Read the story in Japanese
TOKYO, Japan – On a small percentage of flights, despite everyone’s best intentions, something unplanned happens. A passenger gets sick or a flight has a long delay.
After the cabin attendant attends to these kinds of situations, the senior cabin attendant writes up a report so ground staff can follow up – such as making sure there’s a wheelchair waiting at the gate or re-arranging onward travel.
Logging a single case, depending on complexity, can take an hour or more, taking time away from other inflight duties. Now Japan Airlines (JAL) is developing an AI app that can generate these handover reports by cabin attendants simply typing in a few keywords and phrases and checking a series of boxes – which can all be done while in the sky, even when connectivity isn’t dependable.
“The JAL-AI Report makes our cabin attendants’ jobs more productive,” said Keisuke Suzuki, a senior vice president of JAL’s Digital Technology Department. “They can spend more time on customer service instead of doing administrative work.”
Keisuke Suzuki, a senior vice president in charge of JAL’s Digital Technology Department, said the JAL-AI Report will generate reports faster, giving cabin attendants more time with passengers. Photo by Noriko Hayashi for Microsoft.
The JAL-AI Report is being developed using Microsoft’s Phi-4 small language model or SLM, which requires less computing power than the large language models or LLMs most generative AI tools run on, so it can be used offline on a device for specific tasks.
Cabin attendants who have tried it say it can slash the time for writing operation reports by up to two thirds, say, from one hour to 20 minutes, or from 30 minutes to 10 for simpler cases.
Click here to load media
The app also translates the reports with one tap from Japanese to English, a task needed on international flights.
Making an AI app work offline
Japan’s flagship carrier operates a fleet of 227 planes flying worldwide and serves 66 countries and regions including code sharing. Last year, it ranked sixth amongst the world’s best airlines for customer satisfaction according to Skytrax. Its current group chief executive Mitsuko Tottori is the first woman to lead the airline, having risen up the ranks from cabin attendant.
The JAL-AI Report is being developed with the help of Microsoft’s Azure AI Foundry and using Microsoft’s Phi-4 SLM.
While LLMs are good for tackling complex tasks that need advanced reasoning and analysis, SLMs can handle simpler tasks and run locally on a device rather than the cloud.
Japan Airlines is building an AI app, the JAL-AI Report, for cabin attendants to report situations on board for ground staff to act on. Photo by Noriko Hayashi for Microsoft.
They can also be fine-tuned with less data. The JAL-AI Report is fine-tuned on 100 previous reports, said Ryuto Ikeuchi, an AI engineer from Headwaters, which together with Fujitsu, are the system integrators for the project.
The goal is for these apps to be usable in environments with weak Wi-Fi, such as the outdoor ramp and inflight.
Manabu Yamawaki is leading JAL’s generative AI charge. Photo by Noriko Hayashi for Microsoft.
JAL chose the Phi-4 small language model because “even though there is Wi-Fi inflight, some areas have bad connection,” said Manabu Yamawaki, manager of security planning in the System Management Department of JAL, who is leading the airline’s generative AI charge.
Tech in flight
Takako Ukai joined JAL as a cabin attendant 35 years ago. She says she likes meeting and talking to people from around the world.
The business has changed over the years. With the advent of low-cost carriers, passengers now expect more from full-service carriers like JAL, she said, adding: “The challenge is how to serve better than expectations.”
“In the past (if something happened), we verbally conveyed information about connecting passengers to the crew of the next flight. Now we have to report it on a tablet, as an official report so (the cabin attendant on) the next flight knows and ground staff knows.”
Ukai is currently a member of the airline’s employee experience team, providing a cabin attendant’s point of view to JAL’s digital transformation team.
Right now, when there is an episode, the senior cabin attendant fills out a template on a tablet. This includes a section of free text where they type in chronological order what happened. To do this, the senior crew member may have to interview the cabin attendant involved and maybe also the passenger involved.
Takako Ukai, cabin attendant, at JAL’s training facility at Haneda Airport. Photo by Noriko Hayashi for Microsoft.
“There are frequent interruptions to serve passengers, so you can’t do it all at once. Sometimes, you have to rework many times,” Ukai said.
The JAL-AI Report app speeds it up by taking cabin attendants through a series of checkboxes – whether the category is medical, flight delay, etc. – and also what the situation was – stomachache, fever, maintenance and so on. The cabin attendant then types a series of keywords or phrases in bullet points briefly noting what happened, for example – “Fever.” “Seat 3H.” “Moved seat and lay down.” “Requesting to go clinic.”
The AI might have questions – such as whether a doctor was called, or whether the captain or ground staff were told, to prevent omissions in reporting.
Once that’s done, the cabin attendant taps on a button to generate the complete report. Another button translates from Japanese to English if needed.
The app can cut the time for reporting down to about 20 mins from an hour, Ukai said.
Better quality reports
Of the 1,000 flights JAL flies each day, a small percentage involves such report creation when an event requiring a handover occurs, said Yamawaki. These reports get sent to relevant departments – from security to customer service and other kinds of ground staff.
Yamawaki’s remit has grown over the years as the use of electronics has grown on board – from software security to inflight entertainment to Wi-Fi and now generative AI.
He thinks that in addition to saving time, the JAL-AI Report could help improve quality, as some cabin attendants currently write in more detail than necessary.
Chief cabin attendants Maya Tanaka and Takako Ukai at Haneda Airport in Tokyo. Noriko Hayashi for Microsoft.
Once the proof-of-concept period is over at the end of March, he said, the challenge is making sure the system works well offline.
In the future, he would like the JAL-AI Report to be able to receive verbal accounts from those involved – cabin attendants or passengers, transcribe and summarize the information and generate a report.
“Improved voice handling is high priority,” he said.
The JAL-AI Report app is part of a wider roll-out of generative AI across JAL that began in mid-2023. All of the group’s 36,500 employees now have access to AI tools grouped under JAL-AI Home on the Microsoft Azure OpenAI platform for administrative tasks like drafting emails, summarizing and translating documents and more.
JAL sees opportunities to “put generative AI at the center of the business and bring changes in operations and customer service,” Suzuki said. “We are excited to have the AI and humans work together.”
Top Image: Chief cabin attendant Maya Tanaka tests out the JAL-AI Report at Japan Airlines’ training facility in Haneda Airport. Photo by Noriko Hayashi for Microsoft.
GIPHY App Key not set. Please check settings