Skip to main content
Aamina Mughal

Blog7 min read

What business intelligence actually does

I nodded along to the phrase for a year before I understood it. Here is what it means in plain words, where it gets used, and the one idea that changed how I read numbers.

For most of a year I nodded along whenever someone said “business intelligence”. I pictured a screen full of charts, probably blue, probably in an office I had never been to. Then I started building something like it myself, and found out that the screen full of charts is the last small part of the job.

What the phrase actually means

Business intelligence is the work of turning what an organisation records into something it can decide with. That is the whole idea. A shop already records every sale. A college already records every enrolment. Recording is not the same as knowing, and the gap between the two is where this work lives.

It usually runs as a loop of five stages. Someone has a question. The numbers that might answer it live in more than one place, so they get collected. They arrive in shapes that do not match, so they get cleaned and combined. Then you decide how the pieces relate to each other, and only then do you show the result. And showing the result almost always produces a better question, so the loop starts again.

The five stages of a business intelligence cycleA question leads to sources, then cleaning and combining, then modelling, then showing the result. An arrow returns from the end to the beginning, because the answer usually changes the question.A questionsomeone must decideSourceswhere numbers liveClean & combinemake them matchModelhow they relateShow itchart, table, pagethe answer usually changes the question
The five stages. The dashed arrow is the part nobody warns you about.

The part that surprised me most is where the effort sits. I had assumed this was mainly about choosing charts. On the project I am working on now, collecting and cleaning took weeks. The first chart took an afternoon.

Where it gets used

These are the examples that made it click for me. None of them are my own work — they are the sort of question the technique is built for, and I found them easier to hold onto than a definition.

A shop

Which products sell together?

Puts the right things on the same shelf, and stops money sitting in stock nobody wants.

A café chain

Which hours actually need more staff?

A manager can feel that Saturday is busy. The data says whether it is busy at nine or at two.

A hospital

Where do patients wait longest?

Finds the one step in a process that holds up everything after it.

A college

Which students stop logging in?

Someone can reach out in week four instead of finding out in week twelve.

Four ordinary questions. Each one is answerable only once the recording has been turned into something comparable.

What all four have in common is that the answer changes what somebody does on Monday. That is the test I now apply to my own work: if a chart does not change anyone’s decision, it is decoration.

The idea that changed how I read numbers

Here is the one thing I would tell myself a year ago. The average is not the typical case, and the difference matters enormously.

Imagine a city measuring how long it takes to close a service request. Most get closed quickly. A handful drag on for months, because they need a contractor or a decision from somebody senior. Those few pull the average upward, away from almost every actual request.

A right-skewed distribution where the mean sits far above the medianMost requests close within ten days, but a small number take up to a hundred. The median is ten days and the mean is twenty point four days, so the average is roughly double the typical case.07142128020406080100days taken to closemedian 10 dayshalf finish faster than thismean 20.4 daysdragged right by the long tail
Most requests close inside ten days, but a small tail runs to a hundred. The median is 10 days; the mean is 20.4. Illustrative numbers, made up to show the shape — not measured data.

Look at where the mean sits. There is barely any data near it. If you promised people an average of twenty days, you would be describing a case that hardly exists — worse than useless, because it sounds precise.

The median answers a more honest question: what happens to the person in the middle? Both numbers are true. They just answer different questions, and picking the wrong one is an easy way to mislead people without ever writing down anything false.

What it cannot do

Three limits I keep having to relearn.

It describes what happened, not why. A dashboard can show that sales dropped in March. It cannot tell you that a competitor opened across the street. The explanation always comes from somewhere outside the data.

It inherits every flaw in its inputs. If half the entries were typed in wrong, a beautiful chart makes the wrong answer more convincing, not less.

A dashboard is an argument. Somebody decided what went on it, what got left off, and which comparison sits at the top. That is a set of choices, and it is worth asking whose choices they were.

What I am doing with it

I am building my first real one now, on a dataset I put together myself about how online attention spreads across platforms. It is not finished. The cleaning stage took far longer than I expected, which — going by everything above — seems to be the normal experience rather than a sign that I am doing it wrong.

The project it comes from has the methodology, including the parts that did not work.

All posts