A "real-time" dashboard that only refreshes at midnight isn't real time — it's a report with better lighting. And this is what many managers buy without realising. At the end of this guide you'll find a checklist to distinguish a true operational panel from a decoration, plus a matrix for choosing the 6 to 8 KPIs worth a screen on the wall.
The position I defend, after installing BI in footwear factories, garment makers and distribution centres in the North: the problem with real-time BI is almost never Qlik, Power BI or the tool. It's the latency of the upstream data. If production logging enters the system once per shift, no dashboard turns it into real time. Before choosing the colour of the chart, fix the source. And there's a raw proof of this on the ground in Portugal: only 53.7% of companies in Portugal used ERP in 2025 (INE, 2025). Without an integrated core, BI runs on scattered Excel sheets — and no panel can save that.
What you need before you start
The question I ask in the first meeting isn't "which charts do you want". It's "where does each number come from and how often". The answer separates the projects that live from those that die in the third month. Before opening Qlik Sense, put your house in order.
- An ERP with sales, stock and production data in the same place — the ERP MULTI or equivalent.
- A real data entry frequency defined per source: per shift, per hour, per second.
- A person responsible for each indicator — not "management", a person with a name.
- A clear decision: what action changes when the number changes? If none, it's not a KPI.
- Machines or terminals that log production without manual intervention, when the goal is OEE.
- Defined access — who sees margins, who sees salaries, who sees everything.
Step 1 — Separate "real time" from "useful time"
Not every indicator needs seconds. The cadence of the data has to match the cadence of the decision. The warehouse manager in the Lousada/Paços corridor decides picking by the minute and won't let go of the radio because of it; the CFO reviews margin once a week. Forcing real time where it doesn't serve overloads the system and tires whoever's looking. A panel that flickers every second with a number that only matters monthly is expensive noise.
| Indicator | Decision cadence | Latency required |
|---|---|---|
| Line OEE | Per shift / per minute | Seconds to minutes |
| Critical stock level | Throughout the day | Minutes |
| Overdue orders | Daily | Hours |
| Margin per customer | Weekly / monthly | Daily is enough |
| Staff turnover | Monthly | Weekly is enough |
Classify each indicator in this grid before you put it on a screen and refuse real time for anything decided weekly. Latency is an engineering choice, not a month-end vanity.
Step 2 — Choose 6 to 8 KPIs, not 40
The most frequent mistake we see in projects: the first dashboard arrives with 40 metrics because no one wanted to cut theirs. Each manager defends their number the way one defends a seat at the table. Nobody looks at 40 numbers. An operational panel lives or dies on the first column the eye meets — and the eye finds three, at most four, before giving up.
Use this matrix. Score each KPI candidate from 1 to 5 on each criterion. Only those totalling 16 or more get in.
| Criterion | Question |
|---|---|
| Actionability | Does anyone change anything when this number moves? |
| Reliability | Is the source automatic or does it depend on someone typing correctly? |
| Frequency | Is it worth looking at more than once a day? |
| Comparability | Does it have a target or history to know if it's good or bad? |
A detail the manuals don't mention: the criterion that fails the most KPIs is reliability, not actionability. In a garment maker in the Vale do Ave, almost all the "interesting" indicators fell because the source was manual and arrived truncated at the end of the shift. The number was useful on paper and useless on the screen. To choose the right indicators by function, our guide to KPIs for industrial directors has lists by area that save half a meeting of discussion.
Step 3 — Connect the sources to the associative model
Here's the concrete advantage of Qlik Sense over a tabular report: the associative model. Click on a footwear reference and the whole panel reacts — sales, stock, manufacturing orders, customer, colour, size. You don't need to know the question in advance. Follow the data wherever it takes you.
In a footwear factory in Felgueiras, a sample collection has 800 to 1,200 SKUs across three axes — colour, size, last. International buyers appear twice a year, men in August, women in February, and they want to cross-reference everything in front of them. A fixed report can't cope with that combinatorics. An associative model can: filter the last, watch the rest react, and answer the question the buyer has just invented.
A dashboard doesn't answer the questions you were already asking. It's there to ask the questions you didn't yet know you had.
Qlik Sense integrates natively with the ERP MULTI, MAXIRETAIL and the KORA suite — inside the INFOS house, without fragile bridges snapping every time a version changes. For per-second production capture, connect KORA Productivity: without automatic logging on the shop-floor terminals, the OEE on the screen is fiction with good typography. And grant autonomy with Self-Service BI without opening the door to those who shouldn't see margins.
Step 4 — Govern access before publishing
Self-service doesn't mean everyone sees everything. The line logger doesn't need the margin per customer; the salesperson doesn't need the salaries. The GDPR and Law 58/2019 make this an obligation, not goodwill — staff data must have restricted and logged access.
Define access profiles by function and not by person — when someone changes post, they change profile, and you're not chasing loose permissions. Log who accesses sensitive data: HR, pay, appraisal. And document the source of each number in the panel itself, with the table name and the time of the last update. That's what ends the end-of-meeting argument — "your Excel says otherwise" — before it begins.
Common mistakes and how to avoid them
- Pretty dashboard, old data. The chart refreshes but the source only comes in per shift. Remedy: audit the real latency of each source before promising real time.
- KPI without an owner. The number has been red for three weeks and no one acts. Remedy: each indicator has a person with a name and an associated action.
- Real time where it doesn't serve. Monthly margin on a per-second panel just burns resources. Remedy: use the grid from Step 1.
- A metric nobody trusts. If the source is manual and fails half the time, the panel dies at birth. Remedy: automate the capture before showing it.
- Confusing the tool with the culture. Buying Qlik doesn't make a company data-driven. Remedy: start with the first decision you'll now make with data, not with the software.
For the strategic framing — when data really replaces the founder's intuition — it's worth reading strategic planning with BI and the first step towards a data-driven organisation.
The next step
Choose a decision you make in the dark today — just one — and build the minimal panel that lights it up. If you get that right, the rest of Qlik Sense follows naturally. What doesn't scale is the 40-number panel that no one opens after the second week. A good dashboard isn't the one that shows the most — it's the one that makes someone, on the afternoon shift, change a decision they made yesterday out of habit.
Sources
- INE — Survey on the Use of Information and Communication Technologies in Enterprises, 2025 (use of ERP, data analysis and cloud in Portugal).
- INE — Trade Statistics, 2024 (trade turnover).
- Regulation (EU) 2016/679 (GDPR) and Law no. 58/2019, of 8 August (national implementation of the GDPR).
- Qlik — Official Qlik Sense documentation (associative model and data governance).
Frequently asked questions
Does Qlik Sense refresh data in true real time?
Qlik Sense is a visualisation tool. Real time depends on the latency of the upstream data — if production logging enters the system once per shift, no dashboard turns it into real time. Fix the source before choosing the colour of the chart. Without an integrated ERP, BI runs on scattered sheets and loses credibility.
How many KPIs should I put on an operational dashboard?
Between 6 and 8 KPIs. A panel with 40 metrics is one nobody looks at. The eye finds three or four numbers before giving up. Use a matrix of actionability, reliability, frequency and comparability. Score each candidate from 1 to 5 — only those totalling 16 or more get in.
What is the biggest mistake when implementing real-time BI?
Forcing real time where it doesn't serve. Not every indicator needs seconds. The warehouse manager decides by the minute; the CFO reviews margin once a week. A panel that flickers every second with a number that only matters monthly is expensive noise. Align latency with the cadence of the decision.
How do I know if a number is really a KPI?
Ask this question: does anyone change anything when this number moves? If no action changes, it's not a KPI — it's just a pretty metric. An indicator without actionability consumes resources and generates no operational value. Always define the action before putting it on the screen.
Do I need an ERP to use Qlik Sense?
Technically no, but in practice yes. Only 53.7% of companies in Portugal used ERP in 2025. Without an integrated core, BI runs on scattered Excel sheets — truncated, out-of-date, manual data. An ERP with sales, stock and production in the same place is the foundation that makes BI live beyond the third month.
How do I control who sees sensitive data in Qlik Sense?
Define access profiles by function, not by person. The line logger doesn't see margins; the salesperson doesn't see salaries. Log who accesses sensitive data — HR, pay, appraisal — as required by GDPR and Law 58/2019. When someone changes post, they change profile automatically.
What is the advantage of Qlik Sense's associative model?
Click on a reference and the whole panel reacts — sales, stock, orders, customer, colour, size. You don't need to know the question in advance. Follow the data wherever it takes you. In a fixed report, 800 SKUs across three axes is impossible combinatorics. An associative model answers the questions you didn't yet know you had.