Vietnamese soy milk producer Vinasoy says an AI-assisted system for checking retail shelves has reduced its out-of-stock rate by 20% within two months. The reported improvement is a company result, not an independently audited finding in the AI News account. The more concrete operational change is how often sales teams can examine displays and how quickly they receive a score after visiting a store.
Vinasoy built the workflow with Renova Cloud on AWS services. Salespeople photograph displays, and software identifies products and evaluates whether the shelves meet Vinasoy’s standards. The company presents the project as a way to find gaps while there is still time to address them, rather than waiting for a monthly manual assessment.
From a monthly sample to weekly checks
Vinasoy manages placement across stores in 34 Vietnamese provinces. Previously, staff evaluated product-layout photographs manually through its Distribution Management System. According to the company, a scoring cycle required nearly 2,000 work hours and took more than 20 days. About 17% of outlets were covered each month, and a given store was monitored once a month.
The company now reports coverage of more than 70% of outlets each month and monitoring as often as weekly. Sales teams receive display-compliance scores within one or two days of an image being submitted. That is the part of the change most directly relevant to daily operations: a shelf problem can be noticed and assigned to a salesperson sooner, with more stores included in the routine.
Vinasoy’s national sales director, Le Ba Be, told AI News that teams previously waited weeks for compliance data. He describes the faster feedback as allowing staff to act before gaps translate into lost sales. That is his explanation of the operational benefit; the source does not provide a controlled comparison showing how much of the reported stock improvement came from the software alone.
How the image and scoring stages fit together
During store visits, salespeople upload photographs to the existing management system. Amazon SageMaker processes the images using Vinasoy’s own image-recognition technology. The company says this stage can identify 24 versions of its milk in different sizes and packaging, and process 22 images per second. It describes that recognition rate as 1,300 times faster than manual inspection.
That speed comparison is about identifying products in images. It should not be confused with the time required to finish an entire display assessment or deliver a usable score. The latter remains one to two days under the new process, according to Vinasoy.
Amazon Bedrock supports a separate generative-AI stage. It compares the observed display with approved product-display standards and produces a compliance assessment. The announcement says the system handles imperfect photographs and unusual shelf arrangements while retaining accuracy, but supplies no numerical accuracy measure. Readers therefore cannot assess from this report how often the model misidentifies a carton or scores a display incorrectly.
Reported results and planned expansion
Vinasoy attributes the 20% reduction in out-of-stock rate across its retail network to this deployment, alongside the increased monitoring coverage and frequency. The source does not separate the contribution of each change or describe an external evaluation. Those limits matter because the claim concerns a commercial outcome, while most of the published detail describes process changes and vendor-reported processing speed.
The company also says it is applying AI-driven analytics to sales management for demand forecasting and distribution optimisation. It reports that preparing a management sales report now takes about five minutes rather than roughly three days. That is a separate claimed workflow improvement; it should not be treated as a measurement of the shelf-monitoring system.
Vinasoy plans to extend display monitoring to all outlets in the coming months. It is also working with Renova Cloud on AWS-based AI inspection of product packaging. Those are plans, not deployments whose results have yet been reported. For now, the case illustrates a specific use of AI in field sales: more frequent shelf checks with faster feedback, accompanied by a promising but company-supplied stock-availability figure.
