Changing into data analytics online, without a fantasy salary

The internet will tell you that a nurse in Geelong, a teacher in Logan, or a 485-visa holder in Auburn can become a “data scientist” in twelve weeks and start on $130,000. Open SEEK. Sort by date. Read the first ten analyst ads in your city. Count how many want a senior who already lives in SQL, how many want a citizen for a cleared agency, and how many will look at a Udemy PDF. That is the labour market. The brochure is a different document.

This piece is for people changing into analytics from something else — admin, finance, nursing, teaching, warehousing, a degree that is not statistics. It is not a ranking of bootcamps. The careers hub collects the rest of that argument. Here the question is narrower: what can you learn online that a hiring manager will still care about after the certificate expires in their memory?

The honest answer is a short list of tools, a portfolio that looks like work, and a salary expectation that would not embarrass you in a conversation with payroll.

The job is not “data science”

Australian job ads abuse language. “Data scientist” is used for a person who will maintain a dashboard. “Analytics engineer” is used for a person who will write SQL that a dbt model could have written. “Business intelligence” still means Excel in half the mid-market.

For a career-changer, the entry jobs that actually exist are usually some mix of:

• Reporting analyst / MI analyst. Refresh the pack. Explain why this week is not last week. Live in Excel, Power Query, a BI tool (Power BI is the public-sector and corporate default; Tableau appears; Looker if someone has already paid for Google).
• Data analyst. SQL against a warehouse or a messy production database, plus the pack. Python if the team is young or the ad is in a tech firm.
• Operations or people analyst. The same work, domain-flavoured. Your old job can be an advantage here if you can quantify it.
• Graduate or junior in a bank, insurer, supermarket or agency. They will train you if you can already join two tables without crying.

“Machine learning engineer” and “applied scientist” are not those jobs. They want a degree, a publication, or three years of production models. An online course that puts “AI” in the title so it can charge $8,000 is not a bridge. It is a product.

Read the essential criteria, not the headline. If the ad wants “5+ years”, “Snowflake and dbt in production”, and “stakeholder management at director level”, it is not a career-change role even if the salary looks like a movie.

SQL, Excel, Python — in that order, for most people

A lot of syllabi reverse this because Python photographs better.

Excel is not beneath you. In a council, a hospital network, a state department or a mid-size retailer in Adelaide or Newcastle, the analyst who cannot build a robust workbook is not an analyst. Learn Power Query, pivot tables that do not explode, XLOOKUP, and how to stop a file becoming a second source of truth. If that sounds dull, good. Dull is employable.

SQL is the filter. If you cannot write a JOIN, a GROUP BY, a window function, and a query that does not scan a fortune, you will fail the take-home. Free and cheap resources are enough to start: Mode’s SQL tutorial, a community edition of PostgreSQL on your own machine, then a public dataset. Paid courses are justified when they mark your queries and make you rewrite them. A video that never asks you to type is a documentary.

Python is a lever, not a personality. Pandas, a bit of visualisation, and enough command line to run a notebook will cover most junior ads that mention it. You do not need to invent a model. You need to clean a file, join it, and not lie about missingness. If you already have Excel and SQL and a domain, add Python when three ads in your city name it. Do not add it because a syllabus is padded with a week of “intro to machine learning” that you will not be allowed to use.

R still appears in research, health and some government shops. If your target is a hospital or a uni, look at the ad before you pick a language tribal war.

Ignore “AI prompt engineering for analysts” as a core skill. It is a side door. If you cannot check the output, you are not analysing.

What a portfolio has to look like

Hiring managers do not owe you a read of your Coursera completion email. They will look at a repo or a folder if you make it easy.

One good piece beats six tutorials. A good piece has:

1. A question a real workplace would ask. “Which clinics missed their target, and is it a coding problem or a staffing problem?” not “I explored the Titanic dataset.”
2. Ugly data. Missing values, a late-arriving dimension, two definitions of “active customer”. If you only ever use Kaggle’s clean tables, you have practised the exam, not the job.
3. A decision. A chart is not a conclusion. Write the paragraph you would send to a manager at 4.40pm.
4. A method you can defend. Why that grain, why that join, why you dropped those rows. Interviews live here.
5. A tool they use. A Power BI file and a SQL script will travel further in Australia than a Streamlit app nobody asked for — unless the ad is for a product-analytics shop in the inner city that wants the app.

Public data that works: ABS, data.gov.au, state open-data portals, Bureau of Meteorology, ACNC charity data, court and tribunal statistics if you can keep your claims humble. Local government expenditure is a gift. It is messy and political and nobody can accuse you of leaking.

Do not invent a “client”. Fabricated case studies are obvious and they train you to lie. If you need a stakeholder, use a community organisation that will give you a real question, or analyse a public process (ambulance ramping, rental bond data, train punctuality) and be precise about what you cannot know.

Host the work where a stranger can open it without a login you control. GitHub is fine. A PDF of screenshots is what people send when the project does not run.

Certificates: which ones a human still opens

The Google Data Analytics Certificate is the supermarket brand. Applicant-tracking systems parse it. Some graduate programs list it as “nice to have”. It will not compensate for an empty repo. Use it as a twelve-week scaffold if you need structure, then throw away the sample projects and do your own.

University microcredentials and graduate certificates (the ones with a unit code you can credit later) matter more for people who already have a degree and want a paper HR understands. They are slow and they are priced like university — often $2,500 to $4,500 a unit, more for a full graduate certificate. They make sense if your employer will pay, or if you are aiming at the public sector and need the brand.

TAFE diplomas and certificates in data, business analytics or statistics sit on the AQF. That can matter for funded places and for managers who still think in qualification levels. Read the unit list. If there is no SQL and no messy-data assessment, it is a business diploma wearing a hoodie.

Marketplace certificates are souvenirs unless the skill is the point. Read whether Udemy certificates are worth it before you treat a coupon as a strategy. A finished $20 course in SQL is worth more than an abandoned $3,000 brand.

Coding bootcamps that have bolted “data” onto a web syllabus should be read with the same suspicion we apply to their job-outcome slides. The dedicated piece on coding bootcamp outcomes is the longer version. If the capstone is a classification model on a famous dataset, you have bought a party trick.

Government hiring is not corporate hiring

The Australian Public Service, state departments, and local government hire a lot of analysts. They also hire slowly, and they write criteria as if a selection panel will die if a verb is missing.

What is different:

• Citizenship and clearance. Many ongoing APS roles still want Australian citizenship. Roles that touch welfare, tax, health, law enforcement or national security can want a Baseline or higher clearance, which is a process, not a course. If you are a permanent resident, you will find more doors in state government, universities, and labour-hire contracts than in a cleared Commonwealth agency. Read the ad. Do not spend $4,000 on a “government data” microcredential if the knockout criterion is a passport you do not have.
• The selection criteria. They mean it. “Demonstrated ability to communicate insights to non-technical audiences” is a work sample, not a personality claim. Write the STAR paragraph with a number.
• Tools. Power BI, Excel, SAS in some shops, a warehouse you will not choose. Python is welcome in newer teams and ignored in others. A Tableau-only portfolio can look fashionable and still miss the brief.
• Pay. APS and state bands are public. A junior analyst is not on a start-up fantasy package. The trade is leave, super, and a job that survives a funding cycle if you are good. That is a feature if you have a mortgage in Blacktown or a child in care.

Corporate hiring is faster, vaguer, and more impressed by a repo if the team is technical. Banks and insurers will still ask for a degree. Retail and mid-market will often take a career-changer who can already do the work and sit with finance without performing. Agencies will take a junior who is cheap and curious. All of them will test SQL if they have been burnt.

Labour-hire and contracting are how a lot of people get their first government desk. The rate looks high until you price super, downtime and the fact that you can be ended on a Friday. It is still a door.

Visa holders and career-changers: the unromantic bit

If you are on a temporary visa, a certificate is not a sponsorship strategy. Employers who will sponsor want someone who already does the job. A 12-week course does not make you “in shortage”. Check the current skilled lists and the actual occupation code before you tell yourself a story about points. Student-visa graduates on a 485 should treat analytics study as a skill, not as immigration advice — this desk does not give that advice, and a bootcamp that implies it does is selling more than tuition.

If you already work in Australia in an adjacent role — finance officer, roster clerk, claims, school operations, warehouse systems — the fastest path is usually inside the building. Offer to own a report that currently lives in someone’s inbox. Ask for SQL access, even read-only. A lateral move at $85,000 with a desk you know is worth more than a speculative jump to a “junior” role that pays $72,000 and wants two years of Snowflake.

If you are coming from nursing, teaching or trades, your domain is an asset in health, education and field operations analytics. Do not hide it under a generic “passionate about data” line. Hide the opposite: the idea that your old job was a mistake. It was inventory. You already know which numbers are lies.

Career-change timelines that do not involve luck look like six to eighteen months of evenings, not a summer. The people who do it in twelve weeks either had a quantitative degree, a job in the next room, or a definition of “analyst” that includes unpaid internships.

Salary, without the screenshots

Fantasy numbers travel because they are screenshots from a different country, a different year, or a different seniority.

In 2026, a realistic first analytics seat in most Australian cities is often advertised in a band that starts in the $70,000s and stretches toward $95,000 if you already have a domain and a degree. Melbourne and Sydney pay more and cost more. Regional roles can be kinder to applicants and meaner on the tool stack. Contract roles quote a day rate that looks like wealth until you count the gaps.

The $120,000 to $150,000 ads are for people who can already model a P&L, own a warehouse layer, or sit in a bank with six years. You can get there. You will not enrol your way there in a quarter.

When a course page shows an alumni salary, ask: employed in what, where, after how long, and did they have a degree and a job already? If the page will not say, treat the number as decoration.

How to study so you finish

Analytics courses die in week four when the videos outrun the practice. The antidote is a weekly artefact, not a streak.

• Two sessions a week that produce a query or a workbook, not a “watched” tick.
• One dataset you return to until it bores you. Boredom is expertise.
• A person who will look at your SQL and be rude. A forum is a weak substitute; a colleague or a meetup in your city is better.
• A finish line that is a repo and an application, not a certificate email.

If you cannot keep a calendar, read how to finish an online course before you buy another beginning. The market will happily sell you a second orientation.

A short working checklist

Before you pay, and again before you apply:

• Five live ads in your city, tools highlighted, citizenship requirements circled.
• Excel and SQL you can demonstrate without a tutorial open. Python only if those ads demand it.
• One portfolio piece on ugly public or workplace data, with a paragraph a manager could use.
• A salary band you would say out loud to a friend who works in HR.
• A credential only if it unlocks a filter you have already seen in writing — AQF, a graduate program list, an employer tuition rule.

You can change into this work. People do, every year, from roster offices and classrooms and warehouses. They do it with queries, not with a fantasy about a six-figure month in week thirteen. The course is a means. The sample is the product.