Qlik Sense in Portuguese industry: from spreadsheet to decision

In a knitwear factory in Vizela, the monthly production plan lives in a 14-tab Excel file that only Mr. Armando knows how to open without breaking the formulas. When Armando goes on holiday, the company goes blind. This scenario — the irreplaceable human analyst, embodied in a spreadsheet that no one else understands — is repeated in half the Portuguese industry I know.

The thesis of this article is uncomfortable: the problem with industrial BI in Portugal is not technical, it is one of ownership. Companies buy Qlik Sense thinking about pretty dashboards and discover that the real obstacle is letting go of the spreadsheet — because the spreadsheet gives power to whoever controls it. I will show you how Qlik Sense changes that game, what is defensible in numbers, and where projects derail.

1. The real operational problem: the spreadsheet as a hidden operating system

Most Portuguese industrial SMEs do not suffer from a lack of data. They suffer from trapped data. The ERP has the numbers, the shop floor has the notes, the sales team has its own Excel of margins, and the CFO has a third file that reconciles the three by hand, on Sunday night.

I call this the hidden operating system because, in fact, it is what runs the company. The ERP invoices, complies with the law, keeps the records — but the decision comes out of the spreadsheet. When an industrial director decides to subcontract one more workshop in Barcelos, or when the CFO decides to cut a collection reference, the number that supports that decision almost always came out of a manual tab. And that tab has no version control, no audit trail, and does not survive the departure of whoever maintains it.

The month-end ritual

Month-end in a garment factory near Famalicão sounds like this: the ERP exporting CSVs, the printer spitting out reports, and someone copying values from one tab to another because "the column J formula stopped matching the new VAT". The closing that should take two hours takes two days. And when the parent company — Inditex, Decathlon, Tom Tailor — asks for a report on delivery-deadline compliance per order, no one knows how to answer without reopening the same monstrous file.

The classic symptom: three people in the management meeting bring three different figures for "monthly sales". Not because they are lying — because each spreadsheet was built with different rules for cut-offs, returns, exchange rates. The business discussion turns into a formula audit.

This ritual has a cost that is rarely accounted for because it is diluted in the hours of competent people. A controller who dedicates two days a month to consolidating spreadsheets spends around 24 days a year — more than a month of qualified work — doing what an automated load does overnight, without error and without coffee. And that controller is, almost always, the person the company most needs to have thinking, not copying and pasting.

When the management meeting is arguing over which of the three Excel files is right, it has already lost the management meeting.

Why this persists (and it is not laziness)

The spreadsheet survives because it is flexible, free at first glance, and — the critical point — because it gives autonomy to whoever masters it. Mr. Armando does not hide the file out of malice. He hides it because that file is the proof that he is indispensable. Taking away the spreadsheet without giving back a leading role to the person who maintained it is mistake number one of the BI projects that fail in Portugal.

There is also a second, more structural reason. In Portuguese family businesses, the investment decision runs through the trio CEO + CFO + IT, and the IT manager is often a self-taught person with 15 years of business knowledge and no formal qualification. That professional built the spreadsheet over a decade, embedded in it business rules that are not written down anywhere, and distrusts — rightly — any consultant who turns up promising to replace it in a PowerPoint. Winning over this ally is half the project. Ignoring them is guaranteeing the passive sabotage that kills rollouts.

The silent costs of this architecture pile up:

• Decisions made on yesterday's data — or last week's — because consolidation is manual and slow.
• Formula errors that no one detects until the customer complains about a wrong invoice.
• Business knowledge locked in one person's head and in a file that does not survive their departure.
• Impossibility of cross-referencing production with margin, or stock with seasonality, because the data lives on islands.
• Zero traceability of who changed what and when — a problem when a parent company's audit asks how a given figure was reached.

According to INE, in 2025 only 53.7% of companies in Portugal used ERP. In other words: nearly half do not even have an integrated core on which to build analysis. In those companies, BI would work on scattered data — and that is why the right order matters, a topic I develop in How to build a data-driven organisation: the first step.

The continuity risk: when the spreadsheet is a person

It is worth distinguishing two types of risk that the spreadsheet creates. The first is operational — the number is wrong and no one notices. The second is one of continuity — the knowledge lives in one person. The second is the most serious and the least discussed in management meetings, because it is uncomfortable to admit that the company depends on an individual.

In a textile finishing factory in the Ave Valley, I witnessed a situation where the only person responsible for the costing spreadsheet for dyeing baths left for a competitor. For four months, the company was unable to reliably calculate the real margin of any dyeing order, because the rules for apportioning chemicals and energy were in a file that no one knew how to interpret. The cost of that blindness, in poorly informed pricing decisions, far exceeded what it would have cost to document the data model in good time.

2. What exactly is Qlik Sense in Portuguese industry

Qlik Sense is a business intelligence and data visualisation platform. Reduced to the essential: it takes data from various sources — ERP, payroll, shop-floor terminals, e-commerce, spreadsheets — loads them into its own engine, and allows them to be explored visually through interactive dashboards where each click recalculates the entire context.

The associative engine: the difference that competitors copied but did not match

What technically distinguishes Qlik is the in-memory associative engine. In a traditional report, when you filter by "customer = Decathlon", the system shows you that customer's data. In the associative model, Qlik also actively shows you what is not associated with that selection — which articles were never ordered by Decathlon, which months had zero activity. That ability to see the void, the negative, is where the business questions no one thought to ask are born.

This behaviour is not cosmetic. In a Felgueiras footwear collection with 800 to 1200 SKUs across three axes — colour, size, last — the difference between "show me what sold" and "show me what stayed put and why" is the difference between managing the collection and suffering it.

Most BI tools built on traditional query models — SQL-based behind each filter — force the analyst to know the question before asking it. You have to build the report to discover the pattern. The associative engine reverses this: you navigate freely through the data, and the pattern emerges from the exploration itself. For those working seasonal collections with hundreds of references, this freedom of exploration is not convenience — it is the only practical way to detect the reference that behaves anomalously in a universe too large for manual inspection.

A brief history — why the architecture matters

Qlik was born in Sweden in the 90s. QlikView, the first generation, was powerful but rigid — each application was a monolith built by specialists. Qlik Sense, launched from 2014 onwards, reversed the logic: dashboards that the business user themselves composes, self-service, with central governance. That change is the one that matters for a Portuguese SME, where there is rarely a dedicated team of analysts and the industrial director has to be able to manage on their own.

The distinction between self-service with central governance and self-service without governance is fundamental and frequently misunderstood. Without governance, each user builds their own version of the truth — and you are back to the problem of the three spreadsheets, now with prettier charts. With central governance, there is a single, validated data layer on which each person explores freely, but the definition of "net sales" or "contribution margin" is the same for everyone. It is this architecture that allows analysis to be decentralised without fragmenting the truth.

Terms you will encounter and what they mean

• Application (app) — the set of dashboards on one domain, for example "Production" or "Sales".
• Associative data model — the structure that links tables from different sources by common keys.
• Load script — the code that extracts and transforms the data before presenting it; this is where the business rules are resolved (returns, exchange rates, cut-offs).
• Set analysis — the syntax for calculating conditional metrics (sales for the same period in the previous year, for example).
• QVD — Qlik's native file for storing already-transformed intermediate data, which speeds up subsequent loads.
• Section access — the security mechanism that restricts what data each user profile can see.

For a practical explanation of what changes day to day, it is worth reading Real-time Business Intelligence with Qlik Sense.

Where Qlik ends and the ERP begins

A common confusion, especially in the decision-making trio of family businesses, is thinking that BI replaces ERP functions. It does not. Qlik does not invoice, does not issue delivery notes, does not communicate to SAF-T, does not do payroll. It consumes the data these systems produce. The boundary is clear: transactional systems record what happened and comply with the law; BI interprets what happened to inform what to do next. Buying Qlik expecting it to solve a certified invoicing problem is buying the wrong tool — that is what AT-certified software is for, not the analytical layer.

3. The landscape in Portugal today

Data-based management is still not universal in Portugal — far from it. The INE figures paint a sober picture.

What these numbers really say

That 45% of companies perform data analysis sounds reasonable until you cross-reference it with the 53.7% that have ERP. The uncomfortable conclusion: there are companies "performing data analysis" without an integrated core underneath — that is, analysis on spreadsheets. It is BI pretending. The growth of 6.4 points in two years shows appetite, but the foundation is still missing in nearly half of the business fabric.

There is a nuance of size here that the aggregate numbers hide. The adoption of analytical tools and ERP is heavily concentrated in larger companies. Micro and small companies — which make up the overwhelming majority of the Portuguese industrial fabric, especially in textile and footwear subcontracting — are significantly below these averages. When you look at a 25-person garment factory in Guimarães that works exclusively for one parent company, the probability of having structured data analysis drops sharply. It is precisely there that the digital divide deepens and where European reshoring will demand capabilities that today do not exist.

Data analysis without an integrated ERP underneath is a spreadsheet with aspirations. The BI engine only shines if the fuel is clean.

The weight of the industrial sector

The Portuguese metallurgical and metalworking sector has more than 23,000 companies and around 250,000 people employed, according to AIMMAP. It is a huge universe of moulds, plastic injection and one-off parts in the Aveiro–Marinha Grande corridor, where each run has costs and margins that vary radically between orders. Without analysis, it is managed by gut feeling. And gut feeling, in a one-off part mould, is expensive when it is wrong.

In textiles and clothing, the Portuguese chain is one of the most complete in Europe — from spinning to finished product — and rests on thousands of companies concentrated in the North, many of them family-structured and with an ageing workforce. The simultaneous pressure from fast fashion, which demands ever shorter response times, and from the EU Strategy for Sustainable and Circular Textiles, which demands batch traceability and a digital product passport, turns data analysis from a luxury into a prerequisite for survival. A brand that asks for proof of origin and composition per batch does not accept "it is somewhere in the Excel" as an answer.

Commerce, for its part, generated €201.8 billion in turnover in 2024, with retail growing 4.7% (INE). Volume is not margin — and it is exactly that gap between invoicing and earning that a good profitability-per-article dashboard makes visible.

Funding and the bottleneck of technical reports

Digitalisation and BI projects fit into instruments such as PT2030, the PRR, COMPETE 2030 and Norte 2030. The application is approved relatively often; where projects get stuck is in the technical execution report — the demonstration that the investment produced the promised result. A poorly scoped BI project is a natural candidate to get stuck in that phase, because "we improved decision-making" is not an auditable metric and "we reduced closing time from 2 days to 3 hours" is.

The lesson we take from following applications is operational: design the project starting from the execution metric, not from the technology. Before writing the application, time the current closing time, count the invoicing errors of the last year, measure the OEE of a line. These baseline numbers are what makes the technical report defensible eighteen months later, when the intermediate body's officer asks for proof of the result. Without a baseline measured on day zero, there is no way to demonstrate the gain — and the funding is at risk of being returned.

The PME Líder seal and digital maturity

IAPMEI awards the PME Líder and PME Excelência seals to companies with robust performance and risk profiles. Increasingly, digital maturity enters the risk assessment that underpins these distinctions, because a company that manages blindly is, by definition, riskier. A governed analytical layer is not only operational efficiency — it is a sign of management quality that financial institutions and international clients read with ever more attention in due-diligence processes.

4. The implementation models: four approaches and their trade-offs

There is no single way to implement Qlik Sense. There are four patterns I see repeatedly, each with a distinct risk and cost profile.

The mistake of skipping steps

The enthusiastic CEO's temptation is to start with the governed platform. It is the equivalent of buying a five-axis CNC lathe for someone who does not yet know how to sharpen a hand tool. The sequence that works is to gain credibility with a tactical dashboard that solves a real and visible pain point — then, with the confidence gained, scale up. Adoption is measured in people who let go of Excel, not in apps published.

The most common failure pattern we have documented: an enthusiastic company buys licensing for the complete platform, commissions the implementation of ten dashboards simultaneously, and six months later has ten panels no one opens because none solved the pain point that hurt most. The energy dispersed, the confidence evaporated, and Excel continued to run the company. The antidote is the discipline of focus: one dashboard, one pain point, one measured result, and only then the next.

The real-time case

The model with shop-floor capture deserves a note. Connecting Qlik to a KORA Productivity solution — which captures production on industrial terminals — transforms TPM from a manual concept into a panel where the production director sees the OEE per machine by the hour. With protocols such as MQTT for equipment telemetry, the data reaches the dashboard before the operator finishes the shift. Here, integration is often done with an iPaaS layer orchestrating the sources.

The difference between OEE by the month and OEE by the hour is not one of degree, it is one of nature. Monthly OEE serves for reporting and for after-the-fact self-flagellation: you discover, on the 5th of the following month, that the injection line was below acceptable availability. There is nothing left to do. Hourly OEE, visible on the panel while the shift is running, allows the line supervisor to react before the problem crystallises into an unrecoverable loss. It is the difference between the autopsy and the timely diagnosis.

OEE calculated by the month is history. OEE calculated by the hour is management.

The choice between scheduled load and near-real time

A technical decision with real operational consequences: not all analysis needs real time. A financial dashboard of profitability per customer works perfectly with a nightly load — yesterday's data is enough for margin and credit decisions. An OEE dashboard for a critical line needs a refresh in minutes. Confusing the two cases costs money in both directions: real-time infrastructure where it is not needed is waste; a nightly load where immediate reaction is needed is useless. The practical rule: align the refresh frequency with the frequency of the decision the dashboard supports.

5. How to assess whether your company needs it

Not every company needs Qlik Sense tomorrow. There is an honest test to know whether you are mature — and another to know whether you are running from the wrong problem.

The signs that you need it (and now)

• The management meeting spends more time validating numbers than deciding on them.
• The month-end closing depends on one person and a file that no one else understands.
• You cannot answer within five minutes "what was the real margin of this customer in the last quarter?".
• The parent company asks for compliance reports and the answer takes days.
• You have ERP, but you export everything to Excel to "be able to work the data".
• You discover profitability problems too late to fix them — the order was already delivered with a negative margin.

The signs that it is not yet the time

If you do not have ERP, or if the ERP is out of date and the source of the data is chaotic, BI will amplify the chaos with colourful charts. First you tidy the house. A pretty dashboard over wrong data is more dangerous than no Excel at all, because it looks reliable. The order is: consolidated ERP MULTI, clean data, and only then the analytical layer.

A second sign of "not yet": if the company has no one willing to take ownership of the tool. BI is not a system that runs by itself on a server — it needs a human owner who asks questions, who refines dashboards, who brings the findings to the management meeting. If there is no such profile, and no intention to train one, buying Qlik is buying shelfware. Technology without human ownership is the most expensive way to change nothing.

Step by step: a five-stage readiness diagnosis

1. Inventory the sources of truth. List all the files and systems where the numbers management uses to decide live. If there are more than five and three are spreadsheets, you have a dispersion problem.
2. Identify the human owner of each critical spreadsheet. Note who knows how to open each file without breaking it. Each name on this list is a continuity risk and a potential ally in change.
3. Measure the real closing time. Time how many hours a month are spent consolidating data manually. This is your defensible return figure for a PT2030 application.
4. Choose a single, measurable pain point. Profitability per article, deadline compliance, OEE of a line. One. The one that hurts most and whose result can be measured before and after.
5. Validate the maturity of the source. Confirm that the data for that pain point exists in the ERP reliably. If it lives only in manual Excel, the first project is to integrate it, not to visualise it.

This exercise takes a week and avoids the most expensive mistake: buying a platform before knowing what you want to ask. Culture counts as much as technology — the subject of How data-driven culture changes decision-making.

The silent-cost table

To convert the diagnosis into a number, it helps to quantify what the current architecture costs. Not in licences — in hours and in errors.

Excel has no monthly invoice. It has an annual invoice disguised as hours of competent people doing machine work.

6. What to choose and why: decision by company size

The right recommendation depends on scale, data maturity and product complexity. An honest matrix:

For the micro and small: less is more

A 25-person garment factory subcontracting for a parent company does not need a governed analytical platform — it needs to answer two or three survival questions. Am I meeting the deadlines I was given? What is my real efficiency per operation? Am I gaining or losing on each type of piece? A single tactical dashboard on the ERP answers this, and the real risk is not technical — it is the time of whoever has to maintain it. In a small structure, the person doing the BI is the same person doing everything else. The project has to be designed to consume minutes, not hours, of that person's attention.

For the medium industrial: the sweet spot

A typical textile factory in the textile sector in the Ave Valley, with around 80 employees, almost always fits the second profile. It has ERP, it has production data recording, and the pain point splits between "I don't know the real margin per order" and "I don't know how much time I lose in setup between colour batches". Here, Qlik on the ERP solves the first pain point in weeks; the second requires integrating shop-floor capture and is phase two.

The typical mistake in this profile is to want to solve both pain points at once. The margin per order depends on data that is already in the ERP — sales, material costs, allocated labour — and is tackleable immediately. The setup time between colour batches depends on fine capture on the shop floor, which often does not yet exist or exists on paper. Mixing the two phases delays the first, which was the quick win capable of buying credibility for the second.

For footwear: the axes problem

In the footwear of Felgueiras, the complexity is not in the sales volume but in the dimensionality. Analysing a collection means cross-referencing colour × size × last × channel × campaign. Generalist ERPs flatten this into flat reports. Qlik's associative model was made to navigate precisely these cubes — and to reveal the SKU that no one ordered in the entire August showroom. On intelligent pricing in these scenarios, Dynamic Pricing gains another factual basis.

There is a seasonal rhythm that footwear analysis has to respect: international buyers visit twice a year — men's footwear in August, women's in February. The decision on which references to keep, cut or reinforce in the next collection is made in a tight window right after each visit. A dashboard that answers within minutes "which models attracted interest but did not convert into an order" is worth, in that window, more than weeks of analysis out of the moment. Seasonality turns the speed of analytical response into a direct competitive advantage.

For distribution: the warehouse cannot stop

In a distributor in the Lousada–Paços corridor, the warehouse manager will fight any rollout that takes them off the radar for more than two hours. The operational lesson: warehouse BI is built on data that is already captured by the operation — from a solution such as KORA Inventory Suite — without asking for new work from whoever does the picking. The dashboard consumes; it does not demand. Cross-referencing this with Just-in-Time and 5S practices gives the warehouse manager numbers they themselves can defend.

The warehouse manager's resistance is not stubbornness — it is operational common sense. Whoever manages a 20,000 m² warehouse with picking and cross-docking knows that any stoppage cascades through the day's delivery chain. The approach that works is to make the BI work from the telemetry the operation already produces: barcode readings, stock movements, dispatch times. No new forms, no asking the operator to enter extra data. The dashboard reveals productivity patterns, dock bottlenecks and stock deviations from data that already exists. When the warehouse manager realises that the panel defends them rather than watching them, they become an ally.

For retail: volume is not margin

A regional food retail chain with, say, 22 stores lives a specific tension: it invoices a lot, but the margin hides in a multitude of articles, promotions, shrinkage and stockouts. The POS generates a deluge of transactions which, without analysis, is noise. A profitability dashboard per article and per store, cross-referenced with MAXIRETAIL data, reveals where the promotion destroyed margin without generating incremental traffic, and where the silent stockout is sending customers to the competition. The obligation of software certification and communication to the AT already produces the data; BI gives it management meaning.

7. Regulatory framework and applicable compliance

BI does not live in a legal vacuum. When it cross-references invoicing, employee and customer data in a single system, it inherits obligations that many forget until the first audit.

Invoicing data and SAF-T

The sources that feed Qlik include invoicing data subject to DL 28/2019 — electronic invoicing, ATCUD, AT-certified software. BI does not replace the certified system; it consumes its data. But if the sales dashboard presents values that do not reconcile with the SAF-T communicated monthly (Portaria 195/2020), you have a problem of internal credibility before you have a tax problem.

The reconciliation between the dashboard and the SAF-T must be a formal step of the project, not an improvised check. If the panel says March sales were X and the SAF-T communicated to the AT says Y, someone will ask which is right — and the answer "it depends on how the returns are counted" undermines confidence in the whole platform. Defining the business rules in the load script so that they match the certified accounting to the cent is what separates a credible dashboard from just another spreadsheet with aspirations.

GDPR and people's data

An HR dashboard that cross-references absenteeism, turnover and performance — of the kind a solution such as pplPortal feeds — handles sensitive personal data. The GDPR and Law 58/2019 require minimisation and access control by profile. No one in sales should see the individual payroll on a panel. Qlik's access governance exists precisely for this; using it badly is a fine waiting to happen. Employee Engagement is measured, but it is measured with respect for privacy.

The principle of minimisation has a direct practical translation in BI: an HR dashboard for top management should show aggregate rates — absenteeism per department, turnover per unit — and not the individual record of each employee. The temptation to "be able to drill down to the person's detail" is precisely what minimisation prohibits when there is no legal basis for that level of detail. Qlik's section access allows the same app to be built with different views according to the profile: the HR director sees what they have to see, the industrial director sees only the aggregate of their unit.

Cybersecurity: NIS2 and ISO 27001

The NIS2 Directive (EU 2022/2555), transposed in Portugal, extends cybersecurity obligations to more sectors, including manufacturing considered critical. A BI server concentrates data from the whole company in a single point — it is, by definition, a high-value target. Hosting it with ISO 27001 controls and perimeter protection is not a luxury. It is the subject of INFOS Security: perimeter protection and SOC and of the silent risks of cybersecurity in the company.

It is worth measuring the exposure clearly: a compromised spreadsheet exposes one data domain; a compromised BI server exposes the entire company, because that is where the domains converge. This concentration is the strength of BI and its risk. The tightening of NIS2 obligations, with its requirements for incident notification within short deadlines and management accountability, turns the decision of where and how to host the analytical platform into a board decision, not an IT department one.

A BI server is the company's safe disguised as a pretty tool. Protect it as such.

AI and the AI Act

When predictive models are added — demand forecasting, anomaly detection — the EU Regulation 2024/1689 (AI Act) comes into play, classifying AI systems by risk. Most cases of industrial predictive BI fall into low or limited risk, but it is worth documenting the classification. On where AI adds real predictability, it is worth reading AI applied at INFOS: turning data into automated decisions.

The practical distinction that matters: a dashboard that shows the past and present is not an AI system in the sense of the regulation. It becomes an AI system when it incorporates models that make inferences — forecasting the demand of the next collection, flagging an order at high risk of delay, detecting a fraud pattern. At that point, even at low risk, it is worth having documented the logic of the model, the data that trains it and the known limitations. Not for bureaucracy, but because a model that decides about people or about credit without auditable logic is a liability waiting to happen.

8. How INFOS approaches this

We work with Qlik Sense as an analytical layer on top of the core we have implemented in Portuguese factories for more than three decades. This changes the conversation: we do not turn up with a generic dashboard, we turn up knowing the data model of textiles, of footwear, of distribution — because we built it in ERP MULTI over thousands of projects in the industrial sector.

The practical consequence is the speed of start-up. When the data lives in an ERP we know, the load script does not start from scratch — the business rules for returns, exchange rates and cut-offs per collection are already mapped. For groups with multiple sites, consolidation relies on Multi Connect, and shop-floor capture on terminals that already talk to the BI.

This point deserves emphasis because it is where most generic BI projects get bogged down. Half the cost and timeline of a BI project is in data modelling — in understanding how "sales" is defined in this company, how returns are handled, how indirect costs are apportioned per order, how the articles of a collection that changes reference between seasons are counted. Whoever turns up with a tool and without knowledge of the sector's data model spends months rediscovering what is already resolved in the vertical ERP. We start out with that map drawn.

We do not sell a platform as an end. We sell the passage from the spreadsheet to the decision — and, honestly, the difficult part is rarely technical. It is convincing Mr. Armando that letting go of the file makes him more valuable, not dispensable. The projects that ignore this human dimension are the ones we write up, years later, as a lesson.

The dashboard is the easy part. The difficult part is the internal politics of who loses and who gains power when the truth becomes shared.

What we learned from the projects that went wrong

Honesty compels us to mention the failures. We have seen projects derail for three recurring reasons. The first: overly ambitious scope from the outset, ten dashboards instead of one, dispersed energy and no clear win. The second: source data that was assumed clean and was not — the BI project turned, halfway through, into an unbudgeted data-cleansing project, with all the friction that generates. The third, the most insidious: the lack of a human owner with a mandate and time to take ownership of the tool, leaving the dashboard to gather dust while Excel continued to command.

From these lessons came our discipline: start small, validate the quality of the source before promising deadlines, and identify the human owner before the first app is built. None of these lessons is technical. All are about change management — and that is why the most valuable part of a BI project happens in the conversations, not in the code.

9. 30/60/90-day roadmap

An actionable plan with verifiable milestones — nothing about "improving decision-making", everything in numbers that a PT2030 application or a management meeting can audit.

Days 1–30: foundation and proof

• Complete the five-stage readiness diagnosis from section 5, with the closing time timed as the baseline figure.
• Choose a single, measurable pain point and map its sources in the ERP.
• Build the first tactical dashboard on that pain point, with real data — not a mock-up.
• Verifiable milestone: management can answer the target question in under five minutes, without opening Excel.

Days 31–60: integration and confidence

• Automate the data refresh from the ERP — the end of manual loads.
• Reconcile the dashboard values with the SAF-T and with the accounting close, to the cent.
• Train the human owner of the old critical spreadsheet as the first advanced user — give them back a leading role.
• Verifiable milestone: the monthly consolidation time falls to a fraction of the baseline value measured on day 1.

Days 61–90: scale and governance

• Publish the second app on the next pain point — typically production or profitability per customer.
• Define access profiles: who sees margins, who sees HR, who sees what — GDPR compliance by design.
• Configure automatic alerts on critical deviations (OEE below the threshold, negative margin per order).
• Verifiable milestone: at least two management decisions of the quarter were made from the dashboard, documented for the technical funding report.

The anti-milestones: what should not happen

As important as the milestones is watching for the signs of derailment. If after 30 days there is still no dashboard with real data and everything remains in the "requirements gathering" phase, the project has entered the swamp of infinite analysis. If after 60 days the values do not reconcile with the accounts and the answer is "we'll fine-tune it later", credibility is already bleeding. If after 90 days the human owner of the old spreadsheet is still keeping the spreadsheet in parallel "just in case", the change did not take. Each of these anti-milestones is a signal to stop and correct, not to accelerate.

Set targets with OKR for each phase — not because it sounds modern, but because a PT2030 execution report survives better with measured key objectives than with adjectives. And if, at the end of the 90 days, Mr. Armando opens the dashboard before Excel of his own initiative, you have won. That is the true indicator that the spreadsheet has ceased to be your company's hidden operating system. On turning that data into bottom-line results, I go deeper in Turn data into profit with Qlik Sense and in strategic planning with BI.

The spreadsheet does not die by decree. It dies when whoever maintained it discovers that the dashboard gives back the time the file stole from them — and when the management meeting, for the first time, discusses the business instead of discussing which of the three numbers is right. That is the day the BI stopped being an IT project and became the way the company thinks.

Sources

• INE — Survey on the Use of Information and Communication Technologies in Companies, 2025 (data analysis, ERP, cloud).
• INE — Commerce Statistics, 2024 (turnover of commerce and retail).
• AIMMAP / Metal Portugal

Frequently asked questions

What exactly is Qlik Sense and how does it work in industry?

Qlik Sense is a business intelligence platform that integrates data from various sources (ERP, spreadsheets, databases) into a single analytical layer. It allows the creation of interactive dashboards where users explore data in real time, without depending on static reports. In Portuguese industry, it solves the problem of scattered spreadsheets by centralising the information.

What is the difference between using Excel and implementing Qlik Sense?

Excel is flexible but fragile: data in silos, no version control, dependence on people. Qlik Sense centralises everything in a single base, with a complete audit of who changed what and when. A controller stops spending 24 days a year consolidating spreadsheets and starts using that time on strategic analysis. The data is always up to date and accessible to whoever needs it.

How much does it cost to implement Qlik Sense in an industrial SME?

The cost varies with the complexity of the data and the number of users. Licences start at around €3,000–5,000 per year per named user. The implementation (data integration, dashboard design, training) typically costs €15,000–40,000. The return appears in 6–12 months through the reduction of manual hours and faster decisions.

Why do BI projects fail in Portugal?

The main reason is not technical: it is political. Whoever controls the spreadsheet loses power when that spreadsheet disappears. If the project does not reconvert that leading role into real analysis, there is passive sabotage. The second mistake is ignoring the internal IT person who built the spreadsheet — that professional is a critical ally, not an obstacle. Third: implementing without documenting the business rules that live in the spreadsheet.

How does Qlik Sense solve the problem of the indispensable person?

By documenting and automating the knowledge that lives in a spreadsheet. The apportionment rules, the margin formulas, the returns cut-offs — everything is recorded in Qlik's data model. When that person leaves, the knowledge stays. In addition, multiple users can access the same data with confidence, eliminating the dependence on a "Mr. Armando".

Does Qlik Sense work if the company does not have ERP?

Yes, but with limitations. Qlik can integrate data from various sources (Excel, databases, APIs, CSV files). Without ERP, the data remains scattered and the quality depends on the source. The ideal is to have an integrated core (ERP or data warehouse) to build reliable analysis. If one does not exist, Qlik helps to organise the chaos, but it does not replace an adequate data architecture.

How long does it take to see results with Qlik Sense?

The first operational dashboards come out in 4–8 weeks. Faster decisions start to appear immediately — an industrial director stops waiting for month-end consolidations. The full ROI (reduction of manual hours, better decision quality) materialises between 6 and 12 months, depending on the initial complexity and the adoption by users.