John Ward, Fractional CTO at Resolutis, authored Chapter 5: Retail Supply Chain: Compute Meets Reality
The Linux Foundation’s State of Edge report is the closest thing the edge computing industry has to a consensus view of where the technology actually stands. It draws on contributions from practitioners and engineers across the global ecosystem, not vendor roadmaps, but ground-level experience of what works in production.
Resolutis contributed Chapter 5 to the 2026 edition. The brief was to address retail supply chain, one of the most demanding environments for edge AI deployment, from the perspective of someone who has spent 25 years engineering at the boundary between hardware and software, from silicon selection through to production inference.
What the chapter covers
The title is deliberate. There is a significant gap between how edge AI in retail supply chain is discussed, in analyst decks, at trade shows, in press releases — and what the engineering reality looks like when you try to deploy compute at scale in a live retail environment.
The chapter addresses that gap directly, working through the practical implementation of Linux-based edge infrastructure in fast-moving consumer goods (FMCG) warehouse and retail environments, where real-time decision-making, continuous uptime, and integration with decades-old legacy systems are not optional features, they are survival requirements.
The core argument is straightforward: the cloud-centric model, which dominated the first wave of retail AI, is reaching its limits. Processing images of supermarket shelves via a round-trip to a cloud API is costly, slow, and brittle. Moving that inference to the edge, to a camera, a handheld scanner, a local Linux server, changes the economics and the performance characteristics entirely.
The specific problems edge compute solves
On-shelf availability (OSA) is one of the most important operational metrics in retail. The industry average of 93–95% OSA translates to an estimated $100 billion in lost US retail sales annually. An OSA increase of just 1% lifts sales by 2–4%. The challenge is that traditional OSA measurement, store associates walking the floor every morning, captures a single point-in-time snapshot that bears little relationship to conditions throughout a trading day. Edge vision systems change this: continuous shelf monitoring, processing images locally, detecting stockouts in real time rather than through periodic manual audits.
Legacy WMS integration is the unglamorous reality that most edge AI deployments have to navigate. The majority of warehouse management systems were built over a decade ago in C and C++, are largely proprietary, and create substantial architectural constraints. The chapter addresses how containerised, modular architectures built on open-source Linux foundations — including LF CIP for long-term kernel support, allow edge AI to be layered onto existing infrastructure without wholesale replacement.
The economics of edge vs cloud inference is quantified directly. Edge AI at the shelf can reduce overall cloud costs by 40% to 90%, driven primarily by eliminating data transfer fees and reducing cloud compute time. For a retailer running vision inference across thousands of stores, the difference is not marginal.
The last-mile task generation problem — turning raw shelf intelligence into prioritised, actionable work for store associates — is addressed through integration with handheld devices (Zebra, Honeywell), planogram compliance systems, and computer-generated ordering workflows. The chapter covers how edge processing on the devices themselves enables offline operation, which is critical in stores where connectivity is inconsistent across large footprints.
Why this matters beyond retail
The patterns in this chapter are not unique to retail. The same architecture, distributed edge compute, local inference, intermittent connectivity, integration with legacy enterprise systems, applies across manufacturing, logistics, healthcare, and any environment where data needs to be processed at the point of origin rather than shipped to a cloud and back.
These are exactly the kinds of deployments Resolutis works on: taking the engineering depth required to make edge AI work in production, and applying it to the environments where it matters most, constrained hardware, regulated data, operational requirements that no proof-of-concept ever accounts for.
Read the report
The 2026 LF Edge State of Edge Report is available now. Chapter 5, Retail Supply Chain: Compute Meets Reality, begins on page 80.