Candidate assessment methods have barely changed since 2024. In that time, AI has changed how people apply for jobs, and the Employment Rights Act is raising the cost of a wrong hire. Assessment now needs to predict how someone will perform, not just confirm what their CV says.
Our new insight report, Candidate assessment has stood still, explains what has changed and what to do about it. It draws on the CIPD Resourcing and Talent Planning Report 2026, supported by Omni, and Omni’s Candidate Experience Report 2026. It also includes a practical guide for hiring managers and talent leaders.
Ask most organisations how they choose new hires and the answer is familiar: look at the CV, then interview. The CIPD data confirms it, 96% of organisations use some form of interview, and 68% treat work history or experience as a key decision criterion.
The methods with the best record for predicting performance are still among the least used. Fewer than a quarter of organisations use work sample tests, and only a fifth use assessment centres or critical reasoning tests. Methods that were good enough in 2019 are now carrying more weight than they were built for.
From 1 January 2027, the qualifying period for ordinary unfair dismissal drops from two years to six months, and the cap on compensatory awards goes. Anyone hire from 1 July 2026 will have that protection on the day the change comes in. The government has also dropped plans for a lighter-touch statutory probation process.
A six-month probation is no longer a quiet way out if a hire doesn’t work. A poor selection decision has to be handled through a fair capability process, with uncapped exposure if it isn’t. Most employers expect the Act to push up their costs, and a sizeable share expect to recruit for fewer roles as a result.
Fewer vacancies mean more applicants per role, and technology isn’t solving the volume problem. 57% of employers say it helps screen out unsuitable applicants, but only 13% say it helps to a great extent. That makes candidate assessment the point where confidence in a hire has to be built: before the offer, not at the end of probation.
Generative AI can produce a polished CV, a tailored cover letter and a near STAR answer in seconds. Omni’s Candidate Experience Report 2026 found that 47% of candidates have used AI during their job search. A rehearsed answer to a competency question now tells you more about someone’s preparation that their ability, and an interview that walks through the CV mostly just confirms the CV.
Employers are reacting. A third now monitor jobseekers’ use of generative AI, and three-quarters of those reject candidates for it at least some of the time. Some have moved away from tests with clear right answers towards ones that are harder to game, which can cost predictive accuracy.
The evidence on cheating is more reassuring than the headlines suggest. Test publishers have seen no rise in cognitive ability scores since generative AI arrived. Deterrence and good design do more than detection, and anyone who gets flagged should be retested under supervision rather than rejected outright.
AI scoring has biases of its own. Large language models (LLMs) tend to reward verbosity unless that is designed out, and in one widely reported study, AI models favoured men’s CVs even after names were swapped.
“Human in the loop” can be more reassuring in theory than in practice. If 1,000 candidates complete an AI interview and recruiters review the top 100 to move 50 forward, the AI has effectively made 900 decisions by itself. Candidates notice too: 29% of respondents to Omni’s Candidate Experience Report said they would drop out of a process due to perceived overuse of AI.
Bolting AI onto a weak process doesn’t fix it. You just reach the wrong answer faster.
The fundamentals haven’t changed in 20 years. Does the method predict performance? Is it fair across groups? Does it make sense to candidates? Work samples, simulations, structured interviews and well-chosen behavioural assessments meet those tests better than unstructured conversations or years of experience, and AI tools should be held to the same standard.
Most organisations aren’t checking. Only 7% tract the predictive validity of their methods, and only 19% measure the return on their hiring. When HR and talent acquisition professionals were asked what drives adoption of new AI talent tools, proven results came last. If you don’t know whether your assessment works, you can’t know whether it’s fair or defensible.
When Omni began an RPO partnership with National Highways in 2024, its Early Talent recruitment team was largely manual. Applications were tracked in spreadsheets, and assessment centres ran on Microsoft Teams without a consistent structure.
Starting with our Talent Maturity Assessment, Omni redesigned the Early Talent assessment process into a tech-enabled, fully managed model. This introduced video interviewing, an in-house designed situational judgement test, and structured virtual assessment centres on TopScore’s digital platform with live scoring and automated scheduling. Hiring managers received training and live reporting dashboards. Candidates could request reasonable adjustments, with support for neurodivergent and first-time applicants.
Since September 2024, more than 2,400 video interviews have been completed and 330 candidates have been assessed through structured virtual assessment centres. Across the wider partnership, National Highways has achieved a 98% fulfilment rate across 115 roles, with record offer acceptance rates and reduced candidate drop out.
Read the full National Highways case study.
Organisations will hire less, pay more per hire and carry more risk when they get it wrong, which makes assessment one of the few areas where modest changes can pay off disproportionately. Our full candidate assessment report sets out the latest CIPD and Omni findings, a step-by-step guide for hiring managers and talent leaders, and how Omni’s assessment services can you build a candidate assessment process that is predictive, fair and defensible. If the honest answer to “does our assessment actually work?” is “not really”, 2027 is a good time to find out.
Work samples, simulations, structured interviews and well-chosen behavioural assessments predict performance better than unstructured conversations or years of experience. Combining methods adds further accuracy, as long as each one measures something the others miss.
AI can help with scheduling, admin and consistency checks, but AI scoring has its own biases. Keep decisions with people, review a sample from across the whole ranking rather than just the top, and monitor pass rates across groups at every stage. Before buying a tool, ask for evidence of its reliability, fairness and predictive accuracy.
Set a clear position on acceptable AI use, and ask candidates to actively confirm it rather than tix a box. Favour live, interactive, job-specific tasks that are harder to prepare for with AI. Anyone flagged for unusual responses should be retested under supervision rather than rejected outright.
From 1 January 2027, the qualifying period for ordinary unfair dismissal falls from two years to six months, and the cap on compensatory awards is removed. Anyone hire from 1 July 2026 will be covered on the day the change takes effect. Probation is no longer a low-risk way to correct a poor hire, so employers need more confidence in their candidate assessment before making an offer.
