Retailers and logistics firms use Azure Vision to read labels, check shelves, detect damaged parcels and track goods across the supply chain. In 2026 the bigger story is change: Microsoft has scheduled retirement of the Image Analysis API and Azure Custom Vision for September 2028, so new projects should be built on the recommended replacements and an architecture that can run at the edge.
Updated October 2026
Who this is for: Heads of supply chain, warehouse and store operations, retail and logistics CIOs, and Azure architects responsible for vision workloads.
The business problem: invisible errors in the supply chain
Most supply chain losses start as something a camera could have seen: a mislabelled carton, a crushed pallet, an empty shelf, a wrong item in a pick bin, a trailer loaded out of sequence. Found late, each one becomes a chargeback, a return, a lost sale or an unhappy customer. Many retailers and 3PLs already have cameras in place; what they lack is a reliable, governed way to turn video into decisions, on a platform that will still be supported in three years.
What changed in 2026
- The service is now called Azure Vision in Foundry Tools. Microsoft's overview states that Image Analysis 4.0 is deprecated and will be retired on 25 September 2028, after which calls will fail.
- The migration guide covers Image Analysis v3.2 and v4.0, and asks customers to plan their transition by 25 September 2026.
- Azure Custom Vision is also being retired, with full support for existing customers until 25 September 2028.
- Spatial Analysis, once promoted for store and warehouse movement analytics, has been retired; its documentation now points to Azure Content Understanding.
Azure vision services: status and replacements
| Capability | 2026 status | Microsoft-recommended direction |
|---|---|---|
| Image Analysis 4.0 / 3.2 (tags, captions, objects) | Deprecated; retires 25 Sep 2028 | GPT models in Microsoft Foundry or Azure Content Understanding |
| OCR / Read for labels and documents | Image Analysis OCR retiring | Azure Document Intelligence Read model |
| Image embeddings and visual search | Image Analysis embeddings retiring | Multimodal embedding models in Foundry (e.g. Cohere Embed, SigLIP) |
| Custom Vision (custom classification and detection) | Retiring; supported to 25 Sep 2028 | Azure ML AutoML, Foundry models or Content Understanding |
| Face | Current | Face API |
| Spatial Analysis | Retired | Content Understanding (video) or custom edge models |
Source: Microsoft Learn pages linked above. Check them again before you commit, as dates can change.
High-value use cases in retail and logistics
Retail
- On-shelf availability: detect gaps and planogram errors from fixed or robot cameras.
- Self-checkout and loss prevention: recognise items and flag mismatches, as in autonomous store technology.
- Returns inspection: grade product condition from photos.
Logistics and warehousing
- Label and document reading: read shipping labels, bills of lading and container IDs.
- Damage detection: photograph parcels at receiving and dispatch to settle claims.
- Pick and pack verification: confirm the right item and quantity before sealing.
- Yard and dock monitoring: track trailers, dwell time and dock safety.
See our wider guide to computer vision in the supply chain.
Reference architecture for vision in Azure supply chain management
docks, aisles, conveyors, shelves
detection, OCR, damage models on site
Document Intelligence, Foundry models, Content Understanding
WMS, ERP, TMS, POS, dashboards
Figure: Vision AI for retail and logistics on Azure. Cameras feed edge models for real-time decisions; cloud services handle documents and complex cases; results update WMS, ERP and POS; a control plane manages models across sites.
Why the edge layer matters: warehouse and store decisions need sub-second answers, video is costly to upload, and owning your models protects you from future API retirements. See real-time video analytics.
Running Image Analysis or Custom Vision today? Get a free Azure vision migration assessment with a replacement map and timeline.
How to start in 30 days
- Week 1: Inventory every call to Image Analysis, Custom Vision and Spatial Analysis; map each to a replacement.
- Week 2: Pick one high-cost error (damage claims, mispicks or shelf gaps) and collect 1,000+ labelled images.
- Week 3: Train or configure the replacement model; compare accuracy with the current service.
- Week 4: Run at one site in shadow mode, connect results to WMS or POS, and measure the error rate change.
Where it's heading: 2020 → 2030
| Period | Vision in retail and logistics |
|---|---|
| 2020 | Task-specific APIs (Computer Vision, Custom Vision, Spatial Analysis previews). |
| 2026 | Shift to generative and multimodal models; legacy vision APIs on retirement paths; edge inference grows. |
| 2028 (scheduled) | Image Analysis API retires and Custom Vision full support ends on 25 September 2028. |
| 2030 (our view) | Vision agents act on events automatically, such as rerouting damaged goods or reordering stock, within set policies. |
How NexaStack helps
NexaStack Vision AI offers real-time video intelligence, precision detection, autonomous visual agents and continuous learning pipelines, with edge-to-cloud flexibility. Use it alongside Azure to run your own models in stores and warehouses, keep video on-site, and avoid being tied to one API's lifecycle. Explore the Vision AI solution and our Azure computer vision guide.
FAQ
Is Azure Computer Vision being retired?
The Image Analysis API (v3.2 and v4.0) is deprecated and retires on 25 September 2028. Face and other Foundry Tools continue.
What is replacing Azure Custom Vision?
Microsoft points to Azure ML AutoML, Foundry model catalog models and Azure Content Understanding.
What are the latest Azure vision trends for retail and logistics in 2026?
Generative and multimodal models replacing fixed APIs, document AI for labels, edge inference in stores and warehouses, and vision agents tied to WMS and POS.
Is migration costly, and what about shopper and worker privacy?
Migrate one use case at a time and reuse your labelled data. Processing video at the edge and sending only events limits personal data exposure and cloud cost.
Want one warehouse or store running retirement-proof vision AI in 30 days? Book a retail and logistics vision pilot with NexaStack.
