Process automation in Portuguese industry: an operational guide

In a garment factory near Famalicão, on the 28th of each month, there is always the same scene: the administrative manager prints out Excel sheets, dashes over to production with a ballpoint pen behind her ear and asks, section by section, "how many pieces did you close yesterday?". The answer comes in round numbers, rounded off in someone's head, transcribed by hand onto another sheet that someone then keys into the ERP the following week. This is not automation. It is the digitisation of a sheet of paper — which is a completely different thing.

The thesis of this guide is simple and uncomfortable: most Portuguese industrial companies do not have a technology problem — they have an unmapped process problem that technology amplifies. Automating a bad process gives you a bad process that is faster and more expensive to fix. Before buying software, you need to understand what you are really doing on the shop floor. And that is where almost all projects fail.

We write this after more than three decades implementing systems in factories in the North, and the lesson repeats itself with an almost stubborn regularity. Technology is rarely the breaking point. The breaking point is human, it is procedural, it is political within the company itself. This guide runs through the problem from the shop floor to the AI Act, without conference-room optimism and without selling shortcuts that do not exist.

1. The real operational problem

Anyone who has been around factories in the Vale do Ave, footwear halls in Felgueiras or warehouses in the Lousada–Paços de Ferreira corridor recognises the pattern. The data exists. It is just in the wrong places: in the foreman's head, in a black-covered notebook, in a WhatsApp group, in a spreadsheet that only one person knows how to open without breaking the formulas.

The pattern is neither laziness nor lack of intelligence. It is the opposite. It is competent people building parallel systems because the official systems do not keep up with the speed and irregularity of real work. The foreman's notebook exists because it works. The official ERP, very often, did not work for that specific case — or it worked in a way that required more clicks than the operator had the patience or the time to give during a shift under delivery pressure.

The hidden cost of repetitive manual work

Repetitive manual work has three costs that rarely show up on the bill. The first is direct time — hours of administrative staff copying data from one system to another. The second, more expensive, is error: a reference swapped in a picking, a batch wrongly identified in finishing, an out-of-date price in a mobile sale. The third, the worst, is decision latency. By the time the industrial director finally knows that the efficiency of a line has dropped, three days have passed and the order is late.

It is worth breaking down each of these costs, because most companies only see the first — direct time — and ignore the other two, which are much larger.

Direct time is the most visible and the least important. An administrative worker who spends two hours a day re-keying transport documents costs, over a year, a few weeks of pure work. It is annoying, but it is manageable. The company gets used to it and folds it into the cost structure without even questioning it.

The cost of error is where the money starts to bleed without anyone noticing. A reference swapped in a distribution picking is not just a returned box — it is the return, the reprocessing, the return transport, the credit note, the phone call from the irritated customer and the erosion of trust that makes that customer ask the competition for a quote on the next purchase. A wrongly identified batch in a textile finishing, when the brand is Inditex or Decathlon and requires traceability, can mean the rejection of an entire order and the loss of a specification for the following season.

Decision latency is the invisible cost that separates factories that grow from those that stagnate. In an operation where production information arrives three days later, the industrial director always governs looking in the rear-view mirror. He corrects problems that have already done their damage. In an operation with real-time capture, the same director sees the efficiency of a workstation drop at 10:30 and intervenes at 11:00. That difference of seventy-two hours versus thirty minutes is, over the course of a year, the difference between meeting deadlines and spending your life firefighting.

The time you spend copying data is the cost you see. The error that data carries and the decision it delays are the cost that kills you slowly — and that never shows up on an Excel sheet.

The anatomy of an error in footwear

In footwear, this is visceral. A sample collection has between 800 and 1,200 SKUs, with three axes crossing — colour, size and last. No generalist ERP models this without workarounds. When international buyers arrive (men's footwear in August, women's in February), the factory that does not have the technical sheet, the cost and the availability on a single screen loses business plainly and visibly.

The footwear problem deserves detail because it is the clearest example of why a generalist ERP fails in a vertical sector. Consider a women's shoe model: it exists in five colours, in nine sizes and in two last widths. That is ninety combinations for a single model. A collection of sixty models thus generates thousands of references that have to share a common technical sheet but vary in material consumption, in cost and in stock availability per combination.

A generalist ERP treats each of those ninety combinations as an isolated item, forcing the entry of ninety sheets where a single matrix should exist. Multiplied by sixty models, the maintenance work becomes impossible to keep up to date — and the result is that no one keeps it up, and the system stops reflecting reality. This is exactly where the parallel black-covered notebook is born.

The relentless calendar of international buyers

The Portuguese footwear sector lives on a calendar that does not forgive. International buyers visit twice a year with tight windows. When a buyer from Germany or the United States sits down in the sample room and asks the price of a specific variant in a specific quantity with a specific delivery date, the factory has seconds to respond — not days.

The factory that can, on screen, cross the technical sheet, the current cost (with the updated raw-material price) and the available production capacity, closes the deal in the room. The factory that answers "let me check and I'll call you tomorrow" loses the order to a competitor who answered on the spot — often a competitor next door, in Felgueiras or in São João da Madeira. APICCAPS has documented the competitive pressure on the cluster for years, and the difference is rarely in the quality of the product. It is in the speed of information.

Why the Excel sheet is a symptom, not the disease

Excel is not the enemy. Excel is the first-aid station the company set up because the formal system did not respond to reality. Every parallel spreadsheet is a report of a problem: here is a process that the ERP does not cover, or covers badly, or covers in a way that no one on the shop floor agreed to use.

There is a temptation, among consultants and vendors, to demonise Excel and present it as the villain to be eliminated. It is a lazy reading. The parallel spreadsheet is the most faithful map of the holes in the official system. Anyone who wants to design an automation that works should start by cataloguing all the parallel Excel sheets in the company and asking, one by one: what process does this sheet solve that the official system does not?

The answer to that question is the true specification of the automation project — far more honest than any formal requirements survey done in conference-room meetings.

Before asking "which software do I buy?", ask "how many parallel Excel sheets does my company maintain — and what is each of them telling me about the hole it plugs?"

The classic symptom: the "month-end" that lasts a week

The most honest test of the maturity state of a Portuguese industrial company is to measure how long the monthly close takes. If it takes more than three working days to have reliable figures, the problem is not the accounting. It is that the operational data — production, consumption, stock, hours — is not flowing in real time to the system. It is being reconstructed after the fact, by hand, by tired people.

This test is revealing because it requires no complex technical diagnosis. Any CEO can answer the question "how many days does my month-end close take?" in seconds. And the answer correlates almost perfectly with the degree of real digitalisation of the operation.

A company whose close takes one day almost certainly has data flowing in real time from the factory to finance. A company whose close takes a week is reconstructing the month from fragments — adding up notebooks, reconciling Excel sheets, chasing the foreman to confirm how many pieces went out during the week he was off sick. This reconstruction work not only consumes time but introduces error precisely at the moment when the numbers are going to be used for financial decisions.

The difference that size makes to the same symptom

It is worth noting that the same symptom has different severities depending on size. In a micro-company of ten people, a three-day close is annoying but not fatal — physical proximity allows one to compensate for the lack of a system. In a medium-sized company of two hundred people with three production halls, the same three-day close means that management is permanently navigating with obsolete information, and the compound cost of that blindness over an industrial year is enormous.

2. What exactly is process automation in Portuguese companies

Process automation is the replacement of manual, repetitive, rules-based tasks with flows executed by systems — with human intervention reserved for decisions and exceptions. It is not robotics with mechanical arms (that is industrial equipment automation). It is the automation of information work: the order that comes in and automatically generates the production order, the production entry that updates stock and cost, the document that follows an approval circuit without leaving a mailbox.

This distinction between equipment automation and information automation is more important than it seems. Many Portuguese factories have already invested heavily in equipment automation — laser cutting machines, sewing robots, numerically controlled plastic injection lines. But those machines still produce data that no one captures in real time. The mechanical arm is modern; the record of what it produced is still done by hand on a sheet at the end of the shift. It is information automation that is missing, not equipment automation.

The layers: from RPA to integrated ERP

There is a useful hierarchy of maturity, from the most superficial to the most profound:

• Task automation (RPA): a software robot that copies data between applications that do not talk to each other. Quick to set up, fragile to maintain — if the application changes its screen, the robot breaks.
• Document flow automation: capture, classification and circulation of invoices, delivery notes and contracts with approval rules. This is the terrain of Document Management with cognitive capture and qualified signature.
• Transactional automation in the ERP: the business process lives inside the central system — sales, purchasing, production and finance share the same data without transcription. This is where a vertical MULTI ERP solves what Excel patched.
• Real-time shop-floor automation: capture of production at the terminals, calculation of OEE by the minute, integration with machines via OPC-UA or MQTT.
• Cognitive automation (applied AI): demand forecasting, anomaly detection, planning suggestion. The most recent layer and the most misunderstood.

The order of these layers is not accidental. It reflects a logical sequence of dependency: you cannot calculate reliable OEE (layer 4) without first having the production process living in the ERP (layer 3), and you cannot feed a useful demand-forecasting model (layer 5) without first having clean, historical sales and production data (layers 3 and 4). Anyone who tries to skip layers pays the price in projects that never stabilise.

Where the real value is by layer

Not all layers yield the same per euro invested. Experience on the Portuguese ground shows a clear pattern of where returns concentrate as a function of effort.

Automation is not artificial intelligence (and that confusion costs money)

Many companies delay automation projects because they think they need AI. They do not. The overwhelming majority of automation value in a Portuguese industrial SME is in layers 2, 3 and 4 — deterministic rules, not predictive models. AI comes later, when there is already clean data and a stable process to feed it. Buying AI for dirty data is paying dearly for bad predictions.

This confusion costs money in two ways. The first is paralysis: the company postpones the entire automation project because it thinks it is "not ready for AI yet", when in truth it never needed AI to reap ninety per cent of the available value. The second is waste: the company launches into an AI project on data that does not exist or that is inconsistent, spends budget — often public funds — and produces a model that gives predictions so bad that no one trusts them or uses them.

Automation is doing the same thing without human intervention. Artificial intelligence is deciding what to do. Confusing the two is the fastest way to spend PT2030 budget on a report that no one uses.

A brief Portuguese historical note

The computerisation of Portuguese industrial SMEs happened in three waves. The first, in the 1990s, was accounting and invoicing — tax obligation pulling. The second, in the 2000s, was the ERP expanding into purchasing, sales and stock. The third, ongoing, is the real-time capture of what happens on the factory floor and the integration of the silos. It is in this third wave that Portugal is behind the European average — and it is there that competitiveness against reshoring and Asian competition is played out.

Each wave was pulled by a different engine, and this explains the uneven shape of Portuguese digitalisation. The first wave was pulled by tax obligation — companies computerised accounting because the State required it. The second wave was pulled by management — the owner wanted to know how much he had in stock and how much he owed suppliers. The third wave, the one under way, is pulled by international competitiveness: by anchor customers who demand traceability, by shrinking deadlines, by tightening margins.

The fundamental difference is that the first two waves were pushed by mandatory external forces, whereas the third depends on a voluntary investment decision. And it is precisely because it is voluntary that so many companies postpone it — there is no fine for not capturing OEE in real time, there is only the slow loss of competitiveness that only becomes visible when it is already too late.

3. The picture in Portugal today

The most recent figures dismantle the optimism of the conference slides. In 2025, only 11.5% of companies in Portugal with 10 or more workers used artificial intelligence technologies, up 2.9 points from 2024 (INE, 2025). It is not a collapse — but it is far from the European Union average of 20% (Eurostat, 2025).

It is worth reading these figures carefully, because they measure AI adoption specifically, not digitalisation in general. Basic digitalisation — electronic invoicing, transactional ERP — is much more widespread in Portugal, pushed by tax obligation. The lag concentrates in the more advanced layers of automation and in AI applications, which are precisely the ones that separate competitive companies from those surviving on the margin.

Adoption grows with size — and that is where it hurts

The Portuguese problem has a familiar shape: adoption concentrates in large companies. In 2025, AI was used by 49.1% of large companies, 18.2% of medium ones (50-249 workers) and only 9.4% of small ones (10-49) (INE, 2025). The industrial fabric of the North is made up of exactly those small and medium companies — the 40-person garment factory, the 60-person sole factory, the 90-person regional distributor. They are the ones being left behind, and they are the ones who most need efficiency to compete.

This gradient by size is the most important piece of the whole picture, because it reveals where the competitive fracture lies. Large companies have IT departments, they have budget for consulting, they have staff with recent training. The small and medium industrial companies of the North typically have an IT manager who learned the business over fifteen years, who masters every corner of the operation, but who is alone and without time to lead a transformation while keeping the current systems standing.

The irony is cruel: the companies that would benefit most from efficiency — those competing on tight margins against Asian production — are exactly the ones with the least internal capacity to achieve it. And the gap is not closing; the trend in the data suggests it is widening, as the large ones accelerate and the small ones remain stuck in day-to-day management.

The real barrier is not money — it is knowledge

Asked what holds them back, European companies answer, in first place, lack of skills and knowledge (70.9%), and only then legal uncertainty (52.5%) (Eurostat, 2025). This tallies with what is seen on the Portuguese ground: the obstacle is rarely the price of the software. It is that no one inside the company has the time, training or mandate to map the processes and lead the change.

This datum should change the way companies think about investment. The natural tendency is to budget for the software — the licence, the terminals, the server. But if the dominant barrier is knowledge, then the critical investment is in support, in training and in the internal time dedicated to the project. A company that buys expensive software and does not free up a competent person to lead the change is optimising the wrong variable.

On a global scale, the contrast is brutal: 78% of organisations already use AI in at least one business function and 71% regularly resort to generative AI (McKinsey, 2025). The distance is not technological — the technology is available and cheap. The distance is one of internal execution capacity.

What public funding can and cannot solve

Portugal has, in this cycle, substantial funding instruments — PRR, PT2030, COMPETE 2030, Norte 2030 — that can cover a relevant part of the investment in digitalisation. This attacks the barrier that is not the main one (money) and leaves untouched the one that is (knowledge and execution). We frequently see companies that obtain funding approval and then get stuck at the execution stage, precisely because money does not buy the internal time nor the competence to lead the change.

The typical pattern is the approved project that gets stuck in the technical reports, where the company has to demonstrate results it could not produce because it did not have execution capacity. Funding is a real lever, but it does not replace the discipline of process mapping nor the involvement of the people who will use the system. Those who treat funds as the solution, rather than as a facilitator, end up returning tranches or justifying spending without impact.

4. The five implementation models

There are five practical approaches to automating processes in a Portuguese industrial company. They are not mutually exclusive, but each has a very different profile of cost, risk and time to value. Getting the sequence wrong is the most common and most expensive error.

Comparison of the models

RPA: the temptation of the shortcut

RPA — robotic process automation — is seductive because it promises results in weeks without touching the existing systems. The robot simply mimics what a person would do: it opens an application, copies a value, pastes it into another, presses save. For occasional bridges between systems that have no way of communicating, it has its place.

But RPA carries a hidden debt. The robot depends on the appearance of the screen. It only takes an update to one of the applications — a button that moves, a field that changes name — for the robot to break silently and start producing errors that no one detects for days. In the medium term, a web of RPA robots becomes an expensive technical debt to maintain, precisely because any update to any system involved can break any robot. It is a patch, not a cure. It serves to buy time while the underlying problem — the lack of integration — is being solved, but it should not be treated as the final solution.

Document workflow: the most reliable quick win

Document flow automation is, in our experience, the model with the best relationship between effort and stable return for an industrial SME. The cognitive capture of supplier invoices, with automatic classification and approval circuit, eliminates manual entry, speeds up payments and creates a searchable and legally valid archive with qualified eIDAS signature. The trap is digitising the chaos: automating an approval circuit that was already confused on paper only produces confusion faster. The prior work of simplifying the circuit is what separates success from failure.

Big-bang versus incremental

The CEO's temptation is the big-bang: stop everything over a weekend and switch on the new system on Monday morning. It works in theory. On the Portuguese ground, with ageing teams in garment factories and foremen who master the process but distrust the screen, the big-bang has an extremely high friction rate. The incremental approach — one module, one line, one warehouse at a time — is slower on paper and faster in reality, because it does not collapse the operation when something goes wrong.

The mathematics of risk explains why. In a big-bang, if something goes wrong — and something always goes wrong — the entire operation stops, and the pressure to revert is enormous, often leading to the abandonment of the project and the consolidation of the internal narrative that "the new system doesn't work". In an incremental approach, a failure affects only one line or one warehouse, the damage is contained, it is learned from and corrected before advancing. The incremental also builds allies: each module that works creates internal champions who make the next module easier.

The big-bang is fast on the Gantt chart and slow in reality. The incremental is slow on the chart and fast in reality — because there is nothing more time-consuming than recovering from a stopped operation.

The role of low-code

For companies of great process complexity — multi-factory, multi-country, dense subcontracting chains — the low-code model, such as QAD Adaptive ERP, allows the system to be adapted without rewriting code at every change of business rule. It reduces dependence on heavy development. It does not, however, eliminate the need for someone who really knows how the company works on the inside.

Low-code responds to a real problem of the more complex companies: the business rule changes faster than the traditional development cycle can keep up with. When an international parent company alters a traceability requirement or a new factory enters the structure, waiting months for development is unacceptable. The ability to reconfigure the system without heavy code gives agility — but transfers the responsibility to whoever configures it. If that person does not thoroughly master the business process, low-code merely allows bad solutions to be built faster.

No implementation model survives contact with a warehouse manager who cannot leave the radio for more than two hours. Design the rollout around him, not against him.

5. How to assess whether your company needs it

Not every company needs to automate everything, nor right now. The honest assessment begins by identifying where the pain is real and measurable. A process done once a quarter in twenty minutes is not worth a project. One done fifty times a day, with error, is worth gold.

Signs that it is already too late to postpone

• The monthly close takes more than three working days to have reliable figures.
• There are parallel Excel sheets that only one person knows how to maintain — and that person is going on holiday.
• The industrial director only knows the efficiency of a line at the end of the week, too late to correct it.
• An anchor customer (Inditex, Decathlon, Lacoste) asks for batch traceability and the answer takes days.
• The sales force takes orders on paper that someone re-keys at the office.

Each of these signs has an associated direct cost that can be quantified. The decision point is not philosophical — it is arithmetic. If the sum of hours lost, errors paid for and missed opportunities exceeds the cost of the automation project over a horizon of one or two years, the decision is made. What prevents most companies from doing this calculation is that the costs are dispersed and invisible, while the cost of the project is concentrated and visible.

Step by step of the diagnosis

1. Map the processes that hurt. List the ten tasks that consume the most hours per week in the back office and on the shop floor. Measure with a stopwatch, not by intuition — people underestimate the time of what they do on autopilot.
2. Classify by frequency and rule. For each task, ask: is it repetitive? Is it based on clear rules? How much does an error cost? The candidates for automation are those of high frequency, clear rule and expensive error.
3. Identify the source data. To automate a flow, the data has to exist in a capturable format. If the information is born in someone's head, first the capture is created, then the flow is automated — never the other way around.
4. Estimate the return in hours and errors avoided. Translate each candidate into hours saved per month and cost of error avoided. Prioritise the three with the best ratio. Ignore the rest for now.
5. Check the team's readiness. Those who will use the system every day have to be in the room from the start. Onboarding is not a two-hour training session at the end — it is the joint design of the process from the first week.

The frequency-rule-error matrix in practice

The second step of the diagnosis deserves an explicit method, because it is where prioritisation is won or lost. Each candidate task should be scored on three axes, and only those that score high on all three deserve priority automation.

A high-frequency task but with a decision that requires complex human judgement is not a candidate for full automation — it is a candidate for decision support, which is different. A task of clear rule but low frequency and cheap error can wait indefinitely. The classic error is to automate what is technically easy instead of what is economically valuable.

Quick wins you implement in two weeks

Not everything is a nine-month project. There are quick wins that build internal confidence and free up mental budget for what is bigger:

• Automatic capture of supplier invoices with cognitive reading and email approval circuit — eliminates manual entry and the "where's the invoice to pay?".
• A single dashboard with the five KPIs that the CEO consults every Monday, fed directly from the ERP in Qlik Sense, instead of the hand-made PowerPoint report.
• B2B customer orders via portal instead of email and telephone, eliminating re-keying and reference errors.

Quick wins have a value that goes beyond the direct return: they are the internal political demonstration that automation works. In a company where the foreman distrusts the screen and the CFO distrusts the investment, a visible win in two weeks changes the conversation. It stops being "is this worth it?" and becomes "where do we apply it next?". Choosing the first quick win by its visibility, and not just by its technical return, is a strategic decision that is often underestimated.

6. What to choose and why (decision by company size)

The right recommendation depends almost entirely on the size and complexity of the company. Giving the same advice to a 30-person garment factory and a 400-person textile group is the recipe for wasting money on one of the two sides.

Decision matrix by size

For the small garment factory in the North

In a small garment factory, with experienced but ageing labour, the priority is to have an integrated core that eliminates re-keying and meets tax obligations. It is not the time for AI. It is the time to put sales, purchasing and stock sharing the same data and to ensure certified invoicing and SAF-T communication without hiccups. Resistance to change is real — design the system to reduce clicks, not to multiply them.

Demographics matter in this profile in a way that is rarely admitted out loud. Portuguese garment manufacturing lives on experienced and ageing labour, with decades of craft knowledge but little ease with screens and keyboards. Imposing a system that requires many digital steps on someone who spent thirty years working with their hands is to guarantee rejection. The solution is not to force people to adapt to the software — it is to adapt the software to the people, minimising digital interaction, using code reading instead of typing, large buttons instead of deep menus.

For the medium-sized footwear or textile factory

Here the fastest return is in the real-time capture of the shop floor. A solution like KORA Productivity transforms the manual end-of-shift entry into by-the-minute data — OEE, efficiency per station, scrap. The error one always sees is buying terminals and not designing the interface with those who will touch them. The operator who finds the screen slow hides the tablet behind a box and goes back to the notebook. The technology was right; the adoption process was wrong.

The medium-sized industrial company is the profile where shop-floor capture yields most, because it has enough volume for information latency to hurt seriously, but not so much complexity that it requires a years-long implementation. A footwear or textile factory with two hundred people and several lines gains enormously from seeing efficiency by the minute instead of reconstructing it at the end of the shift. But success depends entirely on adoption by the operator. Each second that the terminal takes to respond is a second the operator loses in piece-rate paid work — and he will do anything to avoid it. The interface has to be faster than the notebook, not just more modern.

For the distributor with a large warehouse

In distribution, the gain is in the warehouse: guided picking, automatic packing list, cross-docking. A warehouse management solution reduces the shipping error and frees the warehouse manager from paper sheets. Golden rule for this profile: any rollout that takes the warehouse manager off the radio for more than two hours in a row will meet active resistance. Do it by zones, outside the peaks.

The warehouse manager in the Lousada–Paços de Ferreira corridor is an archetypal figure who deserves respect, not frustration. He governs the apparent chaos of the warehouse with the radio in hand and a mental map that no system has documented. Taking him off the ground for prolonged training is stopping the warehouse. The right approach is to implement by zones — one aisle, one product family at a time — outside the shipping peaks, with training at the actual workstation and not in a classroom. A mobile sales system for the sales force often completes the picture in distribution, eliminating the re-keying of orders that are born at the customer and die in an error at the office.

For the metal and plastic industry

In the metal and plastic industry — moulds, injection, single piece versus series — the challenge is different from the previous ones. The mix of unit-order production and series production requires an ERP that models both without forcing one logic onto the other. The Aveiro–Marinha Grande corridor is in full transition to Industry 4.0, and here machine-to-system integration via OPC-UA has a direct return, because injection machines already produce data that only needs to be captured and turned into a decision. The error of this sector is to underestimate the difference between digitising the series (easy) and digitising the single piece (which requires capture of real cost per project).

7. Applicable regulatory framework and compliance

Automating processes in Portugal is not a free technical exercise — it is an exercise framed by law. Ignoring the framework turns an efficiency project into a fine risk.

Invoicing and tax communication

Any automation that touches commercial documents has to respect Decree-Law 28/2019 — electronic invoicing, ATCUD and software certified by the Tax Authority. The monthly SAF-T communication, under Ordinance 195/2020, requires the data to flow correctly from the system to the Tax Authority. An ERP that automates invoicing without Tax Authority certification does not automate anything — it creates a tax problem.

This point seems obvious but is a source of frequent suffering. Companies that develop homemade automations or that adopt uncertified tools to touch tax documents discover, often too late, that they have created a non-compliance. The Tax Authority certification of the software is not a bureaucratic detail — it is a legal requirement whose non-compliance exposes the company to fines and to the invalidity of the documents issued. Any automation project involving invoicing, delivery notes or SAF-T communication has to start by confirming the certification of the base software.

Data protection and cybersecurity

Automation implies concentration of data, and concentrated data is a target. The GDPR and Law 58/2019 govern the processing of personal data — relevant above all in HR modules such as pplPortal, where predictive AI for turnover and absenteeism enters the terrain of automated decisions that the regulation watches over.

The NIS2 Directive (EU 2022/2555), transposed by DL 65/2025, extends cybersecurity obligations to many industrial companies that were previously not covered. Anyone who automates the shop floor and connects machines to the network gains a larger attack surface — and obligations to protect it. It is worth reading our NIS2 compliance guide for industrial SMEs before connecting the first sensor. A SIEM is no longer a large-company luxury.

The relationship between automation and cybersecurity is frequently ignored, with serious consequences. Each machine connected to the network, each shop-floor terminal, each B2B portal accessible from outside is a potential entry point. A factory that was essentially offline and that becomes digitalised gains an attack surface it never had — and most do not think about this until an incident happens. ENISA has documented the growth of attacks on industrial infrastructures in Europe, and the manufacturing sector is far from immune. Security has to be designed simultaneously with automation, not added afterwards.

The AI Act and cognitive automation

For those advancing to the AI layer, the EU AI Regulation (Regulation (EU) 2024/1689) has been in force since August 2024, with the prohibitions applicable since February 2025 and the general-purpose model rules since August 2025 (European Commission). AI systems that affect decisions about people — recruitment, evaluation, worker monitoring — fall into risk categories with concrete obligations. Automating performance evaluation without understanding this classification is creating silent legal exposure.

The AI Act introduces a risk classification that changes the way automations that touch people are designed. A system that suggests production planning is low risk. A system that classifies job candidates, that evaluates the performance of workers or that monitors them falls into high-risk categories with obligations of transparency, human oversight and documentation. A company that automates people management with predictive AI components has to know exactly which risk category each function falls into — and design human oversight accordingly. Ignoring this is not just a legal risk; it is a reputational risk with the workers themselves.

The whistleblowing channel and size obligations

It is also worth remembering that Law 93/2021 requires companies with fifty or more employees to maintain a whistleblowing channel — an obligation that many growing industrial companies cross without noticing. When a garment factory or a footwear factory crosses the threshold of fifty workers, it inherits a set of obligations — whistleblowing channel, reinforced GDPR requirements, potential coverage of NIS2 — that require systems to comply with them. Digitalisation, here, is not optional; it is the means to comply with the law without drowning the company in manual process.

Compliance is not the brake on automation — it is the design of automation. Those who treat the law as an annex to the project pay for it twice: once in the fine, once in the rework.

8. How INFOS approaches this

INFOS has been building vertical enterprise software for industry for 35 years,

Frequently asked questions

What is the difference between digitalisation and automation?

Digitalisation is converting a manual process into digital format — for example, printing Excel and then typing by hand. Automation is eliminating repetitive manual work through integrated systems. Digitising a bad process only makes it faster and more expensive to fix. You need to map and optimise the process before automating.

Why do most automation projects in Portuguese factories fail?

The problem is not technological, but procedural and human. Companies automate unmapped and inefficient processes. The breaking point is political within the organisation, not in the software. Without understanding the real work on the shop floor and involving the teams, technology amplifies existing problems instead of solving them.

What is the real cost of repetitive manual work?

It has three components: direct time (hours of administrative staff copying data), cost of error (returns, reprocessing, loss of customers) and decision latency (delayed information prevents rapid action). The cost of error and latency are much larger than direct time, but rarely appear in the company's accounts.

How does decision latency affect production?

When information arrives three days later, the industrial director governs by the rear-view mirror, correcting problems already consummated. With real-time capture, an efficiency drop detected at 10:30 is resolved at 11:00. That difference of seventy-two hours versus thirty minutes, over the course of a year, is the difference between meeting deadlines and constantly firefighting.

Why do generalist ERPs fail in the footwear sector?

Footwear has a complex structure: a model in five colours, nine sizes and two widths generates ninety combinations. Generalist ERPs treat each combination as an isolated item, requiring impossible maintenance of thousands of references. This forces companies to create parallel systems — notebooks, spreadsheets — because the official system does not keep up with the operational reality.

Where is the data in Portuguese factories?

The data exists, but in the wrong places: in the foreman's head, in black-covered notebooks, in WhatsApp groups, in spreadsheets that only one person knows how to open. This is not laziness — it is competent people building parallel systems because the official systems do not keep up with the speed and irregularity of real work.

What is the first step before buying automation software?

Mapping and understanding the current process on the shop floor. Automating a bad process results in a bad process that is faster and more expensive to fix. You need to understand what you are really doing, involve the operational teams and optimise the process before implementing any technology.

Sources

• Statistics Portugal (INE) — Industrial Production Statistics and Structure of Portuguese Companies, sectoral data on footwear and textiles
• Portuguese Footwear, Components and Leather Goods Manufacturers' Association (APICCAPS) — Sectoral reports on Portuguese footwear
• Regulation (EU) 2024/1689 (AI Act) — Legal framework for automation and artificial intelligence systems in an industrial context
• Standard ISO/IEC 27001:2022 — Information security management in automation and ERP systems
• IAPMEI — Support Programme for the Digital Transformation of Portuguese Industrial SMEs