Data analytics, from question to decision.
Eight modules that take you from a vague business question to a defensible recommendation — metrics, cleaning, SQL, statistics, charts, experiments and forecasts, taught in plain English with real depth. Finish with a 25-question exam and an instant certificate.
What you'll be able to do.
Written as demonstrable outcomes, the way the AQF frames them. On completing this course you will be able to:
- CLO1Translate business questions into measurable metrics and KPIs, classifying data types and telling leading from lagging indicators.
- CLO2Source, assess and prepare data for analysis — applying data-quality dimensions, a repeatable cleaning sequence and tidy-data structure.
- CLO3Use spreadsheet techniques and core SQL to filter, aggregate, combine and summarise data.
- CLO4Describe data with appropriate statistics — centre, spread, percentiles and samples — and choose honestly when data is skewed.
- CLO5Select truthful, effective visualisations and assemble them into decision-ready dashboards.
- CLO6Distinguish correlation from causation, design simple A/B tests, build baseline forecasts, and present findings ethically with clear recommendations.
Who it's for: anyone who works with numbers and wants to be the person who can answer "what's actually going on?" — analysts-to-be, marketers, operations and small-business owners. No maths beyond school level, no programming required. Pairs well with Computer Science Foundations.
Eight modules, question to recommendation.
Each module states its learning outcomes, teaches the content, and sets an Apply it task. Everything in the exam is covered here.
Module 1Thinking with data — questions & metrics
- LO1.1Distinguish data, information and insight, and place analytics inside the decision loop.
- LO1.2Classify data: structured vs unstructured, quantitative vs qualitative, discrete vs continuous.
- LO1.3Define metrics and KPIs that are measurable, targeted and owned — and tell leading from lagging indicators.
Core content
Analytics exists to improve decisions, and the whole craft hangs off one chain: data → information → insight → action. Data is raw facts (3,412 rows of sales); information is data with context (sales fell 8% in March); insight is the why-it-matters (the fall is entirely in one product after a price change); action is what you do about it. A number without a question attached is trivia — so good analysts start at the decision end and work backwards to the data they need.
Knowing your material helps: structured data sits in rows and columns; unstructured data (reviews, emails, photos) doesn't. Quantities split into discrete (counted — orders, complaints) and continuous (measured — temperature, time on page). And the numbers a business steers by, KPIs, are only useful when they're specific, measurable, owned by someone and attached to a target — "lift repeat-purchase rate from 18% to 25% by December" beats "improve customer happiness" every time. Watch the tense, too: lagging indicators (revenue) tell you what already happened; leading indicators (pipeline value, sign-ups) hint at what's coming while you can still act.
Module 2Collecting & preparing data
- LO2.1Identify common data sources — transactions, logs, surveys, APIs and open data.
- LO2.2Assess data against quality dimensions: accuracy, completeness, consistency, timeliness, uniqueness and validity.
- LO2.3Apply a repeatable cleaning sequence and structure the result as tidy data.
Core content
Data arrives from transactions (sales, bookings), system logs, surveys, third-party APIs and public open data — and almost none of it arrives ready. Practitioners consistently report that preparing and cleaning data takes the majority of a project — commonly cited as 60–80% — because analysis inherits every flaw upstream: garbage in, garbage out. The data-quality dimensions give you the vocabulary: accuracy (is it true?), completeness (what's missing?), consistency ("NSW", "N.S.W." and "New South Wales" must be one thing), timeliness (is it stale?), uniqueness (the same customer three times under different spellings is a duplication failure) and validity (a postcode of "ABC" fails the format).
Cleaning has an order: profile the data first, fix types, standardise categories and dates, resolve duplicates, then handle missing values deliberately — delete, impute or flag, and say which you did. Treat outliers with respect: investigate before deleting, because sometimes the outlier is the story. The destination is tidy data — each variable a column, each observation a row, one table per kind of thing — the shape every tool downstream expects. And document what you changed; an analysis no one can retrace is an opinion.
Module 3Spreadsheets & SQL — the analyst's toolkit
- LO3.1Use spreadsheet workhorses: sorting, filtering, lookups and pivot tables.
- LO3.2Read and write core SQL: SELECT, WHERE, GROUP BY, ORDER BY, JOIN and aggregate functions.
- LO3.3Judge when a job has outgrown a spreadsheet and belongs in a database.
Core content
Spreadsheets are a legitimate analytics tool — fast, visual and everywhere. Pivot tables are the workhorse: drag a category to rows, a number to values, and you've grouped and aggregated without a line of code; lookups stitch tables together. But spreadsheets have edges: they strain long before their ~million-row ceiling, several people editing at once ends badly, formula errors hide silently, and there's no audit trail. When data grows large, shared and business-critical, it belongs in a database.
That's where SQL earns its keep — a small vocabulary that maps directly onto analytical questions: SELECT picks columns, WHERE filters rows, GROUP BY buckets rows so aggregates (SUM, COUNT, AVG, MIN, MAX) can summarise each bucket, ORDER BY sorts, and JOIN combines tables on matching keys. SELECT state, SUM(amount) FROM sales WHERE year = 2025 GROUP BY state ORDER BY SUM(amount) DESC; reads almost like the business question it answers — "total sales by state for 2025, biggest first". Fifty years old and still the language of data.
Module 4Describing data — statistics that matter
- LO4.1Choose between mean, median and mode — and explain why outliers and skew change the choice.
- LO4.2Communicate spread with range, IQR, standard deviation and percentiles.
- LO4.3Distinguish populations from samples and interpret sampling honestly.
Core content
Three "averages", three personalities: the mean (sum ÷ count) uses every value but is dragged around by outliers; the median (the middle value) shrugs them off; the mode is simply the most common. For skewed money data — salaries, house prices — the median represents "typical" far better than the mean, which is why a few mansion sales shouldn't set your view of a suburb. Centre is only half the description: two datasets can share a mean and behave nothing alike, so report spread — range, interquartile range, standard deviation (how far values typically sit from the mean) — and percentiles, which power statements like "95% of orders ship within 2 days".
You'll rarely measure everyone. A sample stands in for the population, and 200 well-chosen customers genuinely can speak for 40,000 — if the sample is random and representative. That's also where lies creep in: self-selected surveys, survivorship bias and tiny samples produce confident nonsense. Every estimate from a sample carries uncertainty (a margin of error); honest analysts surface it rather than presenting a point as a fact.
Module 5Visualisation & dashboards
- LO5.1Match chart type to question: comparison, trend, relationship, distribution, composition.
- LO5.2Spot and avoid dishonest visuals — truncated axes, cherry-picked windows, 3-D effects.
- LO5.3Design a dashboard that answers its key questions in five seconds.
Core content
Chart choice is grammar, not taste: bars compare categories, lines show trends over time, scatter plots reveal relationships between two measures, histograms show distributions, and composition (pie/stacked) works only with a handful of parts. The same grammar exposes the dark patterns: a bar chart whose value axis starts at 90 instead of 0 turns a 2% difference into a tower; cherry-picked date windows manufacture trends; 3-D tilts distort area. Start bar axes at zero, label directly, and delete every pixel that isn't information.
A dashboard is charts with a job: the few numbers a decision-maker needs, answerable in about five seconds. Lead with KPI tiles (current value vs target, with the trend — a lone number hides direction), support with detail below, and use colour as meaning, not decoration — consistent, colour-blind-safe, with red reserved for "needs attention". The test of a dashboard isn't beauty; it's whether the right person makes a faster, better call because it exists.
Module 6Correlation, causation & experiments
- LO6.1Interpret correlation strength and direction (the coefficient r) — and its limits.
- LO6.2Identify confounders and explain why correlation is not causation.
- LO6.3Design a basic A/B test: hypothesis, single change, random split, pre-chosen metric and decision rule.
Core content
The correlation coefficient r runs from −1 to +1: the sign gives direction, the magnitude gives strength, so r = −0.85 is a strong negative relationship and r ≈ 0 means no linear relationship. What r can never tell you is why. Ice-cream sales and drownings rise together every summer — not because dessert is dangerous, but because a confounder, hot weather, drives both. Reverse causation and plain coincidence (scan enough series and spurious correlations are guaranteed) complete the trap set. Treat every correlation as a lead, not a verdict.
The honest way to a verdict is the experiment. An A/B test changes one thing, splits users randomly between versions — randomisation is what neutralises confounders — and judges a pre-chosen metric against a pre-chosen decision rule at a planned sample size, not whenever the numbers look flattering. When you can't experiment (ethics, practicality), use natural comparisons and downgrade your language: "is associated with", not "drove".
Module 7Forecasting & predictive basics
- LO7.1Decompose a time series into trend and seasonality and build a naive baseline forecast.
- LO7.2Distinguish regression (predicting a number) from classification (predicting a category).
- LO7.3Explain train/test splits, and recognise overfitting and accuracy traps.
Core content
History has structure: a time series is roughly trend + seasonality + noise. Retail spiking every December is seasonality; a steady climb across years is trend; the rest is noise you shouldn't chase. Forecasting discipline starts with a naive baseline — "same month last year, adjusted for recent growth" — because any cleverer model must beat the baseline to deserve its complexity. Prediction problems then split in two: regression predicts a number (next month's sales); classification predicts a category (spam or not-spam, churn or stay).
Whatever the model, the test is the same: hold back a test set the model never saw during training — performance on unseen data is the only performance that counts. A model near-perfect in training but poor on new data has overfitted: it memorised instead of learning. And beware the accuracy trap: when 99% of transactions are legitimate, a "99% accurate" fraud model may simply never flag fraud — judge models by metrics matched to the cost of each kind of error, and present forecasts as ranges, not points.
Module 8Ethics, privacy & data storytelling
- LO8.1Handle personal information responsibly — minimise, de-identify and aggregate (Privacy Act 1988 awareness).
- LO8.2Identify bias entering through collection, history and choice of metric.
- LO8.3Structure findings as a story — context, finding, implication, recommendation — with uncertainty stated.
Core content
Analysts handle other people's information, and in Australia that comes with obligations — the Privacy Act 1988 and the Australian Privacy Principles shape how personal information is collected, used and disclosed. The working defaults: collect the minimum, and aggregate or de-identify before sharing. Mind small cells — a "group" of two people isn't anonymous — and never let convenience (emailing a raw customer table) outrun policy. Bias needs the same vigilance: if history under-represented a group, models trained on that history will under-serve them — that's sampling/historical bias to find and fix, not a fact about the group. And remember Goodhart's law: when a measure becomes a target, people optimise the measure, not the goal.
Finally, analysis only matters if it lands. Structure findings as a story: context (the question), finding (what the data shows), implication (why it matters), recommendation (what to do, with expected impact and your confidence). Decision-makers don't want your 400 rows or your SQL — they want the "so what", honestly caveated. State the uncertainty; it builds trust faster than false precision ever will.
How this course maps to AQF Level 4.
The Australian Qualifications Framework defines each level by three criteria — knowledge, skills, and application of knowledge and skills. This course's outcomes and exam are designed against the Level 4 criteria (the level of the Certificate IV), quoted below from the framework.
| AQF Level 4 criterion | The framework says… | How this course addresses it |
|---|---|---|
| Knowledge | "Graduates at this level will have broad factual, technical and some theoretical knowledge of a specific area or a broad field of work and learning." | Modules 1, 4 and 7 build broad factual, technical and some theoretical knowledge of analytics — data and metric types, statistics (centre, spread, sampling), correlation theory and forecasting concepts (CLO1, CLO4, CLO6). |
| Skills | "A broad range of cognitive, technical and communication skills to select and apply a range of methods, tools, materials and information to: complete routine and non-routine activities; provide and transmit solutions to a variety of predictable and sometimes unpredictable problems." | Tool and method selection (spreadsheets vs SQL in M3, chart grammar in M5), cleaning routine and messy non-routine data (M2), and the communicate-it Apply it tasks train selecting, applying and transmitting solutions (CLO2, CLO3, CLO5). |
| Application of knowledge and skills | "Graduates at this level will apply knowledge and skills to demonstrate autonomy, judgement and limited responsibility in known or changing contexts and within established parameters." | Applied tasks demand autonomous work with judgement inside set parameters — choosing honest statistics for skewed data (M4), setting an A/B decision rule before seeing results (M6), and building a baseline forecast with stated assumptions (M7) (CLO4, CLO6). |
Criteria quoted from the AQF levels published at aqf.edu.au. The AQF also specifies volume of learning for qualification types (typically 0.5–2 years for a Certificate IV); at ~10 hours plus practice, this course aligns its outcome design to Level 4 descriptors but does not approach a qualification's volume of learning — one more reason it is not, and does not claim to be, an AQF qualification.
Plain-English status: this is a non-accredited course. Its certificate is a completion certificate issued by Course Contact, designed with reference to AQF Level 4 descriptors. It is not an AQF qualification (only ASQA/TEQSA-accredited providers can issue those), confers no licence, and Course Contact is not a registered training organisation. Want an accredited Certificate IV in Information Technology (Data Analysis) or similar? We'll match you with providers, free →
From question to decision. Prove it.
25 questions across all eight modules — 80% to pass, options shuffle every attempt, certificate issued instantly.