The pretty dashboard doesn't pay wages. We see managers buying BI to "get visibility" and, six months later, nobody opens the panel. The problem isn't the tool — it's that the panel doesn't answer any decision someone actually makes on a Monday morning. The uncomfortable thesis: most business intelligence projects in Portuguese SMEs fail not for lack of data, but for a surplus of ownerless panels. A KPI without a person who will go and defend it in a meeting is decoration. Start with the decision, not with the chart. This guide gives you a five-step method to turn scattered data into decisions that move the margin, and a matrix to choose which indicator is worth a screen.
There's one figure that frames everything else: only 53.7% of companies in Portugal used ERP in 2025 (INE, 2025). Almost half the business fabric still decides on spreadsheets that disagree with one another. Stacking BI on top of that unstable floor doesn't give visibility — it gives confusion with better design.
What you need before you start
Before opening any tool, you need a data core that speaks with a single voice — ideally an MULTI ERP feeding sales, stock and production without manual exports in between. You need a short list, three to five, of recurring decisions that today are taken on intuition, and to know who owns each one. An owner is a name, not a department. "Production" doesn't lose sleep over any number; António, who chases up late orders, does.
You also need a written definition of each metric. "Margin" means different things to the CFO and to the sales rep — one counts early-payment discounts, the other doesn't, and the meeting is spent arguing over whose number is the right one. And you need a realistic horizon: the first useful panel comes out in weeks, not days, but also not in an eight-month project that nobody can stomach.
Note the contrast with the market: in 2025, only 45% of companies in Portugal did data analysis, even though that is 6.4 points more than in 2023 (INE, 2025). Data-based management is not universal — it's an advantage still to be won, not a given.
Step 1 — Write the question before the chart
Don't open the tool. Open a document and write the question the panel has to answer. "Which footwear references are destroying margin this season?" is a question. "Sales dashboard" is not.
In a garment maker in the North that subcontracts for parent houses, the right question is rarely "how much did we sell". It's "which orders are running late and how much does each day of delay cost in penalties". When Inditex or Decathlon apply a penalty for delay, that amount comes straight off the order's margin — and that's what the panel has to show before the damage is done. The question defines the fields, the granularity and the update frequency. Without a question, you choose granularity at random and then redo everything.
- One question, one panel. Don't stack them.
- The question has to have an owner who acts on the answer.
- If nobody changes behaviour with the answer, cut the question.
Step 2 — Connect the sources without duplicating truths
The classic mistake, the one we see in nine projects out of ten: export from the ERP to Excel, process by hand, paste into the BI. Two weeks later there are three versions of the same sale circulating by email, and the meeting becomes a spreadsheet tribunal. Qlik Sense connects to the INFOS ecosystem — MULTI, MAXIRETAIL and KORA Productivity — and reads the data at the source, without intermediate copies that age between the click and the meeting.
Qlik's associative model lets the user click on a customer and see, at the same time, which products they bought, at what margin and what delays they had — without asking in-house IT for a new query. It's that free cross-referencing that separates real BI from a static PDF report that is born old.
Step 3 — Choose the KPIs that are worth a screen
Not every metric deserves a panel. Use this matrix to decide. If a candidate doesn't score in three of the four columns, file it away before it takes up space and attention.
| Candidate KPI | Has an owner? | Can you act within 48h? | Links to €? | Is the data reliable? |
|---|---|---|---|---|
| Margin per reference | Yes (sales) | Yes | Direct | Yes |
| OEE per line | Yes (production) | Yes | Indirect | Depends on capture |
| Days of delay per order | Yes (planning) | Yes | Direct (penalties) | Yes |
| "Total sales for the month" | No | No | Vague | Yes |
The last row fails. Total sales for the month is a vanity number — it impresses in the report, it doesn't change any operational decision. Note OEE per line: it only qualifies if the "reliable data" column holds, and that depends on there being real production capture on the shop floor, not on someone jotting in a notebook at the end of the shift. Without real-time capture, OEE on the panel is fiction with decimal places. To go deeper on the choice, see our operational guide to KPIs for industrial directors and the reading on where the CFO and the COO disagree.
Step 4 — Design for the meeting, not for the wall
A panel that needs explaining has failed. The five-second rule: whoever opens it has to understand the state in five seconds and know what to do next. Colour only for exceptions — green is noise if everything is green. The eye looks for what's burning, not for what's burning evenly.
A KPI without an owner is decoration. Before designing the chart, write the name of the person who will lose sleep over that number.
In the textile industry of the Vale do Ave, the decision-makers who use BI every day have panels with half a dozen indicators, not thirty. Thirty indicators is what you design to impress whoever pays the licence; six is what you open on a Monday morning with coffee in hand. We explored that design in dashboards decision-makers actually use.
Step 5 — Close the loop with governance
Self-service BI without rules turns into anarchy of numbers. Qlik Sense gives the user autonomy to explore without breaking data governance — who sees what, which definition of "margin" is the official one, who can publish a panel for the whole organisation. It's the difference between a warehouse with marked aisles and one where everyone stacks wherever they fancy.
Without this, each department creates its own truth and the meeting spends forty minutes arguing over whose number is right instead of deciding. A data-driven culture is built with a single source, not with more screens.
- One official definition per metric, written and versioned.
- One person responsible for publishing panels for the whole company.
- Quarterly review: which panels did nobody open? Cut them.
Common mistakes and how to avoid them
The first and most expensive: buying BI before having the ERP talking properly. Stabilise the integrated core first, because BI doesn't fix dirty data — it amplifies it, and now in colour. The second is the orphan panel: assign a name to each panel in the first week or file it away. The third is vanity metrics — if nobody acts within 48 hours on the answer, the KPI doesn't get onto the screen.
There are two more subtle mistakes. Ignoring the TCO: count maintenance, training and the time of whoever creates panels, not just the licence — often the hidden cost is someone's half-day a week maintaining reports nobody asked for. And training everyone at the same time: start with three users who decide, not with thirty who consult. If the warehouse chief on the Lousada–Paços aisle doesn't use the panel within the shift, without leaving the radio for more than two minutes, you haven't trained him — you've just pulled him off the job.
What separates those who invest from those who collect
Choose a decision you currently take blind, write the question, and build a single panel to answer it. One KPI with an owner that changes a behaviour is worth more than thirty charts nobody opens. That's the dividing line, and it has nothing to do with the size of the budget: on one side are the companies that know exactly which decision each screen serves; on the other, those that bought visibility and got decoration.
Sources
- INE — Survey on the Use of Information and Communication Technologies in Enterprises, 2025 (adoption of ERP, big data/analytics and cloud in Portugal).
- INE — Trade Statistics, 2024 (trade turnover).
- Qlik — Official Qlik Sense documentation (associative model and data governance).
Frequently asked questions
Does Qlik Sense work without an implemented ERP?
It works technically, but with limitations. If the data is scattered across spreadsheets and disconnected systems, Qlik Sense will connect to those sources, but it will inherit the same problem: multiple versions of the truth. The ideal is to have an ERP (MULTI, MAXIRETAIL or KORA) feeding the data without manual exports in between. Without that solid base, BI becomes confusion with better design.
How long does it take to implement a useful panel?
The first useful panel comes out in weeks, not in days nor in eight months. The speed depends on having a clear question, accessible data and an identified owner who will act on the answer. Projects that last more than two months tend to fail because nobody can bear the wait and the panel ends up forgotten on the shelf.
What's the difference between a KPI and a vanity indicator?
A true KPI has an owner (a name, not a department), allows you to act within 48 hours, links directly to money or costs, and the data is reliable. "Total sales for the month" is vanity — it impresses in the report, but nobody changes behaviour with that number. If an indicator doesn't pass three of the four columns of the matrix, it should be filed away before it takes up space.
Does Qlik Sense replace an ERP?
No. Qlik Sense is an analysis tool that reads data from an ERP or from other sources. The ERP is the system that records the transactions (sales, stock, production). Qlik Sense turns that data into decisions. Without an ERP, Qlik Sense is dependent on manual exports and loses its main advantage: real-time reading of the single source of truth.
How do you stop a BI panel being forgotten after six months?
Three rules: first, start with the decision, not with the chart — write the question before opening the tool. Second, choose only KPIs with an identified owner who will defend the number in a meeting. Third, design for the Monday-morning meeting, not to impress — six indicators, not thirty. A panel without an owner is decoration.
What data do I need to have organised before implementing BI?
You need: a data core that speaks with a single voice (ideally an ERP); a short list of three to five recurring decisions that today are taken on intuition; the name of whoever owns each decision; and written definitions of each metric (because "margin" means different things to the CFO and to the sales rep). Without this, BI will amplify the confusion.
Does Qlik Sense's associative model offer an advantage over other BI tools?
Yes. It lets the user click on a customer and see simultaneously which products they bought, at what margin and what delays they had — without asking IT for a new query. That free cross-referencing of data, without the need for static reports, is what separates real BI from a PDF that is born old. It's particularly useful in SMEs where agility in decisions is critical.