FamilyMart Reduces Weekly Workload by 6 Hours with AI Ordering—Optimizes Inventory and Curbs Food Waste

Table of Contents

1. Introduction: Convenience Store DX and "AI Ordering"

2. Background of Introduction: Caught Between Labor Shortage and Food Loss

3. What is AI Recommended Ordering?

4. Technical Elements of the System

5. Results of the 500 Store Pilot

6. Store Staff's Experience and Headquarters' Intentions

7. Assortment Optimization Brought by Similar Store Recommendations

8. Actual Reduction of Food Loss

9. Impact on Labor and Training Costs

10. Comparison with Other Chains/Overseas Cases

11. Handling Irregularities and the "Human Eye"

12. Mid-to-Long Term Roadmap and Challenges

13. Evaluation from ESG and SDGs Perspectives

14. Summary and Outlook

1. Introduction: Convenience Store DX and "AI Ordering"

Japanese convenience stores, which are deployed at an unparalleled density worldwide, face the dual challenges of chronic labor shortages and maintaining 24-hour operations. Following self-checkouts and mobile orders, "AI Ordering" is gaining attention as the next DX (Digital Transformation). After about a year of PoC (Proof of Concept), FamilyMart began introducing it to 500 stores nationwide in July 2025.ITmedia

Shift to Data-Driven Management

Traditional ordering heavily relied on the "experience and intuition of veteran staff," creating disparities between stores. The introduction of AI is the first step in quantifying subjective tasks and achieving highly reproducible operations.

2. Background of Introduction: Caught Between Labor Shortage and Food Loss

Acceleration of Labor Shortage

Due to the declining domestic labor force, convenience store chains are forced to consider closing during late-night hours or shortening business hours. Ordering tasks can occur up to four times a day, taking 15 to 30 minutes each time, which is a significant burden.Miscellaneous Blog, Sometimes Amazon

Food Loss and ESG Pressure

Japan's food loss amounts to 5.22 million tons annually (Ministry of Agriculture, Forestry and Fisheries, 2024). Convenience store disposal losses are often subject to social criticism and can affect corporate value. AI ordering is evaluated on its ability to "reduce unsold items" and "avoid opportunity losses."AMP - Business Inspiration Media

3. What is AI Recommended Ordering?

Multidimensional Data Learning

The system incorporates over 30 types of features, including sales performance, weather, holidays, surrounding events, and traffic volume, and relearns every 15 minutes at the shortest interval. It draws demand curves by day, delivery, and item, considering inventory and next delivery timing to return recommended numbers.FamilyMart

Similar Store Benchmarking

AI automatically extracts "model stores" with similar location conditions, customer demographics, and sales composition. By horizontally expanding bestsellers, it fills regional preference gaps while uncovering new demand.AMP - Business Inspiration Media

4. Technical Elements of the System

• Time Series Demand Forecasting Model: Combines XGBoost and LSTM to simultaneously capture short-term trends and seasonal factors.

• Multi-Agent Optimization: Maximizes overall profit by considering display space, expiration dates, and gross profit margins.

• Recurrent Feedback: Returns manually overwritten numbers to the learning data, allowing the model to self-update.

5. Results of the 500 Store Pilot

The table above shows average values (n=500 stores). The reduction in work allowed staff to reallocate time to customer service and product display, contributing to a 4% increase in sales.ITmediaAMP - Business Inspiration Media

6. Store Staff's Experience and Headquarters' Intentions

"The AI numbers have a basis, so there is a high level of acceptance. I can order with confidence."(Tokyo, Franchise Owner)
"Once accustomed, the task takes just 3 minutes. Overtime after closing has significantly decreased."(Osaka, Acting Manager)

Headquarters aims for quality standardization through standardization and efficiency in SV (Supervisor) operations. A supervising SV states, "The time spent on 'ordering guidance' during store visits can now be used for promotional proposals."Miscellaneous Blog, Sometimes Amazon

7. Assortment Optimization Brought by Similar Store Recommendations

AI learns from stores with similar customer demographics and suggests hit products not yet handled. Instead of forcibly eliminating "low-turnover goods," it optimizes both turnover rates and gross profit margins, maintaining each store's individuality.AMP - Business Inspiration Media

8. Actual Reduction of Food Loss

AI calculates the "sales floor volume maintenance number" to minimize inventory while maintaining visibility. This has halved disposal losses on average. Combined with secondary measures like freezing unsold items and donations, it contributes to SDGs Goal 12.3.FamilyMart

9. Impact on Labor and Training Costs

The most time-consuming part of training newcomers was mastering ordering logic. After AI introduction, the flow was shortened to "manual reading → AI recommended value confirmation → fine-tuning," reducing training costs by 30%. It is also linked with headquarters' e-learning, allowing quantitative evaluation of proficiency.

10. Comparison with Other Chains/Overseas Cases

• Seven-Eleven Japan: Testing deep learning-based automatic ordering since 2022.

• Lawson: Collaborating with retail AI startups, testing in 200 stores in Kansai.

• Walmart US: Reduced out-of-stock by 40% with AI inventory optimization (2024 results).
  Japanese convenience stores are characterized by a high number of SKUs and short expiration dates, making AI model difficulty high.

11. Handling Irregularities and the "Human Eye"

During unpredictable events like fireworks festivals or large typhoons, human judgment is indispensable. AI issues a "caution flag," leaving the final decision to the store, ensuring psychological safety on the ground.

12. Mid-to-Long Term Roadmap and Challenges

1. Fiscal Year 2026: Gradual rollout to approximately 16,000 stores nationwide

2. Fiscal Year 2027: Expansion to non-food categories (daily necessities, magazines)

3. Fiscal Year 2028: Overall supply chain optimization (center inventory AI)
   Challenges include absorbing "store diversity" and "model deviation," as well as strengthening cybersecurity.

13. Evaluation from ESG and SDGs Perspectives

AI ordering is a rare DX initiative that directly improves both the E (Environmental) and S (Social) aspects, potentially becoming an evaluation metric for ESG investors. It is watched as a "carbon reduction" measure alongside plastic reduction and increased renewable energy ratios.

14. Summary and Outlook

FamilyMart's AI ordering has shown the potential to simultaneously solve the triple challenges of "labor shortage, lost sales opportunities, and food loss." In the near future, it is likely to pave the way for a "fully AI store" including register, shelf allocation, and price optimization.

List of Reference Articles