Industrial IoT platform rollout for a global manufacturer

Context and constraints

The manufacturer operated plants across multiple countries, and each site had, over time, built up its own approach to capturing machine telemetry. Some sites had sophisticated local historians; others relied on spreadsheets and manual logs. The data existed, but it was siloed by site, which meant that predictive maintenance was limited to whatever a given plant could do on its own, and enterprise-level performance analysis was effectively impossible. Leadership could not reliably compare line performance across sites, identify common failure modes, or share an improvement made in one plant with the rest of the network.

The constraints were considerable and very physical. Industrial environments are demanding: connectivity is sometimes intermittent, equipment spans many vendors and generations, and operational technology networks are rightly protected from the corporate world by strict segmentation. Any platform had to respect those boundaries, avoid interfering with production under any circumstances, and cope gracefully with sites that might lose their link to the cloud for periods of time. There was also a strong, sensible requirement that the rollout should not depend on rebuilding every site's instrumentation at once.

The approach in depth

We designed an industrial IoT platform around a clear edge-to-cloud pattern. At each site, an edge layer collected telemetry from the local machinery, normalised it into a common model, buffered it when connectivity was lost, and forwarded it to a central cloud platform when the link was healthy. This edge-first design meant that local operations never depended on the cloud being reachable, while the enterprise still gained a consolidated view once data flowed upward.

Standardisation was the heart of the approach. We agreed a common information model for assets, signals and events so that a temperature reading or a fault code meant the same thing regardless of which site or vendor it came from. This was painstaking work, because the same physical concept was often represented differently across the estate, but it was precisely this normalisation that unlocked cross-site comparison and enterprise analytics. Without it, the central platform would simply have been a bigger silo.

On top of the consolidated data we built analytics and predictive maintenance capabilities that could now draw on the whole network's experience rather than a single plant's. A failure pattern observed at one site could inform early-warning models applied everywhere, and performance benchmarking across lines became straightforward. We were careful, though, to keep the most safety-relevant logic close to the equipment, treating the cloud as the place for fleet-wide insight rather than for anything that production depended upon in real time.

Delivery phases and sequencing

We piloted the platform at a small number of representative sites first, choosing plants that differed enough to stress-test the common information model and the edge connectivity approach. Proving the pattern in genuinely different conditions early gave us confidence that it would generalise, and it surfaced the awkward integrations, the unusual protocols and the older equipment, while the stakes were still low.

With the pattern proven, we rolled out site by site, treating the edge deployment as a repeatable package that could be installed with minimal bespoke effort at each new plant. Sequencing the rollout this way meant the manufacturer began seeing cross-site value as soon as the first cluster of plants was connected, and the programme could absorb the inevitable surprises of individual sites without holding up the whole estate. Each site's onboarding concluded with validation that its data was flowing correctly and mapping cleanly onto the common model before it was considered complete.

Architecture and technology decisions and trade-offs

We chose an edge-first architecture precisely because of the realities of industrial connectivity. Buffering and local processing at the edge traded some additional on-site complexity for resilience and independence from the network, which for a manufacturing environment is the right exchange: production must never stall because a cloud connection dropped. We respected the separation between operational and corporate networks rigorously, treating the edge layer as a controlled, one-directional bridge that exported telemetry without exposing the operational environment to outside risk.

In the cloud we leaned on managed services for ingestion, storage and analytics so that the manufacturer's teams were not burdened with running infrastructure. A genuine trade-off arose around how much processing to push to the edge versus the cloud. More edge processing meant lower bandwidth and faster local response but greater complexity to manage across many sites; more cloud processing simplified the edge but demanded reliable connectivity. We settled on a deliberate split: time-sensitive and connectivity-independent logic at the edge, and heavy, fleet-wide analytics in the cloud. We also standardised aggressively on the common information model even though it slowed the early sites, judging correctly that the long-term analytics payoff justified the upfront discipline.

Measurable outcomes

The platform turned a collection of isolated plants into a connected fleet. For the first time, leadership could compare performance across sites on a like-for-like basis, identify the best and worst performing lines, and propagate improvements across the network rather than reinventing them locally. We typically see this enterprise visibility surface improvement opportunities that simply were not knowable when each site operated in isolation.

Predictive maintenance improved markedly because models could now learn from the whole estate's experience rather than a single site's limited history. Failure modes spotted at one plant became early warnings everywhere, which we typically find reduces unplanned downtime and lets maintenance be scheduled around production rather than imposed upon it. The repeatable edge deployment also meant that connecting a new site became a routine exercise rather than a bespoke project, lowering the cost of extending the platform as the business grew.

• Edge-first design that buffers telemetry locally so production never depends on cloud connectivity.
• Common information model normalising signals across vendors and sites to enable true cross-fleet comparison.
• Strict OT and IT separation with the edge acting as a controlled, outbound-only bridge.
• Fleet-wide predictive maintenance learning from the whole estate rather than a single plant.
• Repeatable site onboarding packaged so new plants connect with minimal bespoke effort.
• Validation at each site confirming data maps cleanly onto the common model before sign-off.

Lessons learned

The clearest lesson was that the unglamorous discipline of standardisation is what creates the value. It is tempting to rush data into the cloud and worry about modelling later, but a consolidated store of inconsistent data is just a larger silo. The investment in a common information model, painful as it was at the first few sites, was what made everything that followed possible. A second lesson was the wisdom of an edge-first stance in industrial settings; designing for intermittent connectivity from the outset spared us a great deal of difficulty later.

We were also reminded of the importance of treating rollout as a product rather than a series of projects. By packaging the edge deployment for repeatability, we turned what could have been an endless sequence of bespoke integrations into a scalable, predictable programme. The manufacturer ended with not just a platform but a repeatable capability for connecting future sites and acquisitions.

If your factory telemetry is trapped in site-level silos and holding back predictive maintenance and enterprise analytics, we can help you connect it safely. Talk to us about a similar engagement. Email sales@halfteck.com.