Assessment & Talent
When Everyone Looks Excellent: Assessing Talent In The Age Of AI
Translated from the French original. Read the original in French
You receive hundreds of CVs for every position. They are all well written. They all foreground the right skills. They all seem to match the profile you are looking for. And you no longer know who to believe.
The End Of The Signal
This is not a problem with the quality of candidates. It is a problem of signal.
For years, the quality of a CV was correlated with the quality of the candidate. Producing a professional document took time, thought, a real command of one’s own career. The CV was a filter. Those who produced an impeccable file demonstrated, by that fact alone, a form of competence.
That is no longer true.
Artificial intelligence has changed the terms. Anyone can generate an impeccable CV, a tailored cover letter, an impressive portfolio, in seconds. The cost of producing credible information has fallen to zero.
When everyone can look excellent on paper, nobody stands out any more.
The consequence is arithmetic. A company receiving 4,000 applications for a recruitment programme finds itself drowning in a sea of “perfect” files, none of which says who can actually do the work.
What Flipped
The table below summarizes a change many recruiters feel without always putting it into words:
| What used to work | What is happening now |
|---|---|
| A good CV = probably a good candidate | A good CV = the candidate can use AI |
| Read the files to filter | Reading the files filters nothing |
| The challenge: attracting quality candidates | The challenge: identifying the real talent in the mass |
| Assessing = analysing documents | Assessing = putting people in situations |
The problem is no longer finding candidates who look competent. You have thousands of them. The problem is verifying which ones actually are.
That is where tests come in. Not as a “nice to have” in the process, but as the only verification mechanism that still works.
Why Most Tests Fail
You have probably used psychometric tests already. Those instruments have a structural flaw: they ask the same generic questions of everybody. An engineer answers the same items as a management accountant. The result? Abstract profiles nobody really knows how to use.
Take a typical question:
“Faced with a conflict with a colleague, I prefer to:
- A) Avoid the confrontation
- B) Look for a compromise
- C) Firmly defend my position
- D) Ask a superior to arbitrate”
The problem: any reasonably prepared candidate knows B is the “right” answer. The question reveals nothing. It measures the ability to guess what the assessor wants to hear — a useful skill, certainly, but not the one you are looking for.
The Principle Of The Authentic Dilemma
Our experience led us towards a different approach. We start from the real job descriptions to generate situations the candidate will actually live through if they are hired. They are not answering what they “should” do in theory, but what they would do faced with a concrete problem of their future job.
Here is an example for a process engineer position in industry:
You have been in post for 3 months on the firing line. The shift supervisor tells you that thermal consumption has risen by 12% over the last 48 hours with no obvious explanation. The laboratory confirms that product quality remains within specification. The production director wants an update in 2 hours.
What do you do first?
- A) Analyse the history of the kiln parameters over the last 48 hours
- B) Question the operators about recent changes
- C) Check the calibration of the temperature sensors
- D) Prepare a note listing the possible hypotheses for the director
The fundamental difference: there is no universally “right” answer. Every option is defensible. A and B show an analytical approach. C shows methodological caution. D shows a political orientation. The choice reveals how the candidate thinks, not what they can recite.
The candidate cannot “guess” the right answer. They are facing a genuine professional dilemma. And their answer says something real about them.
The Four Dimensions That Matter
Working on large-scale recruitment programmes, we have identified four axes of assessment that produce genuinely informed decisions.
1. How they operate
Not “are they extraverted or introverted” — you will see that in the interview. But: how do they react under pressure? How do they handle ambiguity? Are they a driver or a follower in a group?
An example for an HR position:
You are a junior HR manager on an industrial site. An experienced team leader — twenty years’ service — refuses to apply a new time-recording procedure decided by head office. He says openly that it is “surveillance”, and several operators are starting to listen to him. Your HR director is travelling for three days.
How do you handle the situation?
- A) Wait for the HR director to return and decide together
- B) Call the team leader in for a formal reprimand
- C) Meet him one to one to understand his objections
- D) Send a group email restating that the procedure is mandatory
That scenario reveals several things at once: the relationship to authority, the handling of conflict, autonomy, political sensitivity. A candidate who chooses B has a very different style from one who chooses C. Neither is “better” in the abstract. But one of the two fits your culture better.
2. How they reason
A graduate recruitment programme hires for potential, not for immediate expertise. The question is not “what do they know today” but “how fast will they learn tomorrow”.
We measure the cognitive mechanics: the ability to identify a problem, structure an analysis, spot inconsistencies, synthesize contradictory information.
An example:
The table below shows how maintenance costs have moved on four production lines:
| Line | 2022 | 2023 | 2024 | Change |
|---|---|---|---|---|
| L1 | 120 M | 135 M | 142 M | +18% |
| L2 | 95 M | 88 M | 91 M | -4% |
| L3 | 110 M | 145 M | 180 M | +64% |
| L4 | 85 M | 92 M | 98 M | +15% |
Which line needs investigating first, and why?
- A) L1, because it is the most expensive in absolute terms
- B) L3, because the increase is abnormally fast
- C) L2, because the fall may be masking under-investment
- D) L4, because the increase is close to inflation, suggesting costs are under control
Here we are not testing knowledge. We are observing how the candidate reads data and builds an argument. B and C are both defensible answers, but they reveal different modes of thought. B thinks “anomaly to investigate”. C thinks “risk hidden under apparent good news”. Both are valuable. They are not valuable for the same positions.
3. What they can actually do
A degree is a promise. A technical test is proof.
We verify that the candidate really has command of the fundamentals of their field — not through multiple-choice questions on coursework, but through situations where they have to apply what they know.
An example for a finance position:
You are analysing the accounts of a subsidiary. The CFO sends you the following table:
| Indicator | Y-1 | Y | Change |
|---|---|---|---|
| Revenue | 850 M | 920 M | +8.2% |
| Gross margin | 340 M | 350 M | +2.9% |
| EBITDA | 180 M | 175 M | -2.8% |
| Working capital | 95 d | 112 d | +17 d |
What are the two priority warning signals?
A candidate who has genuinely understood finance will see immediately: gross margin growing more slowly than revenue — margin erosion — and working capital blowing out — a cash problem. One who has merely learned the definitions will miss it.
4. Whether they play fair
An unsupervised online test is an invitation to cheat. We know it, you know it, the candidates know it.
Our approach does not rest on an intrusive webcam. It analyses behaviour: tab switches, abnormal response times, copy-paste patterns. The AI detects the signatures that suggest outside help.
A candidate who switched tabs once for ten seconds generates a weak signal — it was probably accidental. A candidate who left the tab eight times, for two to five minutes at a time, with a copy-paste before each exit, generates a strong signal that justifies investigation.
The candidate knows they are being monitored. They do not know how. That uncertainty is enough to deter most attempts.
What This Changes In The Decision
Tests produce scores. But what counts is what you do with them.
Take a concrete case. You are hesitating between two candidates for an engineering position. Their CVs are equivalent: same school, same internship, same keywords.
| Criterion | Candidate A | Candidate B |
|---|---|---|
| Analytical reasoning | 78/100 | 65/100 |
| Handling ambiguity | 45/100 | 82/100 |
| Technical skills | 72/100 | 74/100 |
| Integrity | 95/100 | 88/100 |
What this tells you:
Candidate A is the more “academic”: strong on analysis, but uncomfortable when the brief is vague. They will be excellent on well-framed assignments, but risk struggling through the first months when everything still has to be built.
Candidate B is the more “operational”: less brilliant on paper, but able to navigate uncertainty. They will adapt quickly to an environment where nobody tells them exactly what to do.
The decision is yours. But you are taking it on real criteria, not on the impression left by a document AI may have written.
The Value Of Feedback
Every candidate assessed can receive a personalized report setting out their profile: results, strengths identified, areas for development, concrete recommendations for progressing.
That report serves two purposes.
For the company: documenting each profile objectively, for the decision and for future follow-up. When you see that candidate again in three years, you will know where they started.
For the candidate: offering tangible value even when they are not selected. A candidate who receives constructive, professional feedback will keep a positive image of your organization — and may reapply with a stronger profile. A candidate who receives a curt rejection email with no explanation becomes a potential detractor.
In a market where the employer brand matters, the way you treat the people you turn down says as much about you as the way you treat the people you hire.
The Change Of Paradigm
The age of AI has made assessment by documents obsolete. A CV no longer proves anything. A cover letter no longer reveals anything. A portfolio can be generated in its entirety.
What remains is the situation. Put the candidate in front of a real problem and watch how they handle it. Not what they say they would do. What they actually do.
That is harder to design. It requires genuinely understanding the positions, creating authentic scenarios, analysing nuanced answers rather than ticking boxes.
But it is the only road left.
When everyone can look excellent, the only way to tell them apart is to make them do the work. The rest is nothing but noise.