AI computer vision powers out-of-stock processes for grocery chain leader
Developing and deploying an automated solution using Computer Vision to detect out-of-stock products, resulting in improved shelf availability and operational efficiency.
The challenge
The traditional, manual approach to inventory management and shelf auditing presented several critical challenges for our client, one of Australia’s largest grocery retailers.
- Manual out-of-stock checks were infrequent, resulting in delayed replenishment and missed sales opportunities.
- Store personnel were tasked with repetitive shelf audits, which diverted their time away from customer service and other high-value tasks.
- There was a limited ability to verify if products were placed correctly on shelves, resulting in an inconsistent shopper experience.
The solution
Mantel was engaged to help our client create a solution, centred on a scalable, automated system using Computer Vision and AI to monitor shelves continuously
- Automated detection: A Compound AI system leveraging multiple computer vision models was used to accurately detect gaps, products, shelf tags, and out-of-stock products from camera images captured in-store.
- Production-ready scale: The solution is a scalable, open, extensible, and continually improving Inventory Management system. It is designed to increase shelf availability and improve productivity across 1000+ stores.
- Store integration: The system integrates with existing in-store cameras and the store monitoring system to deliver hourly out-of-stock alerts.
The outcome
The project’s success was attributed to a strategic focus on continuous improvement, collaboration, and iterative development:
- Continuous improvement: The project demonstrated the importance of continuous improvement when tackling complex AI challenges, as it helps organisations understand the ongoing value delivered.
- Collaboration: Collaboration was key. The team partnered with the retailer, Google Cloud, and the hardware provider to develop an innovative solution that overcame existing constraints.
- Iteration: Iteration led to impact. The team explored multiple approaches and shifted direction along the way to reach the desired level of accuracy and deliver a reliable solution.
