"Industry 4.0" has been a marketing slogan for over a decade. Some of what was promised has landed — connected machines, predictive maintenance, real OEE data — and a lot has not. For mid-market manufacturers (turnover £50–500m, multi-site, mixed-age machinery), the honest 2026 posture is less ambitious than the slides and more useful than the sceptics fear.
Start with the machines you have, not the ones in the brochure
A typical mid-market manufacturer has three generations of equipment. 20–40% modern (OPC UA, Ethernet/IP, MTConnect, clear data model). 40–60% middle-aged (serial, some Modbus, some Allen-Bradley, connectable with work). 10–30% old (air-gapped, no data out, replaceable only on capex cycle). Any programme that assumes all machines are modern fails.
The sequence that works: instrument the modern 20–40% first; bolt-on shop-floor telemetry to the middle-aged machines where the production value justifies; leave the air-gapped minority for the replacement cycle. Do not try to solve the hardest machines first.
OEE — the foundation, done honestly
Every manufacturer has an OEE number. Most of those numbers are estimated, not measured. An honest OEE programme collects availability (machine state logs), performance (cycle time data), and quality (scrap rate) directly from the line, minute by minute, and reconciles them. The first time this is done, the measured OEE is almost always 8–18% lower than the reported figure. This is a feature, not a problem; leaders prefer a measured 64% they can improve to an estimated 72% they cannot.
Quality — the highest ROI category
In our manufacturing engagements, quality initiatives have consistently returned more than any other category. Vision inspection, statistical process control, early-warning models on process data. The pattern we see repeatedly: a 20–30% reduction in scrap in targeted cells within six months, driven by tighter process control and faster reaction to drift.
Predictive maintenance — the honest version
Predictive maintenance is oversold at the general level and undersold at the specific level. The truth:
- For high-value, critical-path machines with rich sensor data, predictive models have genuinely moved failures from unplanned to planned. The ROI is real and measurable.
- For broad-estate predictive maintenance across dozens of machines of mixed age, the data quality and cost of instrumentation rarely repay the investment. Condition-based maintenance with sensor-led thresholds is usually enough.
- The most common failure is "we built a predictive model but no one changed the maintenance schedule". The model is only valuable if it replaces or modifies the existing PM regime.
The data architecture pattern
For a 2–10 site mid-market manufacturer, the reference pattern we ship:
- Edge ingestion (InfluxDB or TimescaleDB on a ruggedised compute at each site)
- Site-local dashboards for the shop floor — latency-sensitive, do not depend on the cloud
- A cloud data platform for cross-site analytics (Snowflake, Databricks, Fabric — all viable)
- A semantic layer for OEE and KPIs — crucial, because definitions drift between sites
- Integration to the MES, ERP, and quality systems — this is where the effort lives
None of this is novel. All of it is the thing that, done well, underpins every more-ambitious programme.
The organisational posture
- Ownership at the plant. A data lead per plant, usually a process or quality engineer with interest. The cloud team supports; the plant owns.
- A cross-plant forum. Monthly, two hours, share wins and learnings. This is where mid-market manufacturers compound.
- A capex conversation linked to the data strategy. New machines specified with data-out capability; not an afterthought.
- Patience. The programmes that land are three- to five-year programmes, not transformational six-month pushes. The "breakthrough in a quarter" framing is what has given Industry 4.0 its bad reputation.
The regulatory overlay
Increasingly, mid-market manufacturers are exposed to supply-chain data demands — traceability, provenance, scope 3 emissions. These are data problems as much as operational ones. The firms that have put the OEE and quality data in place find the supply-chain data demands comparatively easy. The firms that have not are scrambling.