How to Assess Soft Skills Before You Hire (Without Guessing)

The reliable method is boring. Ask every candidate the same questions in the same order, score each answer against a grid written before the interviews start, and put a standardised behavioural assessment underneath it. The method that keeps failing in the research is the one most companies rely on: a free-flowing conversation tells you how comfortable somebody is in a conversation.

The paradox at the centre of hiring

Ask a company today whether soft skills matter and you'll get a yes. Ask how they measure them and most will point you at a personality test.

In a survey of 1,092 French HR managers, 72% reported using personality tests in recruitment, and 92% considered them relevant or highly relevant to the decision.

*Source: NEOMA Business School, survey of 1,092 HR managers.

Which is awkward, because a systematic review of the international literature on that same question found that personality predicts salary, job retention and career satisfaction very modestly, if at all.

*Source: Chaire Nouvelles Carrières, NEOMA Business School — H.R. Insights #03.

So the tools the market trusts most aren't measuring what they're being asked to measure. That doesn't mean soft skills can't be assessed. It means the standard way of doing it is the problem, which is a much more fixable situation than the first sentence suggests.

Why don't standard interviews assess soft skills?

Left unstructured, an interview mostly measures interview skill. That tracks with confidence and social ease. It says very little about how somebody behaves under pressure six months into a mission, which is the part you're actually buying.

There are three failures here, and all of them are documented.

It measures the wrong thing

In a conversation with no structure, the candidate who performs best is the one who's had the most practice at conversations. That's a real skill. It just isn't the skill most roles are hiring for.

It isn't comparable

Give five candidates five different conversations and you haven't compared five candidates. You've compared your impressions of five different conversations, and those impressions are shaped by the order you saw people in, what kind of morning you were having, and how much somebody reminded you of a hire who worked out.

It rewards knowing what you want to hear

In a recruitment setting, self-reported personality assessments reveal less about a candidate's actual characteristics than about how well they've read recruiter stereotypes. Presenting as conscientious, open and agreeable is the obvious safe answer, so most people give it. You would too.

*Source: Chamorro-Premuzic et al. (2010), Educational Psychology 30(2), 221–241.

None of this is an argument against interviewing. It's an argument against interviewing without structure.

What do the standard tools actually predict?

Less than most people assume, and the numbers are worth having in your head before you build a process on top of them.

The Big Five

The most scientifically validated personality model we have: five stable dimensions, thousands of published studies, the dominant reference in occupational psychology. And precisely because it's been studied so heavily, its limits are well mapped. It explains roughly 10% of the variance in job performance.

*Source: Barrick & Mount (1991), Personnel Psychology 44.

General cognitive ability tests

These do considerably better, at around 26%.

*Source: Schmidt & Hunter (1998), Psychological Bulletin.

The MBTI

The most widely used and the weakest of the three. Between 35% and 50% of people receive a different four-letter type when they retake it within weeks. Converting continuous traits into binary categories throws away 26–32% of the information per scale. And its own publisher states plainly that the instrument isn't designed to predict job performance; it's positioned for personal development and team communication.

*Source: Pittenger (2005), Consulting Psychology Journal 57(3), 210–221.

How to read these figures

The 10% and 26% are r² values from meta-analyses, meaning the share of performance variance a tool statistically accounts for. They're not a verdict that these tools are useless. Cognitive ability at 26% is one of the strongest single predictors anybody has. The point is narrower than "tests don't work": if you're using a personality test to predict how someone will work, you're making the whole decision on about a tenth of the picture.

What actually carries the weight?

Behaviour does, and there's a specific number attached to it.

253 experts — researchers, managers, HR professionals, occupational health specialists — were asked to distribute 100 points across the dimensions of individual work performance. They put **63% of the weight on behavioural dimensions**: contextual performance, adaptive performance, and the absence of counterproductive behaviour. Task performance itself took 36.

*Source: Koopmans et al. (2014), Work 48(2), TNO-VU University Amsterdam. Peer-reviewed.

What that figure is and isn't

It's an expert weighting exercise rather than a predictive correlation. These practitioners were asked what they weigh when they evaluate an employee, not what statistically predicts output, and the distinction will matter if you're ever arguing about it with a researcher. What makes it compelling anyway is who answered: people who evaluate employees every week, not theorists.

There's a second finding in the same literature that tends to get skipped. In the US Army's Project A study, covering 4,039 soldiers, cognitive ability predicted technical performance well (r = .33) but had almost no bearing on contextual performance (r = .08). Dependability ran the other way: .30 on contextual, .11 on technical.

Different traits predict different halves of the job. Testing for one half and assuming it covers the other is the structural mistake, and it's a very easy one to make because the first half is the half that's easy to test.

Which soft skills should you actually assess?

Not the ones off a generic list. The ones that predict performance in this role, on this team, in this context.

The narrowing exercise that works: before writing the job description, write down how the person will need to work.

- Autonomy. Will they be handed a problem or a specification? How much ambiguity is normal here?
- Decision-making pace. Does this environment reward fast calls or careful ones?
- Response to conflict. What actually happens on this team when two people disagree?
- Feedback. How direct is it, and how quickly is somebody expected to act on it?

That last one matters more than its reputation suggests. In the Leadership IQ study tracking 20,000 hires across 312 organisations, 89% of hiring failures were attributed to behaviour rather than technical skill. The single leading cause was coachability — the inability to take feedback and adjust — at 26%, followed by emotional intelligence at 23%, motivation at 17% and temperament at 15%. Technical competence accounted for 11%.

*Source: Leadership IQ, [Why New Hires Fail]

⚠️ Worth stating the limits on that one. Leadership IQ is a private consultancy, not a peer-reviewed journal, and the failure causes were categorised by their researchers from manager reports. Treat the 89% as a strong directional signal rather than a precise measurement.

How do you structure an interview to assess behaviour?

Same questions, same order, same scoring grid, every candidate. Structure is the only thing that makes candidates comparable to each other instead of to your memory of the last one.

Four things do most of the work.

Ask about situations, not qualities

"Are you autonomous?" invites a yes. "Tell me about a time you had to make a call without being able to check with anyone. What did you do?" invites evidence, and evidence is harder to fake on the spot.

Decide what a good answer looks like before asking the question

Write down in advance what a strong, an adequate and a weak answer contains. Do it afterwards and you're not scoring, you're rationalising an impression you already had.

Score immediately, question by question

Not at the end. One strong answer late in the conversation will quietly rewrite the memory of everything that came before it.

Ask everyone the same thing, including the obvious yes

It's tempting to relax the process for a candidate who's clearly great. That's exactly the moment it matters, because the impression about to be acted on is the specific thing structure exists to double-check.

What are personality tests worth in recruitment, then?

They're not worthless. Standardised assessments produce results that can be compared across candidates, which unstructured conversation simply cannot do, and that's a genuine advantage worth keeping.

The problem is what gets done with the output. A behavioural assessment tells you where to look. It generates a hypothesis. The interview is where that hypothesis gets tested. Treating the result as a verdict skips the entire part where you find out whether it's true for this person in this role.

There's also a distinction the market tends to blur, and it's worth being precise about. A personality test measures stable traits claimed to hold across every context. A behavioural assessment measures **how somebody acts in situation**: under pressure, in groups, facing uncertainty with incomplete information. The second is contextual by design, which is why it can say something useful about a specific role rather than about a person in the abstract.

That distinction is the basis of the Migbirds approach. Our assessment, the IDC (Inventaire de Dynamique Comportementale), measures behavioural dynamics rather than self-declared traits, and the matching is built on the three dimensions of person-environment fit established in the research literature.

*Kristof-Brown et al. (2005), Personnel Psychology 58 — a meta-analysis finding that behavioural fit predicts satisfaction, engagement and retention, with turnover reduced by 37% and promotions multiplied by 1.8.

How do you involve the team in the assessment?

The people who'll work with this person daily will notice signals a recruiter never sees. But only if they're handed something specific to look for.

"What did you think?" produces "seemed nice", and "seemed nice" is an impression, which is the thing this whole process exists to replace. Give each person one dimension instead. One assesses how the candidate handles disagreement. Another assesses how they explain something technical to somebody outside their field.

Then keep the assessments sealed until everyone has submitted. The second one person says "I loved them", nobody else is assessing the candidate any more. They're assessing whether to agree.

What to do with all of this

Assessing soft skills isn't a question of finding the right test. It's a question of making the process structured enough that you end up comparing candidates instead of comparing conversations.

Three things, in rough order of how much they change:

1. Write down how the person will need to work — before writing down what they need to know
2. Use the same structured questions and the same scoring grid for everyone, including the obvious yes
3. Add a standardised behavioural assessment to generate the hypotheses the interview then tests

None of it eliminates hiring risk. What it does is move the decision from something that was felt to something that can be explained: to a colleague, to the candidate, and to yourself in six months when you're trying to work out what went wrong.

Assess behaviour before you hire, not after

Migbirds evaluates soft skills, motivation and behavioural fit to anticipate whether a person and a mission will actually work together. Before the placement, not after it.

Create a company account