Walk any Swiss production floor and you will find more instrumentation than anyone uses. Machines emit data. Some of it is stored. Almost none of it changes a decision that would otherwise have been made differently.
That gap, between collection and consequence, is the whole of Industrie 4.0 in practice for a manufacturer of 30 to 300 people. Everything else is procurement. This article sets out why pilots stall, what to instrument first, a twelve-week sequence, what to expect from equipment effectiveness, and the Swiss specifics that change the plan: labour cost, employee monitoring rules and the security of machines that were never meant to be online.
Why do Industrie 4.0 pilots fail to scale?
Industrie 4.0 pilots usually fail to scale for organisational reasons rather than technical ones. The pilot was chosen because it was easy to connect, not because it mattered. Nobody's job included acting on the data. And each machine's data lived in its own format, so the comparison across machines, which is where the value was, never happened.
The pilot was chosen for feasibility, not value. It ran on the newest line because that line already had connectivity. The newest line was also the one with the fewest problems, so the measured improvement was small and the business case for rolling out to the older lines never materialised.
Nobody owned the decision. The dashboard exists. It is open on a screen in the corridor. No one's job description says they act on it, so nobody does.
The data model was per machine. Each vendor's telemetry lives in its own schema, and comparing across machines requires work nobody scoped. The comparison is where the value was.
How far have smaller firms got with connected machines?
Not far, and mostly not on the production line. Across the EU in 2021, 26% of small enterprises used the Internet of Things against 48% of large ones, and the most common use was securing premises. Among firms using it, small enterprises were half as likely as large ones to apply it to condition-based maintenance, and less than half as likely to apply it to production processes.
The Eurostat figures are from 2021 and cover the EU, not Switzerland, but the direction matches what we see. The same holds for AI. Eurostat's 2025 survey of AI use found that manufacturers using AI applied it most often to marketing and sales, and to business administration. In Switzerland, the Federal Statistical Office reports that AI use is rising but depends strongly on size, with large firms ahead of medium and small ones, and production among the main areas of use.
None of that is a reason to hurry. A 60-person manufacturer that has not connected its machines has not missed anything that cannot be caught up in a quarter, provided the first quarter is aimed at the right station.
What should a Swiss manufacturer instrument first?
A Swiss manufacturer should instrument the constraint first: the station that sets the pace of the whole line. Not the newest machine and not the most expensive one. Improving throughput anywhere else produces inventory rather than output, because the parts simply queue in front of the station that was already the bottleneck.
If you do not know which station is the constraint, that is the first week of work, and it can be done with a stopwatch and a clipboard before anything is connected. Look for where work in progress piles up, which station everyone waits on, and which machine the planner protects when the schedule slips. In high-mix plants the constraint can move with the product mix, so check it across a few typical weeks, not one good day.
The sequence that produces output in a quarter
Twelve weeks, one station, one decision. That is a slower start and a much faster rollout, because the second station reuses everything the first one proved.
Weeks 1–2. Identify the constraint and baseline it manually: equipment effectiveness by shift, measured by hand. Painful, cheap, and the number everything else is judged against.
Weeks 3–5. Instrument that station only. One machine, one data path, one owner. Retrofit sensors or the controller's own signals are usually enough; a platform decision can wait.
Weeks 6–8. Put the number where the decision is made, at the shift handover, not on a corridor screen. Name the person who acts on it and agree what they do when it drops.
Weeks 9–12. Measure the change against the manual baseline. If it did not move, stop and find out why before adding a second station. The method for turning that result into a case the owners can judge is in how to measure ROI on a digital transformation, and the same quarter-long logic applied to the office is in the first 90 days of a digital transformation.
What is a realistic OEE improvement?
In the Swiss mid-market discrete manufacturers we have assessed, the first honest measurement of equipment effectiveness usually lands well below what management expects, often between 45% and 65% where the assumption was 80%. The gap is almost always unrecorded micro-stoppages and changeover time. These are typical observations, not benchmarks, and any specific line depends on where its losses sit.
Overall equipment effectiveness multiplies three ratios: availability, performance and quality. A line that runs 85% of planned time, at 85% of rated speed, with 95% good parts, scores about 69%, although each figure alone sounds healthy. That is why the hand-measured baseline matters. The losses hide in the gaps between the ratios.
The improvement then comes from the measurement itself before any automation. Making stoppages visible changes behaviour on the floor: changeovers get prepared before the machine stops, and the small recurring jam that everyone works around finally gets a maintenance ticket.
Where the Swiss context changes the calculation
| Factor | Effect |
|---|---|
| High labour cost | Automation payback is faster than EU averages suggest |
| Small batch, high mix | Rigid automation underperforms; flexibility beats throughput |
| Long equipment life | Retrofit connectivity, do not wait for replacement cycles |
| Skilled, stable workforce | Operator knowledge is an asset; instrument to support it, not replace it |
| Data protection and labour law | Anything that can measure individual performance needs care early |
The labour cost gap is large. The Federal Statistical Office puts average Swiss labour costs at CHF 63.62 per hour worked in 2020. Its most recent international comparison, for 2016, put an hour in Switzerland at 55.60 euros against 32.75 in Germany and 34.65 in France. Every hour of manual measurement, re-keying or waiting costs more here, which is why the payback on removing them is faster.
The plants are also small. Of the roughly 92,000 market enterprises in industry and construction that the Federal Statistical Office counted in 2023, about 81% had fewer than 10 employees and about 14,300 had between 10 and 49. That is the scale this sequence is written for: no automation department, a maintenance team of two or three, and an owner who signs every investment.
Does machine data count as employee monitoring?
Machine data counts as employee monitoring in Switzerland when it is designed or used to watch the behaviour of individual workers, which labour law prohibits. Recording the output of a workstation for quality control or planning is permitted in principle, provided it is proportionate and employees are informed in advance. The safe design is to aggregate by station and shift, not by named operator.
The Federal Data Protection and Information Commissioner explains the rule in Article 26 of Ordinance 3 to the Employment Act: monitoring systems aimed at employees' behaviour are banned, while systems used for other purposes, such as quality control recording the output of a workstation, are not. Telemetry linked to a named operator is also personal data under the Federal Act on Data Protection, so transparency duties apply as well.
That row in the table is regularly underestimated. Introducing per-operator data without consultation converts a technical project into an industrial-relations one, in a workforce whose knowledge you depend on. Tell people what is measured and why before the first sensor goes in, and show them the station view they will see at the handover.
How should a manufacturer secure machines it connects?
A manufacturer should secure connected machines by keeping production and office networks strictly separate, allowing remote access only through a VPN with two-factor authentication, keeping an inventory of every connected device, and changing every factory-set password. Most shop-floor controllers were designed for closed networks, so connecting them without these basics exposes equipment that cannot easily be patched.
Those measures come from the National Cyber Security Centre's guidance on protecting industrial control systems, which also recommends testing backup recovery at least every six months and starting with the low-effort, high-impact steps. In the plants we see, the machine builder's remote maintenance access is often the weakest point: find out who can reach which machine, and how, before adding anything new. The office-side priorities are in cybersecurity for Swiss SMEs, and our cybersecurity work covers the boundary between the two.
What about AI on the shop floor?
Only after the data path is reliable and someone owns the decision. Predictive maintenance on a machine whose telemetry has gaps produces confident predictions from incomplete data, which is worse than no prediction because it is trusted.
The sequence is the same one we describe for getting AI past the pilot stage: name the decision, name the owner, build the data path, then the model. The data requirements are set out in the five data foundations an SME needs before AI.
The ERP question
Sooner or later the shop-floor data needs to meet the order book, and that is an ERP integration rather than an IoT project. If an ERP replacement is already on the roadmap, sequence the two deliberately, because instrumenting a process you are about to redesign wastes both efforts. Our view on running that selection quickly is in ERP selection for the Swiss mid-market.
If the first quarter raises a genuinely technical question, such as whether a new inspection method works on your parts, a research partner can answer it before you buy. Innosuisse's innovation cheque funds up to CHF 15,000 of a Swiss research partner's costs for a preliminary study, for organisations with fewer than 250 full-time employees.
Our digital transformation practice runs this sequence one station at a time, and our industrial page covers the wider work with Swiss manufacturers.
Sources
- Eurostat — Use of Internet of Things in enterprises, IoT use and purposes by enterprise size, 2021
- Eurostat — Use of artificial intelligence in enterprises, purposes of AI use in manufacturing, 2025
- Federal Statistical Office (BFS) — Weitere Verwendungen von IKT, AI use in Swiss enterprises by size and area
- Federal Statistical Office (BFS) — Arbeitskosten, labour cost per hour worked, 2020, and the 2016 international comparison
- Federal Statistical Office (BFS) — Kleine und mittlere Unternehmen, enterprises by sector and size class, 2023
- Federal Data Protection and Information Commissioner — Monitoring systems in the workplace, Article 26 of Ordinance 3 to the Employment Act
- Fedlex — Federal Act on Data Protection (FADP), SR 235.1
- National Cyber Security Centre (BACS) — Measures to protect industrial control systems
- Innosuisse — Innovation cheque, up to CHF 15,000 for a preliminary study with a research partner
Related reading
- Digital transformation for an SME: what to do in the first 90 days
- How to measure ROI on a digital transformation before you fund it
- ERP selection for the Swiss mid-market
- Cybersecurity for Swiss SMEs: the five things that actually matter
- Digital transformation consulting
Working on this right now?
Tell us where you are in two questions. A consulting partner reviews every enquiry within 2 business days.
