One AI Team member’s decision to change paths and work on the US app: Why? #BoldChallenge

Mercari has grown to become Japan’s largest marketplace app, and we are continuously striving to accomplish our mission to “Create a global marketplace where anyone can buy & sell.” On this journey, our three company values, Go Bold, All for One, and Be a Pro, have always played a fundamental role. Of the three, we believe that Go Bold in particular is a perfect representation of the Mercari culture. To create a large impact on the world, taking on bold challenges and learning from failures is inevitable. In the “Bold Challenge” series on Mercan, we will shine a spotlight on Mercari members who have continued to strive in our evolving environment and reveal why they chose to take on these challenges.Read the previous article in this series here.

The fourth article in this series features Takuma Yamaguchi, a founding member of the AI Team. The AI Team works on product development using AI, one of Mercari’s technological strategies.

So far, the Mercari AI Team has released AI listing, which uses image recognition to automatically fill in the item name, category, and market price based on an image of the item, and other features such as detection of illegal listings on the app, and image search, which allows users to search using a picture of the item they’re looking for. Yamaguchi, who was on the front lines of the AI Team, is currently working…in the Mercari US development team? Why Mercari US? Tairo Moriyama, the director of the Search/AI Engineering Team, who worked with Yamaguchi on AI-related product development, interviewed him to find out his story.

“I just decided I wanted to do something with data, and applied to Mercari.”

Moriyama: Now that I think about it, I think I’ve been working with you on AI-related product development for about two years now. You joined Mercari a little bit before I did, right? Did you join Mercari because you wanted to do AI-related development?

Yamaguchi: It wasn’t really that I wanted to do AI; I just decided I wanted to do something with data, and applied to Mercari. At my previous job, I created systems for data analysis, so I was deciding between going for an analysis position or an engineering position.

Takuma Yamaguchi (Mercari US development team)

Moriyama: That’s pretty broad.

Yamaguchi: So I joined Mercari in 2016 without a specific position in mind. At first, I joined the product development team, but I was transferred to a different team the next day…

Moriyama: The next day?

Yamaguchi: Yeah. I was assigned to a team whose job was to come up with ideas for Mercari’s next revolutionary concept, and brainstormed new ideas every day. One month later, I started working on software development for the US version of Mercari as a backend software engineer, and did that for about six months. I was involved with a project to reduce the number of inquiries the US help center received from users as much as possible.

Moriyama: The AI Team was established after that, right?

Tairo Moriyama (Search/AI Engineering Team)

Yamaguchi: The AI Team started as a virtual team in December 2016. But at the time, we couldn’t decide on what we would actually do, so we just shared our thoughts on how we could use technology like image recognition and machine learning at Mercari on the company wiki and on Slack. Then, Shunya Kimura (currently director of Mercari’s AI Team) joined Mercari, and it became a distinct team.

AI listing, detection of illegal listings, image search, and creating the foundation for machine learning

Moriyama: As I remember, the team decided that they would work on features for AI-based listing and detecting listings that violate Mercari’s rules. At the time, I was working on a project to improve the user experience for searching on Mercari. I remember reading a wiki page you wrote, thinking “I wonder if this technology could be used in searching?”, and asking you to give me a demo.

Yamaguchi: Right. It’s hard to understand from just documentation, so I created a demo for similar image search and image recognition.

Moriyama: There was definitely a sense that despite how convenient Mercari was, it took a lot of work to list items, so we were looking for ways to make the listing process easier for users. When we showed the AI Team’s demo to the head of the product division at the time, Iyo (currently Merpay CPO), he thought it was really interesting and could be used to make listing easier. This led to the concept of AI listing.

Yamaguchi: It all happened so quickly… I was invited to a meeting to brainstorm the concept, but the discussion ended up being “how much of this can we implement” and “are we going to do it or not?” Normally, developing features using image recognition takes about a year, but we were told to release it within the quarter. I was shocked at how fast everything was moving. We were already half a month into the quarter by the time we had the meeting, too.

Moriyama: After releasing AI listing, you were also involved in feature development for CRE (Customer Reliability Engineering), right? At the time, Mercari was releasing new features for users one after another in rapid succession, and CS (customer service) kept falling behind. So in order to maintain Mercari’s safe and secure environment using technology, we boosted the priority of CS and implemented detection of listings that violate Mercari’s rules using image recognition.

Yamaguchi: I first had the idea of implementing machine learning to improve CS operations when I was working on development for Mercari US. I watched CS members work, and noticed that even when they found illegal items listed, the sellers of those items would do things like put extra keywords in the listing title to make it hard to find the items in search results. Of course, CS members knowledgeable in these areas had an idea of what keywords to use and what to avoid, and were able to find the items anyway, but that specialized knowledge made it difficult for other members to do the same work without high training costs. So we thought, “What if you could search for listings with similar images?” and released this feature.

Moriyama: The releases of AI listing and detection of illegal items marked the real beginning of the AI Team. After that, the image search feature launched in March 2019. This is particularly special because it can be used for so many things; in addition to image search itself, the API for this is also essential for a new feature (coming soon!).

Yamaguchi: That was when AI technology in Mercari really took off. On the other hand, while we were developing features like AI listing and image search, the AI Team was also working to build a foundation for machine learning. In other words, we were creating a base to take in and learn from data, and then actually implement into a feature. So the AI Team has some members working on the foundation for machine learning, and some members working on developing features to solve issues.

Moriyama: Which were you doing?

Yamaguchi: I was always part of the group developing features for Mercari.

Transferring to the Mercari US development team to accelerate the US’s use of machine learning

Moriyama: I always felt as if you and I were comrades in arms, in the sense that we worked together in similar positions. But then I started focusing solely on search improvements, went through a major role change from product manager to engineering manager, and became a director supervising the entire team. You, on the other end, made the decision to transfer to Mercari US.

Yamaguchi: I did. We’re in a stage where we’re working on more AI-related projects than ever before, so to be honest, I almost can’t believe that I made the decision to transfer to working on Mercari US myself. (laughs)

Moriyama: What led up to your transfer to the Mercari US development team?

Yamaguchi: First, I thought that by joining the team, I could accelerate the machine learning features and projects in Mercari US. Like I said earlier, Mercari JP has released a number of features utilizing machine learning, including image recognition for CS and AI listing and image search for users. But I had the impression that Mercari US wasn’t progressing quite as well.

Moriyama: This seems like it would depend heavily on whether or not there’s someone who can lead AI-related technical projects.

Yamaguchi: Generally, teams that work with new technology like AI tend to set their goal as “utilizing the newest technology,” but that alone won’t work. This is something Mercari has to be careful of, too.

Moriyama: Exactly. Our mission is to improve the user experience. To do so, it’s important to keep in mind that we can’t just think from the perspective of AI. Using AI will help us with scalability strategies, but it’s not guaranteed to work. Also, development takes a long time. Whether something will work or not is actually often decided near the beginning. Does the logic behind your project’s solution to the problem look sound? Is it possible to resolve the problem efficiently using machine learning? I think task setting, where you try to answer these questions, is the most important step.

Emphasizing contribution over personal growth

Moriyama: Earlier, I said I was surprised when you transferred to Mercari US, but actually, I had a feeling something like that would happen. Mercari US is a really unique experience you can’t get anywhere else, so I think it’s a great opportunity for you.

Yamaguchi: Personally, I tend to choose the option where I can contribute the most. I also don’t really like to talk about it in terms of my personal growth, or taking on new challenges. (laughs) Honestly, it doesn’t matter to me whether I can grow and learn there or not. I care more about whether I can contribute there. If there were a lot of people at Mercari who could do the same things I can, I would probably quit.

Moriyama: Alternatively, if there were a lot of people who could do the same things as you, you could shift your focus. Personal growth is really just a result of your work.

Yamaguchi: That’s true.

Moriyama: I think one of Mercari’s virtues is that you can hire or train someone to take on your current role at a much higher level than you, pass that role on easily, and then move on to the next challenge. In my experience, as my position changes, I can feel my horizons broadening. In your case, you transferred to Mercari US because you felt that you could contribute there, but for other people, that choice could have been to go to Merpay. Or there may be people who see potential for them to contribute in a completely different position. I think this is a very bold way of thinking about one’s career that represents Mercari really well.

Yamaguchi: It’s a really good aspect of Mercari, isn’t it? I’ve been involved in applications of AI at Mercari for a long time, and I feel like I have a good idea of the methods for using it within the domain of a marketplace app. Now, I need to look at more creative ways of utilizing this technology, regardless of whether I’m working on JP or US. Right now, Mercari US is considering utilizing AI for the price estimation and suggestion features. I think we could finetune some features from Mercari JP and take AI listing one step further.

Moriyama: That sounds great! I think no matter what technology you’re using, what’s important is that you can use it to solve challenges in the business.

Yamaguchi: I agree.

Takuma Yamaguchi

Yamaguchi has been involved in research on machine learning, image recognition, and computer vision for more than 10 years. In addition, he works in system development related to big data and the operations research field, and web application development. In his current position as an engineering manager at Mercari, he actively uses machine learning technology, mainly image recognition, and works every day to make Mercari a better service for its users.

Tairo Moriyama

After graduating from Waseda University, Moriyama launched a recruiting/HR startup before joining BizReach Inc., where he worked on developing a search engine for job postings and improving search features using natural language processing technology. After joining Mercari in November 2016, he worked as a product manager to improve search features in both the US and JP versions of the Mercari app, and released new features using personalization and AI. In January 2018, he established the CRE Team to use technology to improve CS at Mercari. Currently, he is the director of the Data Science, Search, AI, and Tech PM Teams.