Over the last two posts we have laid out the evidence of how utterly messed up our hiring system is, and with the entry of AI its messed up at scale!
A Stanford study found AI hiring tools rejecting Black and Asian applicants at disproportionate rates across 156 companies sharing the same AI hiring vendor. Researchers called it algorithmic monoculture. In Australia, 85% of companies claim to have adopted skills-based hiring while only 1 in 700 of actual hires reflect it. Only 1% of tech graduates are assessed as immediately work ready, despite strong underlying demand for exactly that talent.
Different studies. Different countries. Same root cause.
The tools being used to screen people were not built to measure true human capability. They were built to guess at it, using whatever proxy is fastest to process at volume. A degree. A keyword match. A personality style. A resume format. None of those things are the job. They are signals that correlate, imperfectly and sometimes unfairly, with who has historically gotten hired.
Guessing at scale is the actual problem.
Not bad intent. Not lazy recruiters. A system architecture that was never designed to verify anything. Only to filter quickly.
When you filter quickly using flawed proxies, and then repeat that same filter across 156 employers because they all licensed the same vendor, you do not get faster fair hiring. You get the same wrong guess, repeated at industrial scale, with no one accountable for checking whether the guess was right.
So what does it actually look like to verify capability instead of guessing at it?
It starts with separating the question employers are really asking from the question their tools are answering.
Employers want to know: can this person do the job, and how ready are they to perform from day one?
Most current tools answer a different question: does this person’s paper trail resemble people who got hired before?
Those are not the same question. And the resume was never built to answer the first one.
Verification means assessing what someone can actually do, regardless of how they built that capability. University degree. TAFE certificate. Bootcamp. Self-directed learning. A career change. Time out of the workforce and back in. The pathway should not be the filter. The capability should be. And it should be verified, not asserted.
It also means making that verification visible to the person being assessed, not just the person doing the assessing.
Right now, when an AI hiring tool rejects someone, they often never know it happened. Let alone why. That is not a side effect of automation. That is the actual design problem. A system that cannot explain its own decision to the person it rejected is not a hiring tool. It is a black box with a job title.
This is the thinking behind what we are building at Gradstack.
We do not use AI to build a scoring matrix. That is the same guessing problem, just automated with a new label. We use AI to match against data we have verified is real, not keywords on a resume. Real signals. Real proof. Matched at scale, in a way that is safe and responsible, and inclusive - not another black box making decisions no one can explain.
The VCI® (Verified Capability Index) and VEI® (Verified Employability Index) score what someone can actually do, built from a Digital Human Profile® rather than a resume. Readable consistently by every employer. Not guessed by every vendor’s proprietary model. Backed by human-verified data, which no other company globally can do and here is the best part - it grows with you!
The goal is not to make screening faster. Screening is already fast with AI.
The goal is to make it accurate. So capable people stop getting filtered out by a proxy that was never measuring them in the first place.
Signal over noise.
Australia is about to run one of the largest workforce transitions in its history.
Billions in public investment in skills, training, and AI readiness. That investment will not produce economic output if the hiring layer sitting on top of it is still guessing. The job ad language changing is not enough. The measurement layer has to change.
That is the work in front of us.