Until the biggest skeptics become the biggest AI power users: The Meitetsu Bus shift schedule reform

ALGO ARTIS works on "optimizing complex operational plans" in various fields by leveraging AI and algorithms. One such area is the "shift schedule" that bus operators must create without fail every single day. This plan, which determines which driver is assigned to which route, is an "absolutely non-stoppable plan" where even a single day of gaps leads to service cancellations.

However, there are still many workplaces that rely on the "experience and intuition of skilled staff" to create these shift schedules. Now that the shortage of drivers is becoming critical, how to change this personnel-dependent, high-load work structure across the entire industry has become an urgent issue.

This time, we are spotlighting the shift schedule optimization project at Meitetsu Bus Co., Ltd., which operates 11 offices, 700 vehicles, and over 1,300 drivers, mainly in Aichi Prefecture. The initiative with ALGO ARTIS began in April 2023, and full-scale operation of the "AI Shift Schedule" started in March 2025.

From the staff member who was once the strongest skeptic to becoming the one who uses AI better than anyone else—we asked four staff members about the background of the implementation and the transformation of the workplace.

The work that only craftsmen could handle behind the "absolutely non-stoppable regional lifeline"

What exactly is the "shift schedule work" at Meitetsu Bus?

Mr. Okada: The shift schedule is like a daily shift plan that supports the foundation of bus operations. If there is a gap even for one day, it leads to service cancellations, so someone must continue to create it every single day. At Meitetsu Bus, shift managers at each office create them on a rotation basis, but it took 8 to 10 hours, and sometimes a whole day, to create one day's worth.

It is a plan that is not straightforward enough to require that much time, isn't it?

Mr. Horii: To give an easy-to-understand example, the difficulty lies in having to keep building a plan in a state where "there are 100 courses but only 80 people (scheduled to work that day)." If there are enough drivers, you just assign them as decided, but such cases are rare.
Now, including us, there is a shortage of people in the entire industry, so how to manage—such as asking for holiday work or breaking up one course and connecting it to another to reduce the number—is where the skill of the shift manager comes into play.

Mr. Fukushima: What takes up time in particular is the time calculation based on laws and regulations (standards for improving working hours, etc., for automobile drivers). Since the upper limit of the driver's restraint time was shortened from the previous 16 hours to 15 hours, the difficulty of building it by hand has increased even further. Because there are also regulations for rest periods for drivers, the person in charge of the next day cannot start work until I finish building the plan. There were times when it took until late at night to barely make it in time.

You mentioned building it by hand; were you formulating the daily plan manually in the past?

Mr. Horii: Yes. It was called "hand-building" because it was created manually using paper, pencils, and calculators. When creating the plan, we even considered the personality of each individual driver. We would be in trouble if an accident occurred because we forced them to take a shift, and we also have to ask for their cooperation in the future, not just for that day.

Mr. Ono: You cannot do this work well unless you have a 24-hour clock in your head. It is a highly difficult task to memorize the rules for restraint time and rest periods and instantly perform calculations for everyone. Moreover, because regional characteristics and unique rules are completely different for each office, the "method that worked at that office" cannot be used as is at another office. It was truly a craftsman-like job that relied on the experience and intuition of the person in charge.

So the high hurdle for acquiring skills and the heavy burden of creating plans were the issues.

Mr. Okada: As a result, other issues were also arising. Shift managers are supposed to be in a position where they must be in the closest contact with drivers, but they were so desperate to refine the plan that they stayed at their desks all day from the morning and had almost no opportunity to talk to the drivers. An image of "I don't want to be assigned to be a shift manager" had become established among the younger generation who saw their seniors like that every day.

"There is no clear evidence. But I felt a potential that made me want to entrust it to them."

The project with ALGO ARTIS to improve shift schedule creation work started in April 2023. How did this initiative come about?

Mr. Okada: It started when it was introduced by the DX person in charge at our parent company, Nagoya Railroad. In fact, we had consulted with major vendors and several startups before that.
From the beginning, I had given up on solving the automation and efficiency of shift schedules with existing services. I was looking for something that would make me feel a new possibility that would overturn that common sense.

What potential did you feel in ALGO ARTIS?
Mr. Okada: First, the focus on the algorithm called "simulated annealing" was unprecedented and I thought it was interesting. And the personality of everyone. Even though it is a group of sharp people with excellent knowledge and skills in algorithms, when I talked to them, I didn't feel any unapproachability at all, and they treated us sincerely. There was a discussion within the company about "whether to do it with ALGO ARTIS or not," but since there was no one who had ever done shift schedule efficiency improvement in the first place, no one could say for sure that "it can definitely be done."
However, while having meetings with ALGO ARTIS, there was an atmosphere that made me think, "Shall I entrust it to these people?"

Mr. Ono, you have been involved since the project started. How have you been promoting the project?

Mr. Ohno:It wasn't the current system from the start. Initially, we tried a method where we passed data entered into Excel to an algorithm and checked the returned results. While adjusting constraints and settings in that state, I think we went through trial and error more than 1,300 times by my own count.
I frequently communicated by phone with Fujiyoshi (ALGO ARTIS algorithm engineer Junya Fujiyoshi) to refine our direction together.

I heard it didn't get on track immediately.
Mr. Ohno:Getting the system to take root in the field was not straightforward. To get them to master it, I once spent a month and a half visiting a certain office every day to explain it. Even then, there were times when we had to pause it temporarily.
So, I changed my approach: I would first find offices with an atmosphere of wanting to change, then visit them on both weekends and weekdays, gathering feedback from the field to refine the requirements. Sometimes I even had the ALGO ARTIS team come to the site to help organize opinions together.
I believe we are where we are today precisely because I kept saying, 'Eventually, the era of doing this by hand will be over,' and repeated that steady work.

Until the office that was the biggest resistance group masters AI better than anyone

I heard that one of the offices that showed the most resistance to using the system at the time was the Ichinomiya Office, where Mr. Fukushima and Mr. Horii were.

Mr. Fukushima:I was the one who resisted the most even within Ichinomiya (laughs). I wasn't used to it, and inputting data took time, but the quality of the output was still lacking, so I thought, 'This isn't even a reference' and 'Manual assembly is definitely faster and more reliable.' I even closed my laptop in the middle of the work.

Mr. Horii:At the time, we were groping in the dark regarding how to use it and how to input conditions. 'The same person's name is duplicated across multiple courses' or 'The name is entered even though it says Mr./Ms. XX cannot work any more than this'—that was the state of things, so it's understandable why that happened. Before we knew it, Ichinomiya was being called the most behind.

How did things change from there?

Mr. Fukushima:At one point, I steeled myself and decided, 'I will not do any manual assembly.' Once I decided to let the AI think up the draft, I had to start by inputting the conditions for all crew members. I spent the first two to three hours just on that. Even then, it didn't produce satisfactory results immediately.

However, I don't remember which month it was, but after continuing to do it, one day a moment came when I thought, 'It's actually giving me good answers.' From there, it suddenly became unstoppable.

Mr. Horii:Even though he said, 'Just for tomorrow is fine,' he tried to plan the whole week, and I had to tell him, 'Don't go that far' (laughs).

Mr. Fukushima:It's strange even to me. The reason I got so absorbed in it was also because I had discoveries like, 'There's a way to assemble it like this.' The AI sometimes comes up with new ways of assembling that I hadn't thought of myself. There was a view there that I would never have seen with manual assembly alone.

Time calculation verification work became zero, and the 'expressions' of the staff changed

Now that full-scale operation has started from March 2025, what changes have you seen in the field?

Mr. Fukushima:The most helpful thing is that the verification work for time calculations has completely disappeared. Previously, we spent a lot of time calculating and reviewing the possible start time for the next day for everyone while looking at what time the previous day's shift ended. After the operation started, the system does all the calculations based on laws and regulations, so that verification work is no longer necessary in the first place.

In my case, I felt that about 20% of the completed plan was the part I had manually adjusted. I make adjustments for the last less than 20% due to sudden absences or individual condition changes, but other than that, I can use what the AI assembled as is. There were many times when no corrections were needed at all.

Mr. Horii:What I feel has changed more than the numbers is the expressions of the shift managers. Before, they were always looking down with difficult expressions. Now, the time it takes to finish assembling is faster, and they have more time to talk to the crew members little by little. I think it's approaching the way it should be.

You mentioned that the burden of creating plans and the high hurdle of acquiring skills were issues. Seeing multiple offices, what changes do you feel?

Mr. Ohno:Those who have learned how to use it are using it very well. On the other hand, the reality is that there are still some who haven't fully learned it. After all, there are parts you can't understand without using it to a certain extent to master AI, and it's not easy to do trial and error repeatedly while handling normal duties. I do feel that there is an aspect where 'a certain amount of habituation and experience is required.'

Mr. Okada:The question of 'Is a system that chooses its users really a good system?' is one I have always held, not just this time. Precisely because AI is something you can master by gaining experience, I believe that continuing to improve both operations and the system so that anyone can use it is one of the future challenges.

Also, if the AI can try more autonomously to perform advanced tasks like 'course connection and course splitting' that are done when there is an extreme shortage of personnel, the range of use should expand further. The current system is not yet a finished product, but that is precisely why I believe it has further room for growth.

It seems that changes for the next generation are also emerging.

Mr. Horii:Recently, new staff members who don't know how to create rosters manually have started joining us. Since they rely solely on AI and have nothing to compare it to, they are actually able to utilize AI quite naturally. We are also seeing younger people who are interested in trying it out because it's an 'AI-based roster,' so I have high hopes for the coming generation.

“I hope this becomes a way to save the entire industry.”

Please tell us what you expect from ALGO ARTIS in the future.

Mr. Okada:The struggle of creating duty rosters is not unique to Meitetsu Bus. Bus companies all over Japan are facing the same problem. In fact, many companies have come to visit us after learning about our efforts, but the reality is that there are hurdles to implementation, mainly due to costs. I hope you will continue to advance the packaging of this solution so that it can become a way to save the entire industry. That is my honest feeling.

Finally, do you have a message for other transport operators facing the same challenges?

Mr. Ohno:I believe that those who survive are not the strongest or the smartest, but those who can adapt to their environment. I think the shortest path is for those with the strongest will to change to start moving first.

Mr. Okada:I understand the anxiety about costs and the fear of changing familiar ways of doing things. However, duty roster management is a task that will undoubtedly continue to arise every single day, and it is not something that can be stopped. There is definitely value in accumulating improvements, even if they are small.

Mr. Fukushima:I was the person who insisted more than anyone else that 'manual creation is faster.' But now, I use AI to an extent that I can't believe compared to when I first started. Just try it. I think that alone will change your perspective.

*The content of this article is current as of April 2026.

Related Articles: