In three minutes, I generated three CVs for three unrelated roles: IT project manager, GTM specialist, customer success manager. I am none of those things. Each of those applications would comfortably pass the first stage of most hiring processes.
That is the environment Talent Acquisition teams work in today. Hundreds of AI tools promise candidates a perfect CV tailored to your job ad, automatic applications to hundreds of roles, live interview assistance and real-time answers to technical tests. All of it in minutes, and largely undetectable.
This creates two problems.
Problem one: noise
In a 2025 ResumeBuilder survey, 78% of candidates said they had used AI to write their CV or cover letter. If your process filters on CV screening, cover letters, phone screens, take-home assignments or online tests, there is a good chance you are no longer evaluating the candidate. You are evaluating their tools.
Signal detection theory, formalised by Green and Swets in 1966, describes this precisely. A radar operator must decide whether a faint trace on the screen is an aircraft or interference. Two things decide the outcome: their sensitivity, and the threshold at which they call it a signal. Hiring works the same way. The signal is a candidate’s real ability to perform in the role, and we only see it filtered through CV writing skills, keyword optimisation, interview coaching and now generative AI.
When uncertainty rises, people do not just become less accurate. They change strategy. Some recruiters turn conservative and reject faster, missing strong candidates. Others turn permissive and let more through, raising the risk of poor hires. Both are reactions to the noise rather than decisions about the candidate.
Problem two: volume
AI has not only made applications look better. It has made far more of them appear. LinkedIn data from 2025 shows applications up 45% year on year. Applying now takes two minutes: find the role, click, generate everything, send.
Herbert Simon’s work on bounded rationality explains what happens next. Time, attention and information are finite, so when the options multiply, we adapt. Maximisers review everything to find the best candidate, and hit fatigue and diminishing returns. Satisficers take the first profile that looks good enough, saving energy but missing stronger people further down the pile. At hundreds of applications per role, both strategies cost you talent.
AI did not break hiring
It is tempting to blame the technology. But AI mainly exposed something already true.
Last year’s CIPD data on UK selection methods shows previous work history is still the most used filter, ahead of CV-based interviews and application screening. Decades of predictive validity research rank these among the weakest predictors of actual job performance. The methods we rely on most are the ones telling us least.
In an ideal world you would combine many methods. In the real world, recruiters, budgets and candidate patience are all finite. While you are still assessing, someone else is making an offer.
The signals are shifting too. The World Economic Forum’s Future of Jobs report estimates that around 44% of the core skills required for jobs today will change by 2027. Hiring purely on current skills is short-term thinking. What matters is how fast someone learns and adapts as the role evolves.
Signal compression
So the equation looks impossible: more information about candidates, earlier, in less time.
The way out is signal compression: rather than adding steps or collecting more data, you use science-based assessment to capture the most predictive signals for a role in fewer, far more informative data points. Personality, motivation and cognitive ability, measured properly, say more about future performance than a CV ever will, and no prompt can generate them.
The effect on the hiring pipeline is immediate. Filtering happens on evidence rather than presentation, so every minute recruiters and hiring managers spend afterwards goes to candidates who really fit the role.
Candidates gain as well. A CareerBuilder survey found 58% abandon processes with more than three stages. A compressed process gives them an early, objective signal on their fit, instead of discovering after four rounds that the role was never right.
What it delivers
Companies applying this approach with AssessFirst report hard outcomes: hiring errors cut fourfold at Wavestone, team performance on new hires up 18% at Punch Pubs, turnover halved at Lacoste. The method takes around 30 minutes of a candidate’s time and none of the recruiter’s, scores 4.6 out of 5 on satisfaction and is completed by 86% of applicants. In a world where anyone can produce the perfect application, the smartest companies stop filtering first. They assess first.