Here at Mercari, most members have already started using AI in their daily work as we move toward becoming AI-Native. However, some employees may still not fully understand what Mercari truly means by “AI-Native”—both in terms of what an AI-Native company looks like and the people it refers to.

How exactly are Mercari’s AI-Native employees using AI? In this volume of Yurino’s AI Question Corner—presented by new hire and AI Task Force member Yurino—we speak with data engineer @hira. Since transferring to the engineering organization of Japan Business (JB) in January 2026, @hira is now responsible for building systems that accumulate and organize data and also put it to work. What has become clear to @hira by using AI is her evolution from a person who analyzes data to a person who converts analytical thinking into patterns. This work is not just about improving efficiency—it’s about turning thought patterns into systems. In this article, we probe the background and current status of this challenge.

More than just coughing up numbers—looking at the work of a data analyst

@Yurino: So @hira, in what situations have you used AI in your work as an analyst?

@hira: The thing I’ve used AI for the most is organizing my thoughts in order to ask the right questions.

I often receive requests for specific kinds of data, but I’ve always made it a point to understand why the requester needed those particular numbers and what kind of decision-making they were tied to. Therefore, I’ve used AI to compile communications in Notion and Slack. Once I understood the background and context of each initiative, only then would I propose what data was really needed. In some cases, I’ve also determined that what an initiative probably needed was actually a different data set. As an analyst, I’ve always believed that starting with an understanding of what an initiative wanted to achieve was essential.

@Yurino: So in other words, AI enables you to quickly understand context, allowing you to make proposals based on more in-depth analyses. Could you share what made you start using AI?

@hira: I was motivated to start using AI because I wanted to reduce careless mistakes, such as errors when writing queries (code for data extraction). By using GitHub Copilot (an AI tool that makes code suggestions), I was able to harness code completion, which not only reduced mistakes but also improved my overall work efficiency. From there, I expanded my use of AI beyond just coding, using it for things like understanding context and creating documents.

@Yurino: Has AI changed the way you conduct analyses, and does it affect the results?

@hira: The most significant change has been in the clarity of my output. Previously, I would simply write down or speak the information I wanted to convey as I understood it, which sometimes made it difficult for others to grasp. By using AI tools, I’ve been able to tailor the structure and wording of my communication to suit the audience—whether I’m speaking to PdM members or sales managers—which has improved the clarity of my delivery. I feel that by supplementing my skillset, AI has allowed me to focus more on essential analyses and proposals.

@Yurino: You mentioned that not only are individual employees using AI, but that Mercari is also promoting its use at the organizational level. Which of the company’s initiatives to promote generative AI would you say have been particularly impactful?

@hira: The most impactful initiative has been the automation of A/B testing. When we’re releasing a new feature, we compare two versions of the app, one with the screen as it appears after the new feature is added, and one with the screen as it appears before the feature is added. The comparison is conducted under the exact same conditions to verify whether key metrics such as purchase rates improve. Conventional A/B testing used to involve the manual creation and monitoring of metrics, which took a lot of time. While a system was provided for analysts to register metrics themselves, it was not user-friendly for non-engineers, so I used AI to develop a system that would make the registration process more accessible. Furthermore, based on those metrics, I also used AI to enable queries to be generated automatically for in-depth analysis. Probably the biggest change was achieving a state where registering metrics allowed our employees to execute everything from judgment to in-depth analysis seamlessly.

The challenge of transforming a wealth of insights compiled as a data analyst into an organization’s system as a data engineer

@Yurino: It sounds like the true value of data analysts will also change as AI becomes able to support analysis to a greater extent. How do you think this role will evolve going forward?

@hira: I believe the role of the analyst is changing from being a person who analyzes data to being a person who incorporates analysis patterns into systems. As AI makes it possible for anyone to handle data, there is a risk that the patterns of analytical thinking that analysts have compiled to date will not be leveraged and the data will instead be used in low-quality decision-making. Even if things appear to progress quickly at first, if the necessary decisions cannot be reached, overall efficiency will decrease and the insights in the data will go unused.

For example, in the analysis of Mercari’s gross merchandise value (GMV), there are some implicit thought patterns, such as approaches for breaking down the data and methods for formulating hypotheses. This thought process is precisely what makes analysis valuable, so it’s important to verbalize it as a pattern and make it reproducible. I believe the future role of analysts will be to refine the thought processes that AI cannot replicate, convert them into patterns that can be applied by AI, and turn them into shared assets of the organization rather than hoarding them as the personal weapons of analysts. 

@Yurino: So you’re saying that it’s becoming important to transform the patterns we think up so that they can be reproduced by AI. What exactly do you mean by “thought processes that AI cannot replicate”?

@hira: Broadly speaking, there are three aspects.

The first is “creating data.” AI can only handle the data you give it access to, so if there is no data to begin with, you first have to design a system for data collection from scratch. For example, if you want to know how the user feels about something, you have to start by designing a survey to obtain that information.

Second is “defining what to measure and how.” If you don’t define which metrics to use to break down and evaluate the data, AI will not be able to analyze it correctly either. Imagine you’re looking at increasing GMV, for instance. In order to proceed, you need to consider which is more important, increasing the number of users or the purchase amount per user. To determine your next set of actions, you must first be able to break down the metrics.

Third is “prioritizing.” AI can show results, but deciding what to prioritize is a role that should be handled by a person. For example, if the goals of different teams conflict, they have to figure out how to find common ground. I believe these three points are the thought processes that we need to train AI to apply.

@Yurino: So then, the preliminary stages of analysis—namely, deciding what to collect and what to prioritize—are important, correct?

@hira: That’s right. We have to systematize the tacit knowledge that analysts have cultivated so far and create a system to incorporate them into AI. With that in mind, I transferred to a position with the engineering organization in January of this year. If the job of an analyst is to create, define, and prioritize data, then the next step is to systematize all of those patterns and embed them into a system that the entire organization can use. I see that as the role I fulfill in my current position of data engineer.

@Yurino: So as a data engineer, what challenges do you want to take on going forward?

@hira: Broadly speaking, I want to embed the analysis system with more than just a definition of data. I also want to include ideas on how to conceptualize data. Currently, I’m developing a system that generates reports, written in natural language, detailing the reasons behind the sales fluctuations for each seller on Mercari Shops. Rather than just outputting numbers, I want to establish a standard where anyone can make judgments from the same perspective. This involves building evaluation criteria that prompt the user to consider a particular next action if they see certain numbers fluctuate.

For example, if one of our members working in sales is responsible for hundreds of stores on Mercari Shops, it is difficult to check the movements of all sellers in detail every month. To handle this, an analysis system will automatically extract the sellers to pay attention to and the factors involved, and then suggest actions that should be proposed to the sellers. If we can suggest where to focus, decision-making should become much faster on the ground. The point I’m driving at is that as a data engineer, I only provide the systems. In doing so, I assist our sales teams and the actual data analysts by enhancing what points should be highlighted and how to make suggestions. I will hand off the patterns of my thought processes to AI so that they can be reproduced across the entire organization. This is the challenge I want to take on now as a data engineer.

AI-Native means making thought processes reproducible

@Yurino: Could you describe the sort of person you consider to be AI-Native?

@hira: A person who can verbalize their thought processes and judgment and embed them into a system so that AI can reproduce them. AI is quite a stabilizing force within the framework we have. That’s why it’s important to first hone the thought process elements that AI can’t replicate, and then systematize them into patterns that can be utilized by AI. An AI-Native person is also someone who can not only use AI themselves, but also spread their approaches to other people and teams in a usable form. I believe that is the true meaning of AI-Native.

@Yurino: To wrap things up, please share a message with the Mercari members who are gearing up to take on new challenges using AI!

@hira: First, I would like you to think about how you would break down your tasks if you were to “put them in the hands” of AI. I recommend starting by verbalizing your daily repetitive tasks and decision-making criteria as patterns. On top of that, I encourage you to use AI to make these processes reproducible for others as well. When others try to use the systems you’ve created, you may find there are areas where your explanations weren’t as clear or complete as you thought.

Let’s all take the AI-Native plunge together at Mercari!

Photography: Tomohiro Takeshita

Check out other articles from this series!