, we’re testing new technologies as soon as they emerge and refining our approach through constant trial and error to pursue what the ideal should look like.

To drive this ambitious project, we’ve divided roles clearly:

• VP/Project Management (@narico): Tracks tech trends and leads the team.
  
• Enablers: UX engineers and others who support technical implementation.
  
• Area Experts: Specialists who explore ideal workflows.
  
• Reality Checkers: Members who test practicality and keep us grounded.
  
• Bridging Experts: People who connect design and engineering with a shared language.
  

Through this teamwork, we’re aiming to fundamentally transform how Mercari creates products.

@gomichan: So it’s not just about using AI within the existing flow—it’s about rethinking the flow itself.

@narico: Exactly. The essence is to use new technology the day it’s released, incorporate the concepts into our goals, and decide what’s worth building and what isn’t. By staying flexible and working closely across teams, we want to uncover what we truly need to build.