Take a job description from 2021. Now look at the same one, updated in 2026 with the help of a generative AI tool. It has often tripled in length. It includes skills that no one really knows how to evaluate, ambiguous wording borrowed from the latest trend reports, and a reassuring but misleading impression of completeness.
AI does not lighten the workload, it intensifies it. This is the finding of a recent study published in the Harvard Business Review. And when AI discreetly expands the scope of recruitment tasks, it results in an inflation of criteria, a more complex selection process and, paradoxically, final decisions that are as intuitive as ever, if not more fragile than before.
The real question is not ‘which skills should be listed?’ but ‘which skills actually predict success in this position, in this organisation, at this precise moment?’ These are two very different questions. And the method for answering them consists of three steps.
AI gives you more criteria. Not better decisions.
Generative AI tools can produce competency frameworks in minutes. That's precisely where the problem begins.
These frameworks confuse two fundamentally different categories of skills:
- ‘Visible’ skills: those that circulate in job postings, LinkedIn articles, and industry studies. Easy to list, difficult to assess, often fleeting. ‘Mastery of generative AI’, ‘cognitive agility’, ‘transformational leadership’: phrases that sound good, but which no one knows how to measure in concrete terms.
- Predictive skills: the skills that, in your organisation, in this specific role, distinguish high performers from others. Less spectacular on paper, but measurable and sustainable: learning ability, emotional management, responsiveness under pressure, rigour in execution.
When everything becomes a priority, nothing is. AI amplifies the noise if you don't have a method to filter it. Here's ours.
The 3-question method for choosing what really matters
These three questions are not just another checklist. They are filters. Each one eliminates what does not deserve to be in your model, leaving only what truly predicts success.
Question 1.
Who really performs well in this role, and what sets them apart?
Do not start with a generic job description. Do not start with a market benchmark. Start with your own data.
Identify two or three employees who are truly successful in this role today. Observe their behaviours. How do they react to ambiguity? How do they deal with the unexpected? What do they do differently from others, in a systematic and observable way?
It is these behaviours, not trendy skills, that are your first selection criteria.
Question 2
Can this skill be assessed objectively and reproducibly?
A skill that cannot be measured is not a recruitment criterion. It is a wish.
The distinction is radical. A declarative skill expressed verbally, such as ‘I am comfortable with change’ or ‘I am results-oriented’, has no predictive value if it is based solely on self-declaration. On the other hand, these same traits, when assessed using scientifically validated tools such as psychometric tests, structured role-playing and behavioural analysis, become reliable and defensible data.
A simple rule: if you cannot assess a skill objectively and reproducibly, remove it from your
model. It does not protect you. It gives you the illusion of a rigorous framework.
Question 3
Will this skill still be relevant in 24 months?
This is where the angle becomes strategic and where AI poses a specific problem.
In a context of technological acceleration, the lifespan of a technical skill is shrinking. What is ‘essential’ today may become obsolete before the end of the trial period. Recruiting on the basis of volatile skills means optimising for the short term at the expense of your team's resilience.
Core competencies such as logical reasoning, adaptability, behavioural rigour, emotional self-regulation, retain their predictive value over three to five years. Technical skills, on the other hand, must be monitored and reassessed regularly.
A robust recruitment model consciously balances these two levels. And it is built to last.
How it works: a predictive model for a SaaS sales position
Let's look at a real-life example: recruiting an Account Executive for a fast-growing SaaS company with 120 employees in a competitive software market. The initial job description listed 14 skills. After applying the three questions, five criteria were selected.
01. Define criteria based on internal performers
Analysis of the three best AEs in the team. Result: perseverance in the face of a long sales cycle, ability to learn a complex product environment independently, constructive handling of rejection, strong alignment with a culture of rapid and transparent decision-making. These five criteria form the basis of the model.
02 .Selecting the appropriate assessment tools for each criteria
Personality test to assess behavioural traits predictive of sales performance. Reasoning test to measure independent learning ability on a technical product. Short role-play scenario involving customer objection management. The combination of multiple sources makes the model reliable and distinguishes a data-based decision from one based on impression.
03.Build the matching score
The results are cross-referenced in a matching model. The final score is a probability of success, weighted according to the organisation's specific criteria and adjustable according to your priorities. It is transparent, explainable and challengeable, which is fundamental for adoption by HR teams and decision-makers.
This model can be replicated in other positions. Each recruitment refines it. And it becomes a long-term strategic asset, far more valuable than a skills catalogue that is updated every year.
The real benefits.
Fewer recruitment mistakes. Faster onboarding, because new recruits are truly suited to the role and culture. Better retention, thanks to verified suitability, not just assumptions. Decisions that managers can defend, based on data rather than intuition.
And above all: less time wasted sorting through profiles based on criteria that don't predict anything.
AI can either increase your workload or help you focus on what's important. The difference lies solely in the method you choose to apply before letting it run.
Without rigorous filtering, generative AI tools produce noise, abundant criteria, generic benchmarks, and decisions that are still intuitive behind a veneer of rigour.
A predictive model rooted in your real-world circumstances allows you to recruit based on evidence, not trends. To select fewer criteria, but the right ones. And to build teams that last.
Want to build your first predictive model?
Request a Key Predict demo and discover how our tools transform your recruitment criteria into reliable decisions.
