AI-based Production Management
System in Retail Stores
About the Client
This article describes a real-life project. However, we cannot disclose our client’s name & the project details for privacy purposes.
Our client is one of the largest retail store solution providers in Japan, offering comprehensive technologies and services to optimize store operations. With thousands of retail outlets under management, the client continuously seeks advanced solutions to improve efficiency, reduce costs, and deliver a better shopping experience.
Project Overview
The project focuses on developing an AI-powered system that uses in-store photos to automatically analyse product displays and shelf arrangements. By replacing manual inspections with automated image recognition, the client aims to streamline daily store management and improve overall accuracy.
Industry
Retail
Technology
Computer Vision;
OCR; Generative AI; AWS
Country
Japan
Timeline
01/2024 - 06/2024
Challenges
The client faces several obstacles in managing product placement
and shelf monitoring across thousands of retail stores:
Labor-intensive process
Store staff currently spend hours manually checking shelves, capturing photos, and filling out daily reports. This repetitive work not only consumes significant time and resources but also distracts employees from customer-facing tasks.
Diverse product categories
With thousands of SKUs ranging from food and beverages to household goods, tracking placement and inventory consistency across all categories becomes increasingly complex. Ensuring compliance with display standards is especially difficult in large-scale operations.
Subtle product variations
Small differences in packaging color, design, or size between similar items often cause confusion during manual inspections. Even experienced staff members occasionally misidentify products, which leads to reporting errors and stock management inefficiencies.
High image volume
Each store generates hundreds of shelf images daily. Processing this massive amount of data manually slows down decision-making and delays the client’s ability to detect stock-outs, misplaced products, or compliance issues in real time.
Inconsistent reporting quality
Because the process depends heavily on human observation, data quality varies from store to store, making it challenging for managers to gain a reliable, consolidated view of operations.
Solution
RikkeiSoft develops an AI-based Production Management System that replaces manual shelf inspections with automated, AI-driven recognition and reporting. The solution is designed for scalability, accuracy, and real-time insights:
Computer Vision for Shelf Analysis
The system uses deep learning–based computer vision models to detect and segment every product on store shelves, even in cluttered or low-light conditions. It identifies product boundaries, counts units, and captures placement patterns. This ensures that all products—including those partially hidden or angled—are accurately recognized.
OCR for Price Tag Recognition
OCR technology extracts and digitizes text from price tags, allowing automated validation of pricing accuracy. It cross-checks against the central pricing database, highlighting inconsistencies and reducing human error. This capability ensures that customers always see correct prices, improving both compliance and customer trust.
Generative AI for Product Matching
A fine-tuned generative AI model interprets OCR and visual data to recognize product names and match them against a central database of enrolled items. This enables the system to handle subtle packaging or tint variations that would otherwise confuse traditional recognition systems. The AI continuously learns and improves as new products are introduced or packaging designs change.
Cloud-based Scalability with AWS
By leveraging AWS infrastructure, the system processes hundreds of thousands of shelf images securely and efficiently every month. Cloud deployment ensures flexibility to expand coverage from hundreds to thousands of stores without additional hardware investment, while maintaining compliance with data privacy and security standards.
Computer Vision Pre-processing
The solution generates real-time reports and visual dashboards that provide store managers and regional leaders with actionable insights. Alerts are automatically triggered when stock-outs, misplaced items, or pricing mismatches are detected, enabling faster response and better operational control.
Results
4,000+
stores deployed
90%
management cost reduction
300,000+
images analysed/month
The solution achieves remarkable improvements in operational efficiency and accuracy. Deployed in over 4,000 retail stores, the system now analyses more than 300,000 images per month, providing managers with real-time insights into shelf conditions. By automating inspections and reporting, the client reduces management costs by 90%, eliminates repetitive manual work, and significantly enhances store productivity. This AI-driven approach not only streamlines daily operations but also ensures more consistent product display, improving both efficiency and the customer shopping experience.