Understand what AI replaces, not whom

HERE is a quiet test for jobs that might survive the artificial intelligence era: Do you wash your hands at the end of the working day?

If you do, it is probably for one of two reasons: You work with your hands – as an electrician, plumber, chef or landscaper – and the grime of craft accumulates in the day.

Or, you work through people, as a leader, adviser or relationship manager, with trust as the medium of exchange.

Both groups may be more resilient to AI-driven change, but neither is immune. Yet, a third category in our workforce does neither, and faces the most immediate disruption.

The recent analysis by multinational advisory company Willis Towers Watson (WTW) moves the automation debate from “which jobs will vanish?” to “which activities will change?”

By deconstructing more than 900 occupation profiles representing over 50,000 jobs from the US Bureau of Labor Statistics O*Net database at the task level, it provides evidence for organisations to act on.

The core findings: 27 per cent of all work activities are amenable to process automation, which involves the substitution of rules-based, repetitive cognitive work with software.

A further 10 per cent can be taken over by AI, specifically reasoning-based systems that improve decision quality, and not merely scripted automation.

Most strikingly, 13 per cent of all roles have more than half of their activities exposed to AI and process automation combined.

These are not marginal jobs. They include accountants, loan officers and compliance professionals; the analytical backbone of organisations around the world.

The limits of the handwashing test

The handwashing test is useful, but not foolproof. Even physical work is increasingly surrounded by automation.

A roofer still climbs, measures and seals, but inspection, routing and scheduling are being reshaped by drone fleets with AI-powered imaging.

This involves surveying entire districts, identifying issues before they become failures and directing crew with algorithmic precision.

Across trades, robotics and wearables can augment strength and reduce injury risk. On construction sites, exoskeletons also generate performance data that feeds optimisation models.

The hands remain necessary, but the workflow around them changes substantially.

The issue therefore is not assessing whether manual labour is immune – it is not – but understanding the tasks that require the skill, dexterity and improvisation which make automation unreliable and uneconomic today.

Yet, that calculus may very well change tomorrow.

The wrong unit of analysis

For decades, workforce planning has treated the “job” as its primary unit. With AI, the wrong reflex is to ask: Which jobs will it replace?

The more relevant focus should be on tasks.

Within almost every role, a mix of tasks exists. Some demand judgment, creativity and relational trust, while others are repetitive and data-intensive, exactly the area in which automation excels.

WTW’s analysis suggests that AI typically augments 20 to 35 per cent of tasks in professional and judgment-intensive roles, improving decision quality rather than replacing work wholesale.

The future of the workforce will be defined not by job elimination, but instead by task reconfiguration and human-machine integration at scale.

This requires a deliberate deconstruction of every role into component tasks – classifying what can be automated or AI-augmented, and what must remain irreducibly human – and then reconstructing roles, workflows and teams accordingly.

Companies such as Moderna have made the structural change by merging human resource management and technology into a single leadership function, signalling that in the AI era, people and technology are inextricably linked.

The cost-trap can prove expensive

In high-cost labour markets – the US, Western Europe, Singapore and Australia – the business case for automating knowledge and physical work is immediate.

More than half of the tasks in structured operational roles being susceptible to automation and AI augmentation is not an abstract risk; it demands a budget-optimisation decision.

However, cost-only logic may prove pound-foolish.

One company running back-office business-process outsourcing in South Asia automated a significant portion of its work with large language models (LLMs) to save costs.

But when AI token costs rose, human processing became cheaper. With the workforce dismantled and institutional knowledge dispersed, rebuilding proved expensive.

Klarna’s experience offers a similar lesson.

Between 2022 and 2024, the Swedish fintech company reported an estimated US$40 million profit improvement, after replacing some 700 customer service agents with an AI-powered assistant.

But by mid-2025, Klarna was rehiring.

The AI handled volume efficiently, but it could not manage the nuanced, emotionally charged interactions that define financial services at its margins.

Klarna’s CEO Sebastian Siemiatkowski acknowledged this, saying: “We focused too much on efficiency and cost. The result was lower quality, and that’s not sustainable.”

By early 2026, it had rebuilt around a hybrid model: AI for the routine tasks, humans for the complex. Its customers wanted to speak to people, not chatbots.

Neither case argues against AI adoption. However, both argue against treating it purely as a cost-arbitrage play.

The more durable rationale, that does not reverse when economics shift, is capability-building.

A board question

The board’s fiduciary responsibility must include oversight of all enterprise-wide intelligence, be it natural or artificial. Stewardship of human capital is a central board-level responsibility.

Allowing AI to reshape the workforce without oversight on capability-building and change leadership, as well as ethical and reputational risk management, is a strategic failure.

A structural shift is already under way.

WTW has found that in several leading companies, the mandate of remuneration and compensation committees has expanded to include talent management, succession planning, organisation development and increasingly, workforce transformation.

Boards should put workforce transformation squarely on the risk register, requiring management to report not just headcount efficiency, but also skills investments – the capabilities that are being built for the next five years.

Indeed, progressive compensation and talent committees are asking HR teams to prepare a task-level AI and automation exposure map for mission-critical roles, as well as top roles by headcount cost.

If management cannot produce this, that is a governance issue in its own right and a gap to be addressed immediately.

“How much money will AI save us?” is a relevant question.

However, boards must also understand the question of “what will we lose if we get this wrong?”

What only humans can do

An observation popularised by Bill Gates is that we overestimate what technology can do in a year and underestimate what it will do in a decade.

On AI and workforce transformation, many organisations are committing both errors simultaneously.

When such a large portion of knowledge-based roles face task-level disintermediation and substitution, the window for proactive governance is not a distant horizon. It may be as imminent as the next budget meeting.

The handwashing test was always intended as a provocation, not a prophecy. The roofer’s workflow is being reshaped by drone analytics. The analyst’s drafts are being accelerated by LLMs.

What endures in every sector, labour market and cost structure is the premium on judgment, relational intelligence and accountability.

It is work that machines still struggle to perform reliably, cheaply and safely at scale.

For boards, the imperative is to govern the transition now: Understand task-level exposure honestly and invest in skills that compound and redesign work, so that people spend their time doing what only humans can do.

The writer is senior managing director and global leader of board advisory at WTW. He is also the chair of the sustainability committee at the Singapore Institute of Directors.