The on-demand work service Mercari Hallo is an app that helps fill sudden labor shortages faced by partners (companies looking for workers) and enables crew (workers looking for jobs) to take on work in their free time.

At Mercari, we’re working on various optimizations to deliver a new work experience to both partners and crew. In particular, we utilize AI and large language model (LLM) technology to provide new experience value, with features such as Simplified Job Listing, which automatically generates job listings based on templates and basic information provided by partners, and job listing personalization based on an individual’s usage history.

In this article, we sat down with Mercari Hallo CEO @Asami, Vice President of Engineering @godriccao, and ML engineer @shido to discuss how Mercari Hallo utilizes AI/LLMs now and their vision for the future.

Realizing a world where finding work is quick and easy for anyone

── First, can you explain Mercari Hallo’s position in the on-demand work market?

@shido：One important factor in the on-demand work market is promptness. Business partners want to secure workers as soon as possible when they need them. Workers are looking for work that they can do immediately. It’s important to meet these needs from both sides.

Another important factor is variety. On-demand work covers a wide range of work opportunities, from waiting on customers at a restaurant to light-duty warehouse work and helping out at events. We aim to match our crew with all sorts of work in line with our concept of “quick and easy for anyone.”

@godriccao：The major difference between Mercari Hallo and other on-demand work services is that we focus on providing users with new possibilities. Many services place emphasis on encouraging users to repeat the same job to improve work efficiency. But we aim to provide an attractive experience to first-time crew and create opportunities to explore new possibilities.

── What initiatives are you working on to achieve that goal?

@Asami：One of Mercari Hallo’s strengths is that we can utilize Mercari Group’s technology. We put our knowledge about user behavior analysis gained through Mercari and our expertise about utilizing AI/LLM technology to use in the Mercari Hallo service. What’s particularly important is how we designed our service with data applications in mind from the start.

@shido：That said, we faced a number of challenges when first launching the service. One was figuring out how to recommend the most appropriate jobs to crew when we hadn’t yet fully understood their behavior patterns; another was how to respond to sudden needs raised by partners. To address these questions, we went with the strategy of first establishing the minimum necessary features, then gradually expanding use of AI/LLMs.

@godriccao：We pay particular attention to balancing technology and human work. For example, when opening up discussions with a new partner, someone from our sales team visits their office and interviews them to get an idea of their specific needs. We then store that information as data, analyze it with AI, and use the results to make a proposal. Through this cycle, we hope to support our sales team to work more efficiently.

Launching the AI/ML Team and developing features to drive the use of data

── How was the AI/ML Team formed?

@Asami：We first formulated our strategy over a period of three months, from July to October 2024. During this time, we paid particular attention to data collection and analysis, when we would implement the technology, and how we would measure the effects.

For data collection and analysis, we needed a very elaborate design. Take crew behavioral data, for example. We needed to be able to track the user journey from start to finish—from job listing browsing patterns to actual applications, employment statuses, feedback after completing a job, and repeat crew rate. This needed complex design beyond what we could get from simple log collection. Also, this data is personal data for crew, which meant we needed to be very careful handling it from a privacy perspective.

@godriccao：Data regarding our partners is just as important. The environment we’ve created allows us to gather and analyze data from various angles, such as job listing patterns, wage trends, efficiency of matching crew to partners, crew retention rate, and satisfaction levels. By comprehensively analyzing all of this data, we can enhance the effectiveness of our matching algorithms.

── Of the features Mercari Hallo already offers, are there any that particularly stand out?

@shido：The Simplified Job Listing feature, which I’d say is the first feature that comes to mind when it comes to AI/LLMs in Mercari Hallo, actually took quite a lot of adjusting to get to where it is now. At first, we faced several challenges, like finding the right balance between length and readability, making sure all of the necessary information was included, expressing each partner’s individuality, and conveying what would make the job attractive to crew.

To address these, we gradually fine-tuned our LLM model. For example, we analyzed the characteristics of job listings that had high application rates, and then incorporated those components into the automatic creation process. We also optimize our templates to highlight the information considered particularly important for each industry and job type.

@Asami：We’re also working to strengthen our initiatives regarding safety, from both technical and operational perspectives. Our system for detecting inappropriate job listings first runs a basic check for certain keywords, then a detailed check including contextual understanding using LLM technology, then an analysis for similarity to past cases. Finally, a human carries out one last check. By combining automated checks with manual checks, we are able to achieve both speed and a high level of safety.

As an on-demand work service, we place the highest level of importance on protecting crew and complying with various laws and ordinances, and we aim for a strong team structure while leveraging technology to do so.

Creating value from three perspectives

── Can you tell us more about feature enhancements you have planned going forward?

@Asami：We’re working on feature enhancements from three perspectives: enhancements for crew, enhancements for partners, and enhancements for Mercari employees. For crew, we’re first working to simplify the initial registration process. We want to create a system that enables crew to sign up for the service easily, with as little information required as possible, while also analyzing behavioral data based on their usage of the service.

At the same time, we’re also working to provide as much information as possible regarding the content of jobs listed on the platform and the work environment, in order to mitigate the concerns people doing on-demand work for the first time may have. Additionally, we recommend jobs to crew based on their experience, and we’re developing features to clarify jobs for people with no experience and to recommend opportunities for crew to gradually improve their skills.

@godriccao：One thing we want to develop for partners is dynamic pricing. This is an extremely challenging area in terms of the technology involved. The same position can have varying levels of demand day to day for many reasons, and the crew available to fill the position changes daily as well. We’re making plans to develop an algorithm that can propose the optimal wages to partners comprehensively considering multiple factors.

@shido：We’re also gradually developing features to support our sales team. We’ve been working on automating training support for sales reps and basic sales activities, and going forward, we plan to utilize AI/LLMs for things like predicting the optimal timing to reach out to partners and generating proposals personalized for each partner. On top of this, we’re also considering introducing a system to make calls to partners utilizing AI/LLMs in the future.

Aiming to become a next-generation workstyle platform

── What specific investments in technology are you making?

@godriccao：Right now, we’re investing in three areas of technology. First, LLMs show results relatively quickly, so we’re using them for things like enhancing the quality of automatically generated job listing text, increasing the accuracy of fraud detection, and optimizing customer support.

In the field of AI, we’re considering how to build job listing personalization systems from a more medium-term perspective. Finally, as a long-term investment, we’re working on developing algorithms to meet the unique needs of the on-demand work industry.

@shido：In order to turn these investments into practical value, we measure their effects using concrete indicators, like the time needed to find a successful match, repeat usage rate of crew, and satisfaction of partners.

── To finish off this interview, please share with us your vision for the future with Mercari Hallo.

@Asami：We’re aiming for Mercari Hallo to be a platform that focuses on exciting discoveries for crew, using AI to recommend surprising jobs, discover latent aptitude, and help crew find new ways to work.

At the same time, we hope to achieve sophisticated personalization, such as providing information tailored to each individual’s traits, supporting crew in building careers, and making recommendations based on our crew’s lifestyles. Of course, in doing so, we will work with other teams to ensure that we keep our users’ personal data private and address bias to guarantee fairness. On top of this, I’d also like Mercari Hallo to be a community where crew with similar interests can connect and where partners and crew can build relationships.

@godriccao：We’re trying to create a new standard workstyle: to find a more flexible and varied way of working that goes beyond the conventional boundaries of full-time and part-time employment. I want Mercari Hallo to be a platform that utilizes technology to help each individual find the workstyle that’s right for them.

I think that with our technology, we can become more than just a platform that matches job listings with candidates, and instead create a future that unleashes the potential of work to create new value.