What is a smart factory, in one sentence?

A plant where machine state is recorded automatically, kept as comparable history, and used in a fixed routine that changes decisions. There is no certification for it and no official statistic counts smart factories.

Is Industry 4.0 the same thing as a smart factory?

In everyday use, yes. Industry 4.0, smart factory, digital factory, IIoT and digitalisation of production are used interchangeably for the same bundle of work. None of them is a defined standard, so ask which of the four layers a supplier means.

How many European manufacturers have already done this?

There is no direct measure. The closest proxy is Eurostat isoc_eb_ain2: 17.3% of EU-27 manufacturers with 10 or more employees reported using at least one AI technology in 2025, ranging from 3.2% in Romania to 39.8% in Belgium. That counts enterprises, not instrumented machines.

Where should a plant with no machine data start?

One line, one question with money attached, automatic machine-state capture on that line, a stop-reason list written with the operators, four to six weeks of baseline, then a daily routine. Analysis and AI come after there is history to analyse.

Do we need to replace our MES or ERP first?

No. ERP records what was produced and MES records how work was ordered and confirmed; neither normally records why a machine stopped for four minutes. The layers are complementary, and starting with a replacement of either is the slowest possible route to layer 1.

Home / Western European manufacturing in 2026: the data

Guide · Smart factory, Industry 4.0, IIoT · 22 September 2026

Smart factory: what actually changes on the shop floor

The term covers everything and promises everything. This page separates the four layers of work, says what each one changes for the people on the floor, and gives the European adoption figures with their dataset codes.

In shortA smart factory is not a standard and nobody certifies one. In practice the term covers four layers: machines that report their own state, a data layer that keeps that history, analysis on top of it, and a routine that turns the answers into decisions. Most failed projects buy layer three before they own layer one. In 2025, 17.3% of EU-27 manufacturers reported using any AI technology.

What the term covers, and where it comes from

Smart factory, Industry 4.0, IIoT, digital factory and digitalisation of production are umbrella terms from industrial policy and supplier marketing. They are not standards. No body certifies a smart factory, and no official statistic measures one. What statistical offices do measure are the components: enterprises using at least one AI technology, enterprises running ERP software, and enterprises using connected devices for production processes. Each is a separate Eurostat indicator, and each is counted at enterprise level, not at machine level.

The vocabulary shifts with the language, from Smart Factory Software and digitale Fabrik to fabbrica intelligente, cyfryzacja produkcji and akıllı fabrika, but the question behind the search is the same everywhere: what would actually change in my plant, and in what order.

The definition this page usesA plant is running a smart-factory project when machine state is recorded automatically, kept as history, and used in a routine that changes what people do. If any one of those three is missing, whatever is installed is a demonstration.

The four layers

The four layers, in the order they have to be built
Layer	What it is	The test that it is real	Typical failure
1. Sensing and machine state	The machine reports running, stopped, idle, and at what rate, without anyone typing it.	Pull yesterday's record for one machine. Does it show the stops the operator remembers, including the short ones?	Only production counts are captured, so stops are inferred, not observed.
2. Data infrastructure	Time-stamped history kept in one place, on one clock, with the part, order and shift attached.	Can you compare the same machine on the same product six months apart, in under a minute?	Data lives in the machine, is overwritten weekly, or has no product or order attached.
3. Analytics	Comparing periods, machines and products; finding which losses repeat and which are one-offs.	Can you name the top three recurring losses on a line and how many minutes each cost last week?	Averages replace distributions, so repeated short stops stay invisible.
4. Decisions and routines	A fixed meeting, an owner per action, and a rule for when a number triggers a reaction.	Name one thing the plant did differently last month because of the data.	Screens are installed, nobody has decision rights, the pilot never ends.

Layer 1 can exist without layers 2 to 4. Layers 2 to 4 cannot exist without layer 1.

The layers are cumulative and they are not equally expensive. Layer 1 is mostly engineering effort and access to signals. Layer 4 is mostly management discipline and costs nothing in licence fees. Layer 3 is where the money is usually spent, and it is the layer that produces the least on its own.

Connectivity, protocols and controller compatibility belong to layer 1 and are a separate subject with its own literature. The commercial question is narrower than it looks: for each machine, is there already a signal that distinguishes running from stopped, and who owns the change if there is not.

What changes for operators, supervisors and managers

Role	Before	After a working project
Operator	Stops are written on a paper sheet at the end of the shift, from memory, if there is time.	The stop is already recorded when it happens; the operator adds a reason in a few seconds from a short list written in their own words. The hour's target and actual are visible at the machine.
Shift supervisor	Handover is a conversation. Whether the line ran badly on Tuesday is an argument between two memories.	Handover starts from the record. The top three losses of the shift are on one screen, with the minutes attached, and the argument is about the cause rather than the fact.
Maintenance	Called when a machine stops. Frequency of small stops is unknown.	Repeated short stops on one station are visible as a pattern, so work can be planned against the station that actually costs the most minutes.
Plant manager	Capacity is estimated from standard times and a utilisation assumption.	Capacity claims, customer questionnaires and investment cases are answered from recorded history, including the hours the plant paid for and did not use.

This table describes what changes when layers 1, 2 and 4 are in place. Layer 3 changes the analysis, not the daily work.

Three things do not change and should not be promised. A monitoring project does not improve machine capability, it does not make an unstable process stable, and it does not reduce headcount by itself. What it changes is which problem the plant works on next week, and how quickly it notices that the problem came back.

The adoption reality in European manufacturing

Indicator	Figure	Year	Dataset
EU-27 manufacturers using at least one AI technology	17.3% (10.6% in 2024)	2025	Eurostat isoc_eb_ain2
Highest in Europe: Belgium, Denmark, Sweden, Austria	39.8%, 38.6%, 32.9%, 32.5%	2025	Eurostat isoc_eb_ain2
Lowest: Romania, Bulgaria, Poland, Hungary	3.2%, 5.3%, 7.7%, 7.8%	2025	Eurostat isoc_eb_ain2
EU-27 enterprises with at least a basic level of digital intensity, all sectors	73.7%	2024	Eurostat isoc_e_dii
Robot density, EU-27 and Germany, per 10,000 manufacturing employees	231 and 449	2024	IFR press release, 8 April 2026
Türkiye: enterprises running an ERP package, 10+ employees, all sectors	28.2%	2025	Eurostat isoc_eb_iip
Türkiye: enterprises using connected devices for production processes, all sectors	5.2%	2021	Eurostat isoc_eb_iot

Sources: Eurostat isoc_eb_ain2 (E_AI_TANY, NACE C, enterprises with 10+ employees), isoc_e_dii, isoc_eb_iip (E_ERP1) and isoc_eb_iot (E_IOTDPP); IFR robot density, data year 2024. Extracted 22 September 2026.

Two cautions before quoting any of theseEurostat broadened its AI question set for the 2025 wave, so the jump from 2024 to 2025 is partly methodological; treat the two years as separate measurements. And every one of these indicators counts enterprises, not machines or plants. A company qualifies if one department uses one tool. None of them tells you how many European machines report their own state.

The spread is the useful part. A manufacturer in Belgium sits among peers where two in five report some AI use; a manufacturer in Romania sits among peers where one in thirty does. The advice that follows is not the same. In the high-adoption countries the practical gap is usually layer 2, history that can be compared over time. In the low-adoption countries it is layer 1, machines that say nothing at all.

The Digital Intensity Index reading is the honest corrective: 73.7% of EU enterprises have at least basic digital intensity, which mostly means e-invoicing, websites and cloud email. Basic digital is not shop-floor digital, and the two are routinely confused in vendor material.

The sequence that works

Pick one line and one question worth moneyNot the whole plant. A question with a number attached: why does this line lose two hours a week, or why does changeover take 40 minutes on one shift and 70 on another.
Get machine state automatically before anything elseLayer 1 first, even if it covers only five machines. Automatic state capture is what makes every later layer trustworthy.
Write the stop-reason list with the operatorsFifteen to twenty reasons, in the words people already use, with a rule for what goes in each. A list written in the office gets used once.
Run four to six weeks without changing anything, and sign the baselineThe baseline is the only defence against the argument that things were always like that. Have production and maintenance sign the same numbers.
Put the routine in placeThe same fifteen minutes every day, the same three numbers, one owner and one date per action. This is layer 4 and it is free.
Only then extend, and only then analyseA second line, or deeper analysis on the first. Analysis on top of four weeks of clean history is useful. Analysis on top of nothing is a demonstration.

Three sequences that fail

Platform first, question later. The system is bought, connected and then searched for a use case. It ends as a screen nobody opens, because no decision was ever attached to it. The question has to exist before the purchase order.
Manual entry as the data layer. Operators type what happened. The data is complete on quiet weeks and full of holes exactly when the plant is busy, which is when the losses are. Manual entry is fine for the reason on top of an automatically recorded stop; it is not fine as the record of the stop.
Dashboards without decision rights. Screens go up, the numbers are correct, and nothing changes because no meeting owns them and nobody is allowed to stop a line on their evidence. The project is judged a failure of the software.

How to tell a smart-factory project from a dashboard project

Ask	Dashboard project	Smart-factory project
Where does the machine state come from?	Typed in, or inferred from production counts.	Read from the machine automatically.
What happens to last month's data?	Overwritten, or exported to a file nobody opens.	Kept, comparable, attached to product and order.
Who looks at it, and when?	Whenever someone remembers.	A fixed meeting with an owner and an agenda.
What triggers an action?	A manager noticing something.	A rule agreed in advance.
What changed last month?	The screen layout.	A named change on a named line, with the before and after.

If four of five answers sit in the middle column, the project is a reporting exercise. That can still be worth having, but it should be budgeted and described as one.

The last row is the only one that matters to a plant director. A smart factory is not a set of technologies, it is a plant where the record of what happened changes what happens next. Everything else on this page is in service of that sentence.

Who publishes this pageThese pages are published by TEEPTRAK SAS, which sells production-monitoring and OEE software. That is a reason to check every figure here against the datasets cited, and it is why this site does not rank vendors. For many plants the honest first step is not a purchase at all: a spreadsheet, a signed baseline and a weekly routine, or the machine data an existing MES or a local integrator can already expose.

Free data

Western Europe Manufacturing Data Pack 2026

Every indicator on this page for 17 European countries and EU-27, each row carrying its Eurostat dataset code and link. Excel.

Get the data pack →

Questions

What is a smart factory, in one sentence?: A plant where machine state is recorded automatically, kept as comparable history, and used in a fixed routine that changes decisions. There is no certification for it and no official statistic counts smart factories.
Is Industry 4.0 the same thing as a smart factory?: In everyday use, yes. Industry 4.0, smart factory, digital factory, IIoT and digitalisation of production are used interchangeably for the same bundle of work. None of them is a defined standard, so ask which of the four layers a supplier means.
How many European manufacturers have already done this?: There is no direct measure. The closest proxy is Eurostat isoc_eb_ain2: 17.3% of EU-27 manufacturers with 10 or more employees reported using at least one AI technology in 2025, ranging from 3.2% in Romania to 39.8% in Belgium. That counts enterprises, not instrumented machines.
Where should a plant with no machine data start?: One line, one question with money attached, automatic machine-state capture on that line, a stop-reason list written with the operators, four to six weeks of baseline, then a daily routine. Analysis and AI come after there is history to analyse.
Do we need to replace our MES or ERP first?: No. ERP records what was produced and MES records how work was ordered and confirmed; neither normally records why a machine stopped for four minutes. The layers are complementary, and starting with a replacement of either is the slowest possible route to layer 1.

Sources

Published by TEEPTRAK SAS, which makes production-monitoring and OEE software. Every figure is sourced on the page. Funding rules, standards and reporting duties change: check the official documents before you budget or commit.