Humans still sign off on 85% of AI-assisted HR decisions in Singapore
Source: CNA Tech
A poll of 1,000 employers finds AI taking on more HR legwork, yet people still hold the pen, and firms want clearer guidance on liability.
Roughly 85 per cent of the AI-assisted HR decisions logged in a new poll by the Singapore National Employers Federation still end with a person approving the outcome. The federation catalogued almost 3,000 instances of AI adoption across 1,000 employers questioned in June and July, publishing the results on Friday.
Bigger employers have moved fastest: 81 per cent of organisations above 200 staff put AI to work somewhere in HR, against 65 per cent of smaller firms at or under that headcount. At the federation's first Employers' Summit, its president Kuah Boon Wee argued that oversight should scale with the stakes, with light monitoring for routine paperwork and closer scrutiny where a livelihood is at risk.
Of the logged uses, 73 per cent sit in high-volume chores such as shortlisting CVs and assessing skills. Adoption clusters early in the employment cycle, covering sourcing, personalised training, internal moves and skills-gap mapping, then thins out sharply at the exit end, where whatever remains tends to be analytical rather than decisive. Only 15 per cent of uses let the model decide unilaterally, and even those leave an HR lead or a line manager able to overturn the call.
Asked to rank their worries, 39 per cent feared leaning on AI would dull HR's own capabilities, while about a third flagged when to trust the machine over judgement (36 per cent), who answers for errors (35 per cent) and thin training (31 per cent). SNEF has answered with a pre-scoped package spanning HR strategy, admin automation, job redesign and skills analytics, delivered through named consultants including BDO and PwC, with the Workforce Development Grant (Job Redesign+) picking up as much as 70 per cent of the bill for SMEs and half for bigger companies, against a S$150,000 ceiling.
Bias is the live test case: US suits against hiring-tool makers Workday and Eightfold AI are being watched for what they signal elsewhere. SNEF chief executive Hao Shuo said employers understand their exposure for discrimination, and 84 per cent report strong familiarity with fair-hiring and AI governance frameworks, including the Workplace Fairness Act due by end-2027. Many already review decision patterns, retrain models and log the reasons humans stepped in.
Why it matters for Singapore: The country's claim to be an AI hub rests less on how quickly firms automate than on whether workers trust the systems shaping their careers. A human counterweight on 85 per cent of the use cases surveyed, reinforced by tripartite norms and an incoming fairness law, is the local answer to a question US courts are still working through. Holding that line as the tools get cheaper is the harder test.
