Singapore's AI strategy needs evidence from frontline workflows

Singapore has refreshed its national artificial intelligence (AI) strategy around AI for the public good, setting out 10 priorities and creating a National AI Council to steer the national agenda. 

Its recent moves also include A*STAR's new Institute of Advanced Intelligence and Computing, which brings together capabilities in AI, data and high-performance computing.
These are important foundations. 

But while researching and writing my book, The Psychology of AI Adoption at Work, I kept returning to a less glamorous question: What happens after an organisation acquires the technology and announces the strategy?

The book draws on more than 100 consulting projects and over 50 interviews with leaders guiding AI adoption.

Across that work, I found that the decisive work begins at the frontline. 

AI adoption succeeds when employees can use a tool inside a recurring workflow, understand its limits, and see that it makes a meaningful part of their job better. 

It stalls when leaders count licenses, prompts, training completions or pilot projects and mistake activity for transformation.

That distinction matters for Singapore's public sector. The true test is not whether an agency can deploy an AI assistant. 

It is whether a permit moves faster, a caseworker catches more errors, a citizen receives a clearer answer, or an employee spends less time correcting weak output.

Gallup's latest workplace data in May 2026 illustrates the gap. 

Among employees in organisations that have implemented AI, 65 per cent say it has improved their productivity and efficiency. 

Yet only 14 per cent strongly agree that AI has transformed how work gets done across their organisation. Individual gains do not automatically become institutional gains.

In my research and consulting, the organisations that bridged that gap did five things consistently.

Start with recurring workflows

Agencies should begin with work that happens often enough to test and measure: summarising case files, preparing first drafts of routine correspondence, comparing policy documents, answering common citizen questions, checking forms for missing information, or organising material for a human decision-maker.

Before adding AI, teams should record the current cycle time, error rate, rework and service quality. 

I have seen how easily enthusiasm can distort evaluation. When no baseline exists, almost any pilot can be called a success because participants remember the impressive output and forget the time spent correcting failures.

Assign a named human owner

Every AI-supported workflow needs a person accountable for the result. 

The owner should define what the tool may do, what data it may access, what requires human review and what must never be delegated.

This is not simply a governance requirement. It is also an adoption tool. 

Employees are more willing to experiment when they know the boundaries and know who will help when the system behaves unexpectedly. 

Without ownership, AI becomes an invisible assistant whose mistakes belong to nobody and whose risks become everybody's concern.

Use frontline employees as co-designers

The people doing the work know where delays, workarounds and hidden risks live. 

They should help choose pilot tasks, test prompts, identify edge cases and decide when an output is good enough to use.

One of the strongest lessons from writing my book was that resistance often contains useful operational knowledge. 

An employee who says, "This will not work," may be protecting professional identity or reacting to uncertainty. 

But that employee may also understand an exception, data problem or citizen need that the project team has missed. Treating resistance only as an attitude problem wastes valuable evidence.

Singapore's Public Service has emphasised transformation, innovation and better services to citizens. 

That ambition will produce stronger results when frontline officers help redesign work rather than receiving a finished tool and a usage target.

Build short feedback loops

A useful pilot should run in weeks, not disappear into a year-long transformation programme. 

Teams can test one workflow, review real examples every few days, and document why employees accepted, changed or rejected AI output.

These reviews reveal weaknesses in source data, instructions, tool configuration and employee training. They also create psychological safety. 

Staff learn that identifying a confident but incorrect answer is not evidence that the pilot has failed; it is part of making the workflow safer.

Measure outcomes citizens and employees can feel

Each pilot should use a small scorecard: cycle time, errors, rework, employee effort, citizen satisfaction and the share of cases requiring escalation. 

High-risk workflows should add measures for fairness, privacy incidents and successful human intervention.

Agencies should also compare results across roles and experience levels. A tool that helps experienced analysts but confuses newer staff may widen internal capability gaps even when average productivity rises. 

The purpose of measurement is not to produce a flattering average. It is to understand who benefits, who struggles and why.

Finally, agencies should share reusable lessons. A short internal record of what worked, what failed, which safeguards mattered and which metrics changed would help other teams avoid repeating the same mistakes.

Singapore already has the strategic ingredients for significant AI progress: national coordination, strong digital institutions, research capability and a public-service culture focused on implementation. 

The next discipline is to connect national ambition to frontline evidence.

That bridge should be built one workflow at a time. 

When agencies measure real outcomes, involve employees and preserve accountable human review, AI becomes more than a technology programme. It becomes a practical method for improving public service.

Adapted from: The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026).

-------------------------------

Dr Gleb Tsipursky is a behavioural scientist, consultant and author of The Psychology of AI Adoption at Work: From Resistance to Results (Georgetown University Press, 2026). He helps leaders address the human factors that determine whether AI initiatives are adopted, trusted and translated into measurable workplace results. He is based in Columbus, Ohio, USA.