The era when artificial intelligence (AI) can change the growth strategy of a company at the grassroots level has arrived. Taking on the bold and challenging task of leveraging AI from the product perspective are Mercari CPO Takaaki Shinohara (@unryu-in) and CXO Mayumi Narisawa (@narico). They paint a vision of the future where AI has permeated Mercari from end to end, and the user experience and corporate value have improved simultaneously.

Mercari has thus far been among the quickest companies in Japan to find value in AI usage and has taken steps to implement these technologies aggressively in its products. In the time that we’ve done this, generative AI and large language model (AI/LLM) technology has also developed dramatically.

Currently, Mercari is leveraging the strengths of our image-driven platform, which holds over 4 billion images of items listed on our marketplace, while also leveraging the infrastructure we’ve built to date in order to draw out the maximum potential of AI. However, there is no shortage of red tape when it comes to fintech area regulations and governance.

One question that Mayumi has put forward is, “Can we make essential functionality and features more fun to use?”
At the same time, Shinohara asserts, “We should value the spontaneous encounters that occur at a flea market.”

What they both want is for Mercari to become more than just an AI company. Instead of this, they are looking to develop Mercari into an “AI-led growth company” that leverages AI strategically to achieve soaring growth. This means creating a dynamic workplace that fosters formative experiences for each and every employee while providing a user experience that exceeds expectations. Mercari is in pursuit of its vision of a soon-to-be future driven by AI.

The AI wave according to Mercari

—To start things off, how has Mercari used AI thus far, and what initiatives are you working on now?

@narico：Since my time as the CPO of Merpay, I’ve also managed a team that promotes AI credit. I managed the organization with the aim of bringing the product and AI as close together as possible. As I was focusing on this, I began imagining what the next big thing would be, and that’s when my eyes fell on debt collection chatbots. The performance of sentiment analysis using AI has also improved substantially, and I thought that we could probably use AI to better empathize with each of our various users’ states of mind concerning repayment.

However, we also ran into some issues. If a chatbot were to ever generate something that it’s not supposed to say, it’s game over. For example, if it made a false statement like, “You can repay loans using cryptoassets,” that would hurt us. In the end, AI can have a big impact when its accuracy is improved, but we can’t ignore the risks it brings.

We actually consulted with the relevant people on Line 2 (department in charge of indirect management including risk management, financial, legal, and compliance issues). They explained to us why using AI wouldn’t work, but we didn’t immediately throw up our hands. Instead, we asked ourselves if maybe there was something else we could do. In the end, we spoke with the person who was CEO at the time, who told us, “Even if we manage to improve the accuracy of the AI and avoid all the various risks, we might only be able to provide the feature to users in certain areas. If that happens, what kind of impact can we expect to have?” When I was asked if I thought we’d make a big impact six months later or a year later, honestly I didn’t feel that we’d be able to just yet, and I gave up for a little while.

Since July, when I became CXO, and up until the end of December, the Product team had a goal of creating a so-called AI-native culture. I’d like to use the experience I’ve acquired in consolidating the different organizations I’ve worked with so far to take on the current challenge of Design x AI (design that integrates AI). I want to stick closely to our road map and move it ahead.

Specifically, we have started a trial project called “pj-ai-creator” with internal AI design engineers and AI Lab members. To start, we’ve been focusing our efforts on generating banners that are characteristic of Mercari. This also leads to a lighter task burden for designers.

AI is good at analyzing sentiments and images, so in the not-too-distant future, I think that AI will be able to design something like what a designer would create. I think there’s a lot of potential, especially in the graphic design field. First, we had to train an image generator AI on the large volume of design data that our designers have designed to date. Next, we had to create an illustration that was like a 3D model; all of this was an education for me as well. At first, it had a hard time with things like generating hands, but as the training process progressed, the AI model gradually started to generate cute illustrations that would appeal to people, and the 3D-model illustrations started to resemble the work of our designers. I have to say… the rate at which AI grows is astounding.

In the future, I’d like to provide this technology to sellers. We’re still at the early stages of creating the structure, but we’re also thinking about a service that would allow both B2C and C2C users to create their own banners and to customize their items. At first, I think we’ll probably have a series of templates that people can use to suit their own needs, but we should be able to create more flexible templates as LLMs evolve.

Ultimately, I also think it will be possible to launch them for UI designs as well. When you investigate the UI of an app, it’s just a combination of images and text. If we become able to use AI to generate the UI, we might be able to sell that externally as a tool.

@unryu-in：To begin, our goal is to verbalize and visualize the product roadmap. As we realize the future described in this roadmap, we will choose the most exciting method that best captures current trends from among multiple approaches.
Just to give you an analogy, in Tokyo you can walk from the Roppongi area to Hiro. However, you can also use a rental scooter to travel the same route more quickly.

Next, we have to look at the scenarios in which we use AI. For users to use Mercari, there are some features that we must provide and others that we can provide to make their user experience better and help them sell their items more effectively; AI can help with both required functionality and improvements.

However, that said, we have no intention of completely automating communication between users. We are not even thinking about getting rid of open comment fields or restricting users to using menus of automated responses.

We will leverage AI in order to concentrate on the tasks and fun parts unique to Mercari, such as users communicating with each other and searching for bargains. This will take care of some of the essential things that must be done to use the product, minimize work costs, and maximize our user’s opportunities for lucky finds.
Ultimately, I think that we would like to leverage AI to maximize the fun our users have using the product.

@narico：There are also issues around implementing AI. For one thing, cost is an issue; implementing AI costs a ton of money. We face a dilemma where, even though we implemented AI to cut costs, it could instead end up costing a lot of money. That would be a clear sign that our priorities are in the wrong order.

However, Mercari is the type of company that is willing to make investments for the sake of making our UX fun. We will do our best to use AI on the essential parts of the UI/UX, but usability issues encompass a range of areas. Simply improving operability is not a solution. So, can we instead take the essential parts of the platform, like listing and shipping, and make them more fun to use? We believe that this is where AI has potential.

Mercari is an image-driven service that functions due to the over 4 billion images of items that our users have listed. There has to be a way we can leverage this strength. AI Listing Support is one of the best examples of this. Released in September 2024, this feature allows the user to completely list an item in as few as three taps simply by taking a picture and then choosing a category.

Moreover, I sense potential in the direction of making the platform more convenient by using LLMs to ascertain users’ emotions. Instead of just improving searchability as we have thus far, surely we can increase the overall amount of fun users have based on categories and keywords. We could accomplish this by making recommendations that are based on some form of unspecified emotion or situation that ends up leading users to what they are looking for.

A new-era marketplace experience interwoven with AI

── How would you like to use AI to develop Mercari’s product?

@unryu-in：I think the thrill of Mercari comes from the enjoyment people experience from spontaneous encounters in a flea market-like environment. This is where I would like to use AI to both improve the ability of buyers to discover items and place the rare finds that sellers list somewhere they’ll be seen by the right people.

As for whether it’s all that easy to get an item where it will be discovered, I believe we can leverage AI to increase the likelihood of this happening. To paint a clearer picture, we’d like to be able to create something where we provide the seller information like “this item will sell for such and such price” and “buyers are looking for this item,” based purely off of what the seller points their camera at.
I think we can also use AI for buyers as well, such as by using it to make suggestions for putting together outfits or helping a user buy furniture based on what they put in front of their camera lens.

@narico：Yes, exactly. I think there are other improvements we can make in order to bring items into view using AI/LLMs. For instance, if a user takes a picture in their living room to use in a listing and there’s a children’s toy or something else cluttering the background, the object could be blurred out automatically.

── Contrarily, are there any issues associated with using AI?

@unryu-in：There are still issues with using AI for customer support. Starting sometime around 2018, we launched a suggestions guide, but determining what to use as the source information caused complications. When we collected feedback from our user community, we found that there was a risk that the guide would produce answers that were in a sort of gray area, and some of the information was not appropriate to use as an official answer from Mercari. We have to consider carefully to what extent we use AI to express an official company position.

@narico：I feel like using AI in the fintech area is especially difficult. The debt collection chatbot that I mentioned earlier is another example. Currently, it tends to end up being used in areas that are relatively low risk. If you try to use it for something bolder, there’s a high risk of it infringing on laws and ordinances.

And like @unryu-in said, there are some difficulties with using AI as a legal personality. From the user’s perspective, the statements of an AI model are treated the same as the statements of a company employee. If an AI model misspeaks, it’s treated as an official opinion of the company. Managing this risk is a major topic.

What does “AI-led growth” look like according to Mercari?

── To wrap up our interview, could you tell us about Mercari’s vision for aiming to become a company that wields AI?

@unryu-in：What we are aiming for is to become more than just simply an AI company. Our idea is to become a company whose growth is led by AI.

However, if you apply this to everything we categorize as essential, this ends up being kind of boring for the user. How do we draw in the people who will become our loyal fans, while keeping them from feeling inconvenienced? I think that is one very important thing to note about using AI.

@narico：Yes, exactly. People are saying that the period for building our infrastructure is over. So, going forward, it will be important to see how far we can accelerate growth using AI.

@unryu-in：For engineers, exciting technology is what motivates them as well. And that is the sort of environment I want us to create. Talented engineers are often attracted to a development environment that has a grasp of the latest trends, but I believe that going forward AI-led growth is going to be a keyword that engineers value.

@narico：We have called ourselves an IT company for a while. So in the same way, I didn’t feel there was anything unusual with us calling ourselves an AI company. However, taking it one step further, I think that becoming an AI-led growth company is the most ideal form that we should strive for. I think that’s the extent to which we should use AI.

Also from the perspective of hiring engineers, I have a feeling it’s going to be a keyword of the next era of IT. Companies will leverage their own technologies to achieve growth. By holding up such a passionate vision, I trust that we’ll be able to create an environment that is exciting for engineers.

However, AI is at best just a means to achieving our mission. Our objective is to provide our users with a good experience. In order to accomplish this, we will use AI to accelerate our growth. That is the type of AI-led growth company that we are aiming to become.