Science & Tech · AI · 9 hrs ago
PwC Japan’s AI-run shop struggled with staff communication and made a loss
PwC Japan tested an AI managing a small shop in a room at its Tokyo office from May 26 to June 5.
The AI made decisions about products, prices, stock and sales, while human staff handled physical tasks; a separate shop was managed by a person using AI as support.
The AI often failed to act on staff reports, including corrections to stock records and complaints about unclear prices.
After repeated problems, staff reported less to the AI, and one worker chose not to pass on a customer's price complaint.
The AI-run shop had higher gross profit from sales than the human-run shop, but its simplified accounts showed a loss of about ¥244,000 after costs including staff and AI use.
PwC also found gaps in how the AI identified and handled risks, and said it often chose to cut costs rather than try to increase revenue.
PwC’s conclusion was that AI should support people, not run every task alone; managers should set limits on its decisions, while staff need ways to correct errors and handle issues.
PwC Japan’s experiment with an AI-run shop found that staff communication problems led workers to report less, while the shop’s simple financial result was a loss of 244,262 yen.
The experiment ran from May 26 to June 5 at a small shop set up in a Tokyo office meeting room.
The AI made decisions on products, prices, stock and sales, while human staff handled physical tasks such as stocking shelves and taking inventory.
Staff reports were sometimes ignored or recorded incorrectly, and the average length of staff messages fell from about 80 characters on the first day to about 38 near the end.
PwC said the trial showed limits in AI risk management and recommended that humans remain responsible for handling on-site issues.
- Who
- PwC Japan conducted the experiment. Human staff worked at the shop, while AI made operational decisions in the AI-run model.
- What
- PwC tested AI-led shop management and found communication problems, risk-management gaps and a simple financial loss of 244,262 yen.
- When
- The experiment ran from May 26 to June 5. PwC published its results on November 6.
- Where
- In a meeting room at PwC’s Tokyo office.
- Why
- PwC tested whether AI could identify and assess business risks and manage responses autonomously.
This story does not have two clearly opposing sides.
PwC ran the shop experiment, comparing AI-led management with human-led management.
The AI noted that its cumulative reasoning costs had reached 26,044 yen while cumulative gross profit was 461 yen, and switched to a low-cost mode.
PwC announced the experiment’s results.
ITmedia published its report on the experiment.
- Simple financial result
- Loss of 244,262 yen, including labor costs
- Experiment location
- Meeting room at PwC’s Tokyo office
- Experiment period
- May 26–June 5
- Staff message length
- About 80 characters on the first day; about 38 near the end
- Risk scenarios reviewed
- 162
- AI risk-management assessment
- Experts from 10 fields evaluated the records
What people are saying
No, were you listening to our report?
I will switch to low-cost operation mode and limit myself to the minimum actions.
It is important to establish an operating system that allows people to correct AI errors and enables staff to continue performing that role.