London-based Edgify has raised $9 million to expand edge-AI software that runs computer-vision models inside physical stores, where the company is trying to add automation to existing cameras, self-checkouts, scales, and point-of-sale equipment rather than replace the surrounding retail infrastructure.

The Series A+ round was backed by Rank Ventures and Mangrove Capital Partners and takes total funding to $25 million. Edgify is already deployed with grocery retailers in Europe and the US, while the latest capital is intended to support further growth in retail and other industries operating large fleets of physical devices.

Its architecture moves part of AI inference away from central cloud services and onto hardware inside the store. That allows systems to identify produce, monitor checkout activity, or detect potential loss events closer to the transaction, while reducing the need to stream every piece of video or sensor data into remote infrastructure.

The approach changes part of the commercial calculation around computer vision because retailers already own large estates of cameras and checkout equipment. A software layer capable of working with those devices offers another route to automation without turning every supermarket refurbishment into a bespoke AI hardware project.

Computer vision moves closer to the transaction

Edgify’s product material describes models running on in-store devices and working with standard cameras to recognise products and analyse checkout behaviour. At a self-service till, the system can compare what appears in front of the camera with what the customer scans or selects, creating another signal for errors or potential loss.

That is harder in practice than the demonstration suggests because not every mismatch represents theft. Barcodes fail, loose produce can be visually similar, customers make mistakes, staff override transactions for legitimate reasons, and crowded checkout environments produce far less controlled imagery than a computer-vision benchmark.

A useful loss-prevention system therefore needs to minimise false alerts as carefully as it detects genuine problems. Excessive interventions slow the checkout, consume employee time, irritate customers, and eventually train staff to ignore warnings, shifting cost from inventory loss into store operations rather than removing it.

Local inference offers another operational advantage where live video is involved because latency and bandwidth can influence whether an alert is useful. Keeping more processing inside the store may also reduce the amount of raw imagery moving into central cloud environments, although retailers still need governance around storage, model updates, security, access, and the information retained after an event.

Retail margins create a harder adoption test

Grocery retailers have little reason to buy AI simply because computer vision has improved. Labour, energy, logistics, store refurbishment, and technology investment compete for capital in a sector where margins remain tight, so automation has to demonstrate that it protects revenue, reduces operating cost, or removes work employees would otherwise perform manually.

Loss prevention offers a relatively measurable use case because retailers can compare transaction and inventory data before and after deployment. Even then, the useful measures extend beyond detected incidents towards actual reductions in shrinkage, employee intervention time, abandoned transactions, customer friction, and the infrastructure cost required to operate the system.

Edgify’s model differs from the autonomous-store concept that drew substantial investment earlier in the computer-vision cycle. Rather than rebuilding the shopping journey around a dense new sensor estate, it is attempting to improve equipment already sitting in ordinary supermarkets, which greatly expands the potential addressable market but exposes the software to years of hardware variation and inconsistent store design.

Those estates also create a management problem at scale. Models have to be updated and monitored across hundreds or thousands of locations, while local failures, software versions, camera positions, and hardware differences can affect accuracy. Edge computing moves inference closer to the device but does not remove the need for central lifecycle management.

Edgify is looking beyond grocery towards quick-service restaurants, distribution centres, and other physical industries, where similar combinations of cameras, sensors, and local compute already exist. Expansion into those markets would test whether the company has built a general edge-AI infrastructure layer or a product whose strongest economics remain tied to supermarket loss prevention.

For now, retail offers a demanding proving ground because the software sits directly inside an existing transaction rather than beside it. The $9 million round gives Edgify more capital to expand, although the eventual adoption case will be decided in store-level operating data — fewer losses, fewer unnecessary interventions, and enough savings to justify adding another managed system to an already crowded retail technology estate.