Every talent-focused organisation knows this rule instinctively: the best predictor of future performance is usually past performance, not a credential on a page. It's why smart hiring teams weight demonstrated track record over polished resumes.
Strangely, Australia's national system for selecting permanent skilled migrants does the opposite. New research combining tax records, Census data and visa histories back to 1990 shows the points test — the mechanism that decides who gets to stay permanently — is scoring migrants on criteria that don't actually predict success, while ignoring the performance data the government already has sitting in its own systems.
It's a talent-matching failure playing out at national scale, and the fix reads like something straight out of a workforce analytics playbook.
A selection model running on the wrong signals
Think of the points test as a national hiring algorithm. Like any scoring model, it's only as good as the variables it weights — and researchers found several of its heaviest-weighted variables carry little to no relationship with actual outcomes.
Criteria such as community language credentials, regional study, "professional year" programs, niche specialist qualifications, and simply having graduated from an Australian university all failed to predict stronger earnings or employment outcomes. Some were tied to worse results.
The international student pathway — a major feeder into the permanent system — came out particularly weak. Migrants who moved from a student visa into points-tested permanent residency tended to earn less and were more often working below their qualification level than migrants who arrived through other channels.
In talent terms: the model is optimising for signals that look good on an application but don't correlate with on-the-job value.
The biggest miss: ignoring your own performance data
Here's the part that should resonate most with anyone who works in talent analytics or workforce planning: a large and growing share of points-test applicants are people already working in Australia — many on temporary or bridging visas, with real, observable employment and earnings history.
And yet the system scores them identically to first-time offshore applicants with zero Australian track record.
That's the equivalent of a company having two years of performance reviews on an internal candidate and choosing to ignore all of it in favour of a generic screening questionnaire. The data exists. It's linked to tax and visa records. It's simply not being used to inform the decision.
Three design flaws, viewed as a scoring-model problem
1. Feature weighting doesn't match outcome data. High-signal variables — age, education, English proficiency, relevant work experience — are underweighted relative to their actual predictive strength, while low-signal variables still contribute points.
2. One model, wildly different labour segments. The same scoring approach is applied across occupations with very different dynamics. It performs reasonably in tightly regulated fields like healthcare, where qualifications are a clean proxy for job-readiness. It performs poorly in fields like engineering and business, where fit, networks and demonstrated capability matter more than checkboxes.
3. No use of longitudinal performance data. For migrants already employed in Australia, the system has months or years of earnings and employment history available — arguably the single strongest predictor in the entire dataset — and simply doesn't factor it in.
What a data-informed redesign looks like
Two reform paths have emerged from the research, and both read like standard practice in any serious talent-analytics function.
Option 1 — Recalibrate the existing model. Cut the variables with no predictive value, and reweight the scoring toward the traits the data actually supports: age, education, English proficiency, and Australian work experience. Same basic structure, better-calibrated weights.
Option 2 — Segment the model by candidate type. This is the more sophisticated option, and it mirrors exactly how a mature talent function would treat two fundamentally different candidate pools:
- External candidates (offshore applicants and recent graduates with no Australian work history) get a traditional points model — age, education, English — but scoped to occupations where the data shows this group genuinely succeeds without local experience, largely regulated professions.
- Internal candidates, so to speak — temporary visa holders already working in their nominated role — get scored primarily on age and actual Australian earnings over the past two years. Real, demonstrated income becomes the core signal, not credentials.
It's a segmentation strategy any workforce intelligence team would recognise: don't score a known quantity the same way you score an unknown one.
Why this matters beyond policy circles
For talent and workforce professionals, this case study is a useful reminder that even large, high-stakes selection systems can run for decades on unvalidated criteria if nobody goes back to check the data against the outcomes. The fixes proposed here aren't radical — they're the same discipline good talent analytics teams apply constantly: test your scoring model against real results, drop what doesn't predict performance, and give more weight to the data you already have on people you already know.
Done well, this kind of recalibration wouldn't just make Australia's migration system fairer. It would make it measurably better at its core job — identifying the people most likely to thrive, contribute and succeed.
Tags & Keywords