The skills illusion: Why most skills-based hiring is built on bad data
TL;DR
Most skills-based hiring relies on self-reported or inferred skills, which are inconsistent, biased, and easy to game. AI that uses this same data doesn’t fix that; it scales it. When leaders can’t trust skills data, they compensate with more interviews, slower decisions, and gut feel. If skills hiring is going to work, skills need to be measured through behaviour, not declared or guessed.
Everyone is talking about skills.
Skills-based hiring.
Skills frameworks.
Skills clouds.
Talent marketplaces.
If you’re a CHRO or Head of Talent, it probably feels like this should already be solved. Skills sound sensible. Objective. Fair. A welcome move away from CVs and pedigree.
But the problem not enough of us are talking about: Most skills-based hiring today is built on data we shouldn’t trust.
And when that data is fed into AI systems and scaled across an organisation, the issue is no longer theoretical. It’s a real business risk.
The original sin of skills hiring: self-declaration
Let’s start with the most common approach to skills hiring – asking people to tell you what their skills are.
CVs.
LinkedIn profiles.
Internal skills databases.
Employee self-assessments.
The logic is straightforward. If people declare their skills, we can match them to roles.
The problem is that humans are wildly inconsistent in how they describe themselves. Not because they’re dishonest, but because skills are:
- contextual (“good communicator” means different things in different roles)
- socially shaped by confidence, norms, and culture
- described unevenly across gender, background, and language
In the data, the pattern is clear and persistent:
- some people over-claim while others under-claim
- women and underrepresented groups are generally more conservative
- some cultures avoid self-promotion almost entirely.
And, let’s be honest, that’s bias, neatly wrapped up as data.
Inferring skills from CVs doesn’t solve it; it amplifies it
To get around self-reporting, many organisations turn to AI to infer skills instead.
Keyword extraction, ontology mapping, skills taxonomies, and match scores.
These approaches, while gaining wild popularity for their efficiency gains, rest on a second shaky assumption:
That what someone has done before is a reliable predictor of what they can do next.
Decades of organisational psychology tell us this is only weakly true. Job titles, tenure, and career paths are blunt instruments. They say very little about how someone thinks, adapts, learns, or performs under pressure, all things that matter in modern roles.
And, there’sthe elephant in the room that must be acknowledged. Anyone can now generate a flawless CV for almost any role using generative AI.
So we’ve landed in an odd place:
- we know CVs are a poor proxy for capability
- we know inferred skills are built on noisy signals
- and yet we’re automating decisions on top of them
This is classic garbage in, garbage out. Except it’s faster, and at enterprise scale.
The huge hidden cost: interview inflation
There’s a downstream effect of bad skills data that rarely shows up on dashboards; in fact, it’s often not considered at all.
When hiring managers don’t trust the data upstream, they compensate downstream with interviews.
Roles can quietly stretch to six, seven, sometimes eight interview rounds. Not because leaders enjoy it, but because interviews have become the only place where they feel they can see anything real.
I still remember my first day as a CHRO when my CEO said:
“We’re missing our revenue number because our leaders are spending too much time interviewing. Fix it.”
But it wasn’t an HR problem. It was a business problem caused by low-confidence talent assessment.
When you don’t trust skills data, you force leaders to manufacture certainty, spending time interviewing. And time is the most expensive resource you have.
Why match scores feel reassuring, but don’t help decisions if there’s no insight
This is where matching enters the chat.
On paper, matching is elegant:
Candidate A is a 78% match.
Candidate B is a 72% match.
But ask a hiring manager a simple question: Why is this person a better fit?
Too often, the skills-matching technology can’t provide an answer they can actually use. There’s no behavioural evidence, no explanation they can defend and no insight they can probe.
And that’s why, in reality, match scores don’t change behaviour, nor do they solve the quality or efficiency issue in hiring.
Interviews are not reduced. Decisions remain slow. And critically, they don’t stand up to scrutiny from candidates, regulators, or boards.
Matching without insight is simply false confidence.
A different approach: measure skills, don’t extract them
If skills really matter, and they absolutely do, then the question has been wrong all along.
Instead of trying to figure out how to extract skills from people or documents, talent teams should be asking, how do we measure skills in a way leaders actually trust?
That’s where skills-based hiring either works or collapses under its own weight.
At Sapia.ai, we chose to measure. Because we knew that there was a possibility to scale the best of behavioural science with AI.
So instead of starting with CVs, we start with behaviour.
We built a data-driven competency framework grounded in organisational psychology and validated at scale. Not a static skills taxonomy, but a living model of human capabilities that predict performance across roles.
From tens of thousands of job descriptions and millions of structured interviews, we identified 25 role-agnostic competencies, including:
- Analytical thinking
- Learning agility
- Accountability
- Resilience
- Customer orientation
These are all observable capabilities, designed to be measured through structured, skills-based interviews, not inferred from documents.
The interview is the new CV
Then, we reframed hiring, and flipped it on it’s head.
What if the interview wasn’t the bottleneck?
What if it was the first asset that talent teams could leverage to find the right people for the role and organisation?
A structured, AI-enabled interview, delivered upfront, for everyone:
- the same questions for every candidate
- focused on skills and behaviours, not pedigree
- untimed, inclusive, and bias-aware
- generating evidence, not just scores
Our customers use it as the new CV.
Not static, and not optimised for keywords. It’s dynamic, explainable, and grounded in how people actually think and work.
When you start there:
- hiring managers don’t need eight interviews
- decisions move from gut feel to informed debate
- skills become reusable, not disposable
- talent intelligence for the whole organisation compounds over time
Trust is the real currency of skills-based hiring
This isn’t really a story about AI, frameworks, or even skills.
It’s a story about trust.
If leaders don’t trust the data, they won’t use it.
If candidates don’t trust the process, they won’t engage.
If boards don’t trust the logic, they won’t approve it.
The future of skills-based hiring won’t be won by bigger skills databases or more sophisticated matching algorithms. It will be won by organisations that stop guessing and start measuring. Because skills aren’t something people declare, they’re something people demonstrate.
And once you can see that clearly, hiring gets far more human.
FAQ: Skills-based hiring and talent assessment
What is skills-based hiring?
Skills-based hiring focuses on assessing candidates based on their abilities and behaviours rather than their CVs, job titles, or background. In practice, many approaches still rely on self-reported or inferred skills, which introduces bias and inconsistency.
Why do many skills-based hiring initiatives fail?
Most fail because the underlying skills data is unreliable. Self-declared skills and CV-inferred skills vary widely by confidence, culture, and background, which undermines trust and adoption.
Is AI good or bad for skills hiring?
AI can be powerful, but only if it’s applied to high-quality inputs. When AI is layered on top of weak skills data, it scales the problem rather than fixing it.
How should organisations measure skills instead?
Through structured, behavioural talent assessment. This means measuring how people think, decide, and act in role-relevant scenarios, rather than guessing based on past experience.
Do structured interviews really scale?
Yes, when designed properly. AI-enabled, structured interviews can be delivered consistently and fairly at scale, while generating evidence leaders can trust.
What’s the biggest business benefit of better skills data?
Speed and confidence. When leaders trust the assessment data, they make decisions faster, run fewer interviews, and spend time where it actually adds value.
About Author
