This article is Part 1 of a two‑part series on AI in Tax.
Part 1 focuses on what AI in tax is and where it creates credible value
Part 2 explores governance, ROI discipline and practical use cases
Artificial intelligence (AI) is changing how tax and finance teams research issues, document positions, manage data and deliver work at speed. The opportunity is real, but so are the risks. In tax, where advice must be defensible and outcomes can have regulatory, financial and reputational consequences, organisations need more than powerful tools: they need governance, controls and the right operating model to use AI safely.
Recent market commentary often suggests that AI will rapidly automate professional work. In practice, most organisations are finding that moving from promising pilots to reliable, repeatable deployment is harder than expected, particularly in high-judgment domains such as tax. The differentiator is not whether you adopt AI, but whether you can adopt it in a way that is controlled, explainable and aligned to your governance and risk appetite.
For many tax leaders, the right mental model is: AI can accelerate drafting and analysis, but it still requires careful supervision and clear accountability. Used well, it frees teams from repetitive, low-risk work and improves consistency. Used poorly, it can introduce errors, confidentiality issues and weak audit trails, exactly where tax teams need strength.
Singapore’s policy direction also reinforces why tax and finance leaders should treat AI adoption as a near-term agenda item. In Budget 2026, the Government signalled a strong push to build practical AI capabilities across the workforce, starting with the accountancy and legal professions, alongside broader measures to encourage enterprise adoption. For tax functions, this points to a future where AI-enabled ways of working become more common, and where clients, regulators and boards will increasingly expect strong governance, documentation and accountability for AI-assisted work.
This has also been reflected in recent local reporting, including Channel News Asia’s coverage of Budget 2026’s focus on equipping professionals with practical AI skills and placing trust and human accountability at the centre of Singapore’s AI agenda. For organisations, the implication is clear: capability-building and control design need to move in parallel, so productivity gains do not come at the cost of confidentiality, quality or defensibility.
A useful local leadership perspective comes from DBS Group CEO Tan Su Shan, who has commented publicly on the possibility of AI reshaping even senior leadership roles. The practical takeaway for tax leaders is not whether a role can be “replaced”, but that AI changes how work is produced and reviewed, and that clarity of accountability becomes more important, not less. In high-stakes areas such as tax, governance, documentation and human sign-off remain essential.
There is no question that AI can help with speed and scale in parts of the tax lifecycle. However, many organisations are still working through the practical reality that moving from a proof-of-concept to dependable production use requires strong data foundations, clear process ownership and well-designed controls.
Across industries, experience to date suggests a familiar pattern: quick wins are common in stand-alone, low-risk tasks, while enterprise deployment in complex workflows tends to take longer and deliver value only when organisations redesign processes around the technology and put governance in place.
One useful analogy comes from software development, another domain where AI is heavily discussed. Several studies have found that while AI can reduce time spent on first-draft output, experienced practitioners often spend significant time prompting, waiting, reviewing and correcting what the model produces. In other words, the “review and validation” workload can be the limiting factor, and tax work has similar validation requirements.
At the macro level, the picture is similarly mixed. While adoption is accelerating in many organisations, the impact on measured productivity and workforce demand varies significantly by role and by how well AI is integrated into processes. For tax leaders, the practical implication is to focus less on predictions and more on the fundamentals that determine whether AI improves quality and turnaround time in your own environment.
Other research points to the same pattern: AI can be handy for simple, low-context tasks, but it breaks down precisely where professional tax work really starts. That is not a quirk of current models; it is a structural issue with how these systems operate.
A frequently cited challenge in enterprise AI is the “pilot-to-production” gap: general-purpose tools may help individuals draft, summarise or search faster, but embedding AI into end-to-end workflows that deliver measurable business outcomes is more difficult. Research associated with MIT’s Project NANDA (including its report The GenAI Divide: State of AI in Business 2025, July 2025) highlights this gap and points to common blockers such as brittle workflows, weak integration into day-to-day operations, incomplete data sets, and limited ability for deployed systems to learn from feedback and improve over time. For tax, these limitations matter because the most valuable work is rarely a single isolated task, it is a chain of steps that must be consistent, controlled and reviewable.
The MIT research shows that AI improves productivity mainly in low-skill tasks: transcribing meetings, smoothing text for people with weaker language skills, or handling very simple customer-service interactions. For high-skill work where accuracy is essential, error rates and the risk of hallucinations mean the human time required to catch and fix errors reduces, or even wipes out, any anticipated gains. That is a particularly acute problem in fields like tax, where a single error can have significant regulatory, financial and reputational consequences.
The METR trial (Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity - METR) gives you the micro-mechanics of this dynamic. When experienced developers used top-tier AI tools, they spent less time "doing the work" and more time prompting, waiting, reviewing and cleaning up AI output. They accepted less than half of AI suggestions, typically after careful inspection, and 75% reported needing to read every line of AI-generated code. That is exactly the sort of unglamorous but critical validation work that dominates serious tax practice: checking every figure, reading every paragraph, and anticipating how a regulator or court might interpret what you have written.
Tax advice, especially in a jurisdiction like Singapore, combines statute, case-law, IRAS positions, treaties, commercial context, governance constraints, ethical considerations, and client-specific risk appetite. It also requires a feel for how authorities are likely to react to particular structures or arguments, and an ability to explain tradeoffs to boards, shareholders and founders. That is not a tidy multiple-choice exam; it is closer to the "large, complex, mature codebase" setting where the METR team found that AI slows experts down rather than helping them. In that environment, AI can be a useful tool, but it is very far from being a substitute.
You might think of it this way: AI is quite good at suggesting a generic, textbook answer to a generic, textbook question. Real client work in tax is about recognising why your client's situation is not generic. Textbook answers would be dangerously misleading in their particular case.
When considering AI in the tax, questions remain consistent: what is defensible, what is controllable, and what is appropriate for your risk appetite and governance environment.
Our Tax leaders can help you assess the tax and risk implications of AI-enabled processes, identify priority use cases where AI may improve turnaround time and consistency, and design practical review and documentation expectations so that AI-assisted work remains auditable and fit for purpose.
As investment in AI accelerates, expectations rise alongside it, boards and management teams want faster cycle times, better consistency and clearer insight, without creating new categories of risk. For tax functions, that makes it important to be disciplined about where AI is used, how outputs are controlled, and how value is measured beyond early pilots.
Employee equity incentives, including ESOPs, stock options, restricted share units and their variants, are now firmly embedded in modern talent strategy. Across sectors and jurisdictions, equity is used not only as a form of remuneration, but as a mechanism to align interests, retain key talent and drive long-term value. The concept of shared ownership is compelling. Delivering on that promise, however, depends on the robustness and coherence of the legal and tax framework supporting these arrangements.
Employers in Singapore need to be aware of the key legal and employment tax issues when handling termination scenarios, including routine resignations, performance-related departures, and workforce restructuring. Understanding these considerations is essential to navigate the complexities of ending employment relationships effectively and responsibly.
