Measuring cultural fit, without a grade
Measuring cultural fit with a match percentage does not hold up methodologically and raises legal questions. How to see fit through a better conversation.
The second interview is nearly over when someone asks how she deals with a colleague who does not keep to agreements. She gives the answer she prepared at home: raise it directly first, and go to the manager if that fails. A week later the recommendation form shows a 7.4 next to the words culture match, and nobody on the panel can explain where the four comes from.
Measuring cultural fit does produce something, just not what that number suggests. A meta-analysis of 172 studies found that the alignment between a person and an organisation is firmly related to satisfaction, commitment and the intention to stay, and barely related to how someone actually does the work (Kristof-Brown, Zimmerman & Johnson, 2005). A figure turns that general pattern into a verdict on this one applicant, which is exactly the weight it cannot carry. What does hold is the conversation in which you hear how someone works, and where that would rub against the way things go in your organisation.
What are you measuring when you measure cultural fit?
Cultural fit, called person-organization fit in the research literature, is the degree to which what matters to a person matches what an organisation says matters and acts on. That is a comparison between two things, so the outcome depends on how you establish both of them.
There are roughly two routes. In the first you ask the person directly: does this suit you, do you recognise yourself in it. In the second you measure separately what drives someone and separately what the organisation claims to be, and you calculate the gap. The two routes do not give comparable results. In their review of thirty years of research, Kristof-Brown, Schneider and Su show that the perceived route produces far stronger relationships with outcomes than the calculated route, partly because the same person, at the same moment, provides both the judgement about fit and the judgement about their own satisfaction (Kristof-Brown, Schneider & Su, 2023).
This is not a detail for methodologists. It means the same candidate scores high with one approach and middling with the other, while nothing changed in the interview itself. If nobody on the selection panel knows which of the two routes sits underneath the number, nobody knows what is actually being weighed.
Why does a grade for cultural fit not hold up?
The figures in the Kristof-Brown meta-analysis are the clearest argument in themselves. Person-organization fit relates strongly to job satisfaction and to organisational commitment, and negatively to the intention to leave. Its relationship with overall job performance is close to zero (Kristof-Brown, Zimmerman & Johnson, 2005). So fit mainly predicts how someone will feel here and whether they will stay, not what they will deliver.
On top of that, those relationships are not equally strong everywhere. The same 2023 review cites an international comparison showing that person-organization fit effects are stronger in North America than in Europe. A benchmark that works well in an American research dataset is therefore not automatically a benchmark for a Dutch selection panel.
Then there is the question of what you are measuring against. A culture measurement compares a candidate with the people who work there now and with the way decisions are currently made. Score high, and you resemble the incumbent group. Kristof-Brown, Schneider and Su call it an open question whether a high level of fit benefits the organisation itself, particularly where diversity is concerned. For the individual employee fit is generally pleasant; what an organisation gains from everyone resembling everyone else has not been settled.
None of this is an objection to measuring. A DISC profile or a TMA analysis provides language and structure where otherwise only gut feeling sits, and makes discussable what people would otherwise only sense. It goes wrong at one point: the moment a grade lands on the table, the question in the room changes. Within two minutes the discussion is about whether 7.4 is high enough, rather than about what this candidate would set in motion in this team. The score is the beginning, the story completes it.
What does the EU AI Act say about AI in recruitment and selection?
Now that recruitment and selection increasingly run with AI support, there is a legal side as well. The European AI Regulation, Regulation (EU) 2024/1689, classifies AI systems by risk. Annex III, point 4, covers employment and worker management, and item (a) explicitly names AI systems intended for the recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter applications, and to evaluate candidates. Those fall into the high-risk category under Article 6(2).
High risk, in the regulation, means a package of obligations for whoever provides or deploys such a system: risk management, technical documentation, logging, human oversight, conformity assessment and registration. The date on which that package starts to apply has shifted. Regulation (EU) 2026/1744, the Digital Omnibus, entered into force on 27 July 2026 and sets the core obligations for Annex III systems at 2 December 2027 instead of 2 August 2026. The classification itself did not change, only the clock.
In practical terms for an HR department, the question of which of your selection tools will fall into that category has to be answered now, even though the obligations bite later. Without that overview you will not know in 2027 what you are starting on. This is a description of what the regulation does, not a verdict on any specific system; that assessment belongs with your own lawyer and with your supplier, whom you are entitled to ask.
For us it has had a design consequence. We deliberately build without automated decision-making: no score, no ranking, no rejection filter. The human decides, and that decision has to be explainable to the person who was turned down.
What do you hear once the candidate's alarm is off?
There is one more reason the number is shaky, and it has nothing to do with statistics. The material it is calculated from comes from someone who is, at that moment, frightened.
In early August, Nicol Tadema told Dutch recruitment platform Werf& that employers systematically underestimate how nerve-racking candidates find an application process. The discomfort starts weeks before the first interview, she argues: giving up a permanent contract, an unfamiliar culture, the risk of a manager who disappoints. "In your current job you know the rules of the game. A new organisation is always a leap in the dark." The most stressful phase, in her view, is the gap between sending the CV and the phone call, because that is when candidates lose control and their need to be found competent is put directly to the test.
We call such a meeting an introduction, while we are assessing someone whose alarm system is fully on. People in that state give the answer that sounds right, and that answer is exactly the material the number is derived from. The question "how do you handle conflict" has one correct answer that everybody knows. The question "tell me about a time an agreement was broken and it annoyed you, what did you do and what happened next" does not. The second question produces an event, and in that event sits the way someone works. The same principle shows up in the career conversation no one has: the usable information is in what someone recounts, not in what they conclude about themselves.
How do you see cultural fit without giving it a grade?
Instead of a number, you write down what you heard, in four kinds of notes. Anchors: what drives this person and where that comes from. Tensions: where that collides with how decisions actually get made in your organisation, for instance someone used to deciding alone joining a team that runs everything past one more round. Contribution: what they add that is not there now, which is different from what they supplement. And uncertainties: what you still do not know after two conversations, along with the questions that follow from it for the next round.
Those four notes do something a grade does not. They keep the discussion on the person, they can be retold to a colleague who was not in the room, and they can be explained to the candidate who was rejected. They also stay useful if someone is hired: a recorded tension is a topic for the first conversation after three months, whereas a 7.4 stops meaning anything the moment the contract is signed.
There is a price of admission. In mid-August, Peter Boerman described in Werf& how job adverts are now read as closely by competitors and investors as by jobseekers, because an organisation gives away its strategy in what it recruits for. The same holds facing inwards. Your job advert is a public self-portrait, and the honest test is not whether it sounds attractive but whether it matches what someone finds on day 91. Promise room for initiative in the text and require three sign-offs in practice, and what you measure afterwards is no longer fit but the gap between promise and reality.
At NarraTyx we are working on this in the Cultural Fit module, which is still in development and therefore not yet available. What is already possible is the way of looking: placing the story alongside the result, as we describe on the page from job profile to culture match. Your existing instruments stay exactly where they are; this is the layer underneath.
Which question do you ask in a second interview that cannot be prepared for?
Frequently asked questions
Measuring cultural fit means establishing how far what a candidate values matches what an organisation values and acts on; in the research literature this is called person-organization fit. There are two routes: you ask the person whether they recognise themselves in it, or you measure person and organisation separately and calculate the gap. The two routes give different results, so the first question about any number is which route produced it.
Because the relationship you measure predicts something other than what the number is used for. The meta-analysis by Kristof-Brown, Zimmerman and Johnson (2005) shows that person-organization fit relates firmly to job satisfaction, commitment and intention to stay, while its relationship with overall job performance is close to zero. On top of that, the result depends heavily on the measurement approach, and the effects are weaker in Europe than in North America.
Regulation (EU) 2024/1689 places AI systems intended for the recruitment or selection of natural persons in the high-risk category under Article 6(2), via Annex III, point 4(a). That brings obligations such as risk management, technical documentation, logging, human oversight, conformity assessment and registration. Under Regulation (EU) 2026/1744, in force since 27 July 2026, the core Annex III obligations apply from 2 December 2027 instead of 2 August 2026; the classification itself is unchanged.
In four kinds of notes: anchors (what drives someone and where it comes from), tensions (where that collides with how decisions actually get made in your organisation), contribution (what they add that is not there now) and uncertainties (what you still do not know after two conversations, plus the follow-up questions). Those notes can be retold to a colleague who was not present, explained to a candidate who was rejected, and used as conversation topics once someone is hired.
Sources
- Kristof-Brown, A. L., Zimmerman, R. D. & Johnson, E. C. (2005). Consequences of individuals' fit at work: A meta-analysis of person-job, person-organization, person-group, and person-supervisor fit. Personnel Psychology, 58(2), 281-342. https://doi.org/10.1111/j.1744-6570.2005.00672.x
- Kristof-Brown, A., Schneider, B. & Su, R. (2023). Person-organization fit theory and research: Conundrums, conclusions, and calls to action. Personnel Psychology, 76(2), 375-412. https://doi.org/10.1111/peps.12581
- European Parliament and Council of the European Union (2024). Regulation (EU) 2024/1689 laying down harmonised rules on artificial intelligence, Annex III, point 4. Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2024/1689/oj
- European Parliament and Council of the European Union (2026). Regulation (EU) 2026/1744 (Digital Omnibus on AI). Official Journal of the European Union. https://eur-lex.europa.eu/eli/reg/2026/1744/oj
- Diepstraten, P. (2026, 5 August). Nicol Tadema: 'Solliciteren is retespannend; zet het brein van de kandidaat uit de alarmstand'. Werf&. https://www.werf-en.nl/zet-het-brein-van-de-kandidaat-uit-de-alarmstand/
- Boerman, P. (2026, 12 August). Waarom vacatures allang niet meer alleen interessant zijn voor werkzoekenden. Werf&. https://www.werf-en.nl/waarom-vacatures-allang-niet-meer-alleen-interessant-zijn-voor-werkzoekenden/
