FamilyMart is trialing an unmanned overnight store—can AI really replace the convenience-store clerk’s “eighteen kinds of martial arts”?

FamilyMart is testing unmanned overnight operations at some of its directly managed stores, using self-checkout, self-service heating, and electronic payments to get through the period of the most severe labor shortage in the early-morning hours. Will AI replace more retail labor in the future?
(Background: Japan announced an investment of 1 trillion yen—by 2040, deploying 10 million AI robots across 18 industries to address labor shortages)
(Additional background: Microsoft is putting $2.5 billion into establishing a “Frontier Company,” sending 6,000 engineers to customers’ offices to make AI truly take root)

Table of Contents

Toggle

• Next step—will robots start restocking?
• Even discounts could be calculated by AI
• What really gets stuck has never been technology

Unable to find staff to work the overnight shift, FamilyMart has recently begun posting notices at some of its directly managed stores: from 12:00 a.m. to 5:59 a.m., operations will be carried out without on-site personnel, relying on self-checkout, self-service heating, and electronic payments to get through the entire night shift.

Hong Yaling, secretary-general of the Taiwan Chain Stores & Franchise Association, said bluntly that the labor shortage problem in the chain retail industry is “very severe” over the past two or three years. The monthly wage for the overnight shift—already including labor insurance and health insurance—has reached over 50,000-plus. Even so, franchisees still have to call employees in the middle of the night to get them to come to work. They just hope the workers won’t steal or slack off; whether they can be kept on the job is another question.

Next step—will robots start restocking?

Taiwan’s current self-checkout test is actually fairly cautious. In Japan, Lawson has already gone further.

In June 2025, Lawson opened its first “Real×Tech LAWSON” experimental store in Minato Ward, Tokyo: shelves are equipped with AI cameras that can detect when customers stand in front and appear to be “struggling to choose,” then proactively pop up recommendations. In the back area, robots handle tasks such as frying chicken, preparing food, and restocking drinks. Lawson president Teinobu Takezono spoke very directly: labor shortages are the “biggest, and most important” problem, and the goal is to cut store staff workload by 30% before fiscal year 2030.

The same kind of scene is something Taiwan is practicing, too. At COMPUTEX 2026, Odin, the humanoid robot demonstrated by Yungyuan Intelligent, uses computer-vision recognition to identify product appearances. When products are sold out, it automatically reports back to the system for restocking, and it can even go look on other shelves for the same item. These are still in the demonstration and experimental stage, but the future direction is already clear: the next attempt to replace robots will be “repetitive and physically draining” work such as restocking, inventory counting, and putting items on shelves.

Even discounts could be calculated by AI

For now, the discounts on near-term items in Taiwan’s convenience stores are essentially “rule-based.” For example, FamilyMart’s “Friendly Food Time” gives a fixed 30% off after 5 p.m.; 7-ELEVEN’s i Premium Food is discounted 65% to 20% off in three time periods; and OK Convenience Store’s 60% off “Save the Food Time” follows the same logic.

People set the rules first, and the system then carries them out mechanically. But the U.S. retail giant Target has already shown another path: after introducing AI dynamic pricing, its return on investment surged by several times. In simple terms, AI doesn’t apply discounts from a schedule—it looks at inventory levels, weather, foot traffic, and shelf life in real time, then calculates how much the item should cost “at this moment.”

If this logic were applied to Taiwan’s convenience stores, the same batch of oden might be discounted to 40% to clear stock on a rainy night with few customers, but only reduced to 10% on a sunny day with heavy foot traffic. In the future, discount tables might shift from “X% off at Y time” to “recalculated every second.”

What really gets stuck has never been technology

That said, so far, robots and AI have not truly taken over an entire store. When they get stuck, it’s usually not because the technology can’t do it, but because of those “edge situations”: customers who are drunk and cause trouble, receipt machines jamming paper, elderly people who can’t operate self-checkout, and late-night security disputes.

Taiwan’s convenience stores are currently positioned on the more conservative end of the spectrum: they use self-service devices first to address the urgent issue of not being able to find overnight staff. Japan’s approach, meanwhile, is to actively teach AI to run a store. The distance between the two may be the real battleground for the retail industry over the next decade.