Will training 80,000 staff in AI work?

On 24 September 2026 Singapore's banks, insurers and asset managers committed to train their entire Singapore workforce, more than 80,000 people, in critical AI skills by 2028, and the Institute of Banking and Finance says more than half have already been through it. The group covers a large share of a sector that is itself a large share of the economy. Our view is that the pledge is a good decision taken for a reason that is not productivity. The surveys that ask whether training turns into output are less encouraging: Culture Amp found 71 per cent feel more productive with AI while reported workloads barely move. Skills are the foundation of the work, not the result of it.

Deputy Prime Minister Gan Kim Yong, who chairs the Monetary Authority of Singapore, launched the IBF AI Workforce Co-Lab at the Institute of Banking and Finance's Distinction Evening on 24 September 2026. The institutions, including DBS, OCBC, UOB, AIA Singapore and Prudential, have committed to training their entire Singapore-based workforce in critical AI skills through IBF-recognised programmes by 2028, and Mr Gan said more than half have already been trained. IBF chief executive Carolyn Neo said the pioneer institutions employ a large share of the financial services workforce. "We want AI to translate into better careers for Singaporeans and a more productive workforce," Mr Gan said, adding that AI is already changing customer service, financial advisory, credit underwriting, fraud detection, market analysis and software development.

The Co-Lab begins with three groups rather than everyone at once. Leaders learn to identify valuable AI applications, set governance and lead the workforce change. Wealth managers get courses combining AI skills with client engagement and advisory expertise, with private banking employers committed to the pathway. Operations staff, the largest group, work on process improvement, risk oversight and validating what AI produces. Mr Gan was explicit that this will mostly mean wider responsibilities in an existing role, and sometimes a move to a related or new role. That detail matters more than the headline number, because it describes changing the shape of jobs rather than adding a certificate to the wall.

Culture Amp's first AI at Work benchmark, published on 17 September 2026, drew on a large sample of organisations and employees. Seventy-one per cent said AI tools helped them feel more productive, and the figure was highest among the heaviest users. Yet workloads barely differed across levels of use: the share describing their workload as reasonable was almost the same for power users and for people who did not use AI at all. The lowest-scoring item in the whole benchmark was direction. Few leaders had clearly explained how AI would be used to reach company goals, well below the share who said their organisation encourages experimenting with it, and awareness of internal career opportunities had fallen since the previous year.

Visier's report on 23 September 2026, from a survey of US full-time employees, adds a sharper edge. Forty-eight per cent said they had exaggerated their AI usage or expertise to colleagues or leadership, and many said they felt pressure to use AI at work when they were not confident doing so. Most said their role or career had changed significantly because of AI in the past two years, and a majority were concerned about its effect on their careers. These are US figures and we will not pass them off as Singaporean ones, but the mechanism is not country-specific: when a workplace signals that AI use is expected, part of what it then measures is the signal coming back.

A small firm can take the structure without the scale. The Co-Lab separates people by role instead of running one general course, and it starts with leaders. Read plainly, the point is that someone has to decide what the freed time is for before training can pay for itself. It also names the unglamorous half of the work, checking and validating AI output, which is a skill rather than an afterthought. The version of this for a team of five is a short list of processes with one named owner each, and a written rule about what must be reviewed before anything leaves the building.

We think the Singapore pledge is a good decision taken for a reason that is not productivity. Giving most of a sector's workforce a common foundation in AI is worth doing on its own terms, because the alternative is uneven, self-taught use with no shared standard for what a checked output looks like. But national training programmes have a habit of being counted in completions, and completions are not output.

Where we are unconvinced is the step from training to result. Culture Amp's own numbers, 71 per cent who feel faster and workloads that do not move, are the honest picture of what training alone buys. The banks that get value from this will be the ones that, having trained everybody, go back and redesign or delete the work that never needed a person. That second step is the harder one, and it is not in the pledge.