What Is Data Analytics & Business Intelligence?
Most people start a Power BI course by opening Power BI. That is a mistake. The tool is easy; knowing what question you are answering is the hard part, and it is the part that separates an analyst from someone who makes charts. This lesson gives you the mental model everything else in the course hangs from.
Learning Objective
Before touching any software, you need to be able to place a business question into the right analytical category — because the category determines the technique, the data you need, and how confidently you are allowed to speak about the answer.
- Define data analytics and business intelligence, and explain precisely how they differ
- Classify any business question as descriptive, diagnostic, predictive or prescriptive
- Explain what each analytic type can and cannot legitimately tell you
- Identify where Power BI operates on that spectrum — and where it does not
- Describe the analyst's role in the chain from raw data to a changed decision
Why This Matters
Consider two analysts handed the same file of sales data.
The first builds twelve charts, arranges them on a page, and sends it out. It is colourful. Nobody acts on it, because nobody asked a question it answers.
The second asks: “Why did margin fall last quarter when revenue rose?” She builds four visuals, finds that a discount campaign on Electronics drove volume at near-zero margin, and recommends capping the discount at 15%. Someone changes what they do. That is analytics.
This matters commercially too. Job descriptions say “Power BI Developer”, but interviews test analytical thinking: given this business problem, what would you measure and why? You can learn every DAX function and still fail that question. Module 14 comes back to this directly.
Concept
Data analytics: turning observations into answers
Data analytics is the discipline of examining data in order to answer a question or support a decision. That definition has three load-bearing parts: there is a question, there is data, and there is a decision that could go differently depending on the answer. Remove any one and you are not doing analytics.
The four types of analytics
Every analytical question falls into one of four types. They form a ladder: each rung is more valuable and more difficult than the one below, and you cannot skip rungs — you cannot credibly predict what you cannot yet describe.
Business intelligence: analytics as an operating system
Business intelligence is the practice and infrastructure that makes analytics repeatable and shared. A one-off analysis in a spreadsheet is analytics. A governed semantic model that 400 people query through a refreshed report, where “revenue” means the same thing to everyone, is business intelligence.
The distinction is not academic — it changes what you build:
| Aspect | One-off analysis | Business intelligence |
|---|---|---|
| Lifespan | Answers one question, then it is done | Runs continuously for years |
| Audience | Usually you, or one stakeholder | Many people, often across departments |
| Data | A file someone sent you | A governed, refreshing source of record |
| Definitions | Whatever you decided that day | Agreed, documented, enforced in the model |
| Failure mode | Wrong answer, once | Wrong answer, silently, to everyone, for months |
| Tooling | Excel, Python notebook | Power BI, a semantic model, a refresh schedule |
Where Power BI sits
Power BI is a business intelligence platform that is strongest at the descriptive and diagnostic rungs, has real but limited predictive features, and leaves prescriptive work to you. Being honest about that boundary is part of being good at this job.
Analytics, BI and data science
These three overlap and the labels are used loosely in job adverts. A workable distinction:
- Business intelligence — building the shared, governed reporting layer. Power BI, SQL, data modelling. Answers: what happened, and why.
- Data analysis — investigating specific questions, often ad hoc, often deeper than a dashboard goes. Overlaps heavily with BI; the same person frequently does both.
- Data science — statistical modelling and machine learning to predict and classify. Python/R, experimentation, model deployment. Answers: what will happen, and what drives it, with quantified uncertainty.
Visual Explanation
The four analytic types applied to a single real number. Below is monthly revenue from the course dataset — the same figure you will build in Power BI in Module 6. Watch how the question you ask of it changes completely at each rung of the ladder.
| Rung | The question | What this chart says | Honest limit |
|---|---|---|---|
| Descriptive | What happened to revenue? | Revenue grew from about $98K in Jan 2024 to about $473K in Dec 2025, with a November peak in both years. | Tells you nothing about why. |
| Diagnostic | Why is November always the peak? | Needs decomposition: is it more orders, larger orders, or a specific category? The chart alone cannot say. | Finds associations; proving cause needs an experiment. |
| Predictive | What will next November look like? | Two seasonal cycles is a thin basis. You could extrapolate, with wide uncertainty. | Two years of history is not enough for a confident seasonal forecast — say so. |
| Prescriptive | Should we increase November stock? | Requires the diagnostic answer plus holding cost, stockout cost and lead time — data not in this chart at all. | This is a judgement call, not an output. |
Where the value is — and where the effort goes
Power BI Demonstration
There is nothing to click in Power BI yet — but there is something important to see: where each rung of the ladder actually lives in the product. Keep this map in mind as you go through the course.
| Analytic type | Power BI features that serve it | Covered in |
|---|---|---|
| Descriptive | Measures (SUM, COUNT), cards and KPIs, bar/line/matrix visuals, slicers | Modules 5, 6, 7 |
| Diagnostic | Drill-down and drill-through, cross-filtering, Decomposition Tree, Key Influencers, period-over-period measures | Modules 6, 7, 9, 13 |
| Predictive | Line-chart forecasting, Key Influencers, Q&A; external models via Python/R visuals | Modules 7, 13 |
| Prescriptive | None — this is analyst judgement, communicated through report narrative and design | Modules 8, 14 |
Example Dataset
This lesson is conceptual and needs no download — it uses the sample data shown inline above. The next lesson that requires a file will link it here.
You can still browse every dataset in the course from the dataset library.
Step-by-Step Exercise
This exercise takes about ten minutes and needs no software — a notepad is enough. It is the single most transferable skill in the course, so do it properly rather than skimming.
Diagnostic: Why is Tuesday afternoon consistently our weakest slot? (Needs: the above plus staffing, weather, local events. Power BI: partly — it can show the pattern; explaining it needs external data.)
Predictive: How many pastries will we sell next Saturday? (Needs: 1–2 years of daily sales with seasonality. Power BI: a rough forecast line, yes; a reliable number, no.)
Prescriptive: Should we cut Tuesday afternoon staffing by one person? (Needs: all of the above plus labour cost and service-level tolerance. Power BI: it informs the call; it does not make it.)
Expected Result
You should now have four written questions where each one names a rung, names the data it needs, and names a decision it would influence. If any question fails the decision test, that is a successful outcome too — you have just saved yourself from building something nobody would use.
✓ The descriptive question could be answered by a single well-built visual
✓ The diagnostic question names something you would decompose by (category, region, time, customer type)
✓ You were honest that the predictive question needs more history than you probably have
✓ Every surviving question has a named decision-maker and a named decision
Keep this sheet. In Module 14 you will turn a set of questions exactly like these into a portfolio project brief, and having done the thinking now will make that far easier.
Common Mistakes
Fix: Explore freely for twenty minutes, then stop and write down the two or three questions worth answering. Build only those. Everything else goes in a notes file.
Fix: Use language that matches your evidence: “the Electronics discount accounts for roughly 80% of the margin decline” is defensible. “Caused” usually is not, unless you ran an experiment.
Fix: State the basis and the uncertainty whenever you show a forecast: “based on 24 months, so treat the seasonal component as indicative.” Never present a forecast without its band.
Fix: Ask what would change in your conclusion if you had the extra data. If the answer is “nothing”, you do not need it.
Fix: Ask one question before you build: “what decision will you make differently depending on what this shows?” The answer reshapes the whole project.
Professional Tip
“This report helps [specific person or role] decide [specific decision] by showing [specific metric] broken down by [specific dimension].”
If you cannot complete it, you are not ready to build — and the honest, valuable move is to go back and ask. Experienced analysts do this instinctively. Writing it down until it becomes instinct is the fastest way to get there.
Two further habits worth adopting early:
- Name the rung out loud in meetings. “That's a diagnostic question — I can show you what correlates, but I can't prove causation from this data.” It sets expectations correctly and it reads as competence, not as hedging.
- Write the conclusion first. Draft the sentence you expect the report to support before you build it. Then build honestly and see whether the data agrees. If it doesn't, that disagreement is your most interesting finding.
Knowledge Check
Knowledge Check
5 questions. Answer each one to see why the right answer is right.
A retail director asks: “Our margin dropped 3 points last quarter — what happened?” Which type of analytics does this question call for?
What is the clearest single difference between a one-off analysis and business intelligence?
Power BI draws a forecast line on your 24-month revenue chart, complete with a confidence band. What is the responsible way to present it?
Which of these project briefs is ready to build?
You find that stores with longer opening hours have higher revenue. What can you legitimately claim?
Practical Challenge
Write a defensible one-page analysis brief
You are handed the course retail dataset — 15,067 order lines, 7,607 orders, 896 customers, two full years. The commercial director says only: “Something's wrong with our margins. Have a look.” That is all you get. Turn it into a brief you could actually build from.
- State the one descriptive question you would answer first, and why that one
- State two diagnostic questions that would decompose the margin problem
- For each question, name the specific columns you would need from the dataset
- Write the one-sentence brief from section 10 for the whole project
- Name one thing you would refuse to claim from this data, and explain why
Margin is Profit / NetSales. Anything that moves either number moves margin. There are only four levers, and every margin problem is one or more of them:
- Price — are we selling at lower unit prices?
- Discount — are we discounting more heavily or more often?
- Cost — has unit cost risen?
- Mix — are we selling proportionally more of the low-margin categories? This one is the most commonly missed, and it is often the answer.
The dataset has a column for each of the first three. Mix you derive by comparing each category's share of revenue across periods. Note that the categories in this dataset have very different margins — Electronics is around 13% while Beauty is around 44% — so mix shifts move total margin hard.
Solution
Try the challenge properly before opening this. Reading a solution you have not attempted feels like learning, but it is not.
A model answer
The first descriptive question
“How has profit margin % moved by month across the two years?”
Why this one first: the director's claim is unverified. Before investigating a cause you must confirm the effect exists, establish its size, and locate it in time. If margin has actually been flat and one bad month spooked them, the entire project changes — and finding that out in an hour rather than a fortnight is a real result.
Columns needed: OrderDate, NetSales, Profit.
Two diagnostic questions
- “Is the margin movement driven by mix — are low-margin categories taking a larger share of revenue over time?”
Columns:OrderDate,ProductKey→Category,NetSales,Profit. In this dataset Electronics is by far the largest category by revenue but carries the thinnest margin, so a shift toward it would drag total margin down even with every individual product's margin unchanged. That distinction — total margin falling while no product got less profitable — is the single most valuable thing you can show a commercial director. - “Is discounting deepening, and is it concentrated in particular categories or months?”
Columns:Discount,OrderDate,Category,NetSales. Compare average discount rate and discounted-revenue share over time. The November promotional period is the obvious place to look first.
The one-sentence brief
What you would refuse to claim
You would refuse to claim that reducing discounts would raise profit. The dataset shows what was sold at what discount; it cannot show what would have been sold at a different discount. Cutting a discount may simply lose the volume. Establishing that requires a controlled test — a genuine price experiment — not historical data, however much of it you have.
If you are self-assessing, score against these weights. The same rubric shape is used for every project in Module 10.
Lesson Summary
- Data analytics needs three things: a question, data, and a decision that could go differently. Missing any one of them and you are making charts, not analysing.
- The four rungs — descriptive, diagnostic, predictive, prescriptive — each answer a different question and carry different limits. You cannot skip rungs.
- Business intelligence is analytics made repeatable, shared and governed. Its failure mode is worse than a one-off's, which is why modelling discipline matters so much.
- Power BI is strongest at descriptive and diagnostic work, has real but limited predictive features, and leaves prescriptive judgement to you.
- Observational data supports association, not causation. Match your language to your evidence.
- The one-sentence brief — person, decision, metric, breakdown — is the fastest test of whether a project is worth building.
Syllabus topics delivered
Progress
Marking a lesson complete stores your progress in this browser and updates your BitWithBite learner dashboard. Nothing is uploaded anywhere.