HomeThe Last Reconciliation
Somewhere in Singapore right now, a finance analyst is doing something that will not exist as a job function in five years. She is pulling data from three systems, cross-referencing line items, flagging discrepancies, and sending a report that took four hours to produce to a CFO who will read it in four minutes. Accurate work. Meticulous work. The kind of work that built entire finance departments and shaped the professional identity of an industry.
It is also, with high probability, the last generation of humans whose primary job description is reconciliation.
The tools that automate transaction matching, flag variances, generate first-draft management accounts and cross-check regulatory submissions against source data already exist, are already deployed in leading finance functions globally, and are already producing the productivity gains that make the business case for redesign unanswerable. The question for Singapore's finance community is not whether this shift is coming. It is whether institutions and individuals are redesigning fast enough to stay ahead of it.
Customer service fell first to AI at scale, then operations, then elements of legal and compliance. But finance is distinctive. It is the function closest to the number that governs everything else — the P&L, the balance sheet, the cash position. It carries explicit board and regulatory accountability. And it is the function where the gap between what AI can now absorb and what humans genuinely need to own is widest, and most consequential.
A finance professional working at a glass desk with a dual-monitor setup, abstract data visualisations on screen, cinematic natural light filtering through floor-to-ceiling windows, dark navy interior, shallow depth of field, no text or logos
The Core Shift: Processing to Judgment
A traditional finance team spends a disproportionate share of its time on one thing: processing. Transaction matching. Invoice verification. Period-close reconciliation. Variance reporting. Regulatory return preparation. These tasks share a common profile — high-volume, rules-bound, document-heavy, with measurable right answers. The input is structured data; the output is a verified or flagged state. No ambiguity about what good looks like, and no special judgment required to produce it.
This is the profile AI excels at. Not because large language models are uniquely suited to numerical work — early versions made maths errors that became industry jokes — but because the ecosystem of AI tools built specifically for finance is now far richer than the public narrative suggests. Agentic workflows that reconcile at transaction level from ERP data, models fine-tuned on regulatory filings, automated exception-management systems that surface only the genuinely anomalous: these are deployed at DBS, OCBC, UOB and their global peers today, not in pilot programmes.
The WEF Future of Jobs 2025 report projects, on an approximately reported basis, around 170 million new roles created globally by 2030 and around 92 million displaced. Finance is not uniquely exposed in that projection — but the tasks most vulnerable to displacement look exactly like the processing layer of a finance department. High-volume, codifiable, rules-bound, verifiable.
What is not being displaced is the judgment layer. The CFO deciding whether to draw down a credit facility in an uncertain demand environment is not processing data. She is synthesising incomplete information, weighing probabilities she cannot precisely quantify, and making a call for which she is personally accountable. No AI produces that judgment. What AI can do is give her dramatically better data, faster, so the judgment she exercises is better informed and more precisely targeted.
This is the shift: not replacement, but rebalancing. The finance function's centre of gravity is moving from the processing layer to the judgment layer — analysis, interpretation, business partnering, risk assessment, governance. That movement is being driven by economics: maintaining a large processing-layer workforce is increasingly difficult to justify to boards who can see what the alternative costs.
The Professionals Under Pressure
The cohort facing sharpest pressure is not junior or senior. It is the middle: analysts with three to seven years of experience whose value proposition rests on producing processed outputs reliably. They know the ERP. They close the books efficiently. They prepare the management accounts.
The problem is not that they do this work poorly. The problem is that AI is beginning to do it better — faster, at higher volume, with fewer errors. The displacement pressure is real, and it does not serve anyone to euphemise it.
What the middle of the stack genuinely has — and what AI does not — is contextual intelligence: knowing that the variance in the Asia operations figure is a timing difference, not a structural trend; that the CFO needs a sensitivity analysis for a board presentation next Tuesday; that a difficult conversation about a budget overrun should happen early, not late. The question is whether the individual is building that contextual depth, or spending most of their working day on processed outputs that AI could now produce more efficiently.
The answer determines whether this shift creates opportunity or displacement.
"AI does not eliminate the finance function. It eliminates the finance function's excuse for not doing its most important work."
The Singapore Read: Local Stakes and Honest Assessment
Singapore's financial services sector is one of the largest contributors to GDP, employing a professional workforce concentrated in operations, compliance, reporting, and analysis — precisely the areas where AI-driven redesign is most active. This is not a distant sector impact. It is a direct challenge to a significant share of Singapore's professional class, including much of the graduate workforce that enters finance each year.
The Monetary Authority of Singapore has not been passive. The FEAT principles — Fairness, Ethics, Accountability, Transparency — have governed AI use in financial services since 2019 and are continuously updated. FEAT makes something easy to underestimate legally mandatory: the governance design of AI deployment must be done before deployment, not retrofitted. A Singapore bank deploying AI-assisted credit decisions cannot treat the human reviewer as a rubber stamp. The review must be genuine — the reviewer must be capable of understanding, challenging, and overriding the AI output.
Aerial cinematic view of Singapore's Marina Bay financial district at dusk, warm amber tower lights reflected on calm water, dark navy sky, long lens compression, no text or signage
Singapore's support infrastructure for this transition is more substantial than most finance leaders realise. SkillsFuture funds AI literacy, data analytics, and finance-specific digital upskilling for every working adult. Workforce Singapore runs Career Conversion Programmes that provide salary support and structured training for professionals moving between finance sub-disciplines — from transaction processing into FP&A, from reporting into business analysis, from compliance operations into AI governance. WSG's Jobs Redesign grant co-funds employers undertaking structured role redesign around AI tools. The tripartite model — government, employers, and unions working in structured coordination through the NTUC and affiliated bodies — gives this support a coordination mechanism most markets lack entirely.
The honest assessment: that infrastructure is only as valuable as the planning that makes use of it. The Microsoft 2026 Work Trend Index identifies a "redesign gap" — the distance between productivity gains AI tools are already delivering and the organisational redesign that would translate those gains into outcomes. That gap is real in Singapore's finance sector. Many functions have adopted AI tools for specific tasks; far fewer have systematically redesigned work around those tools or built the redeployment pipelines to absorb displaced capacity into higher-value activity. The window to close that gap at a deliberate pace is still open, but it is narrower than it was twelve months ago.
For organisations navigating the regulatory and grant dimensions of this redesign, FMC Collective works with Singapore's financial institutions on FEAT alignment and WSG scheme structuring.
The Playbook: Four Moves That Actually Work
1. Map the Processing Layer — Then Set It Free
The first move is diagnostic. Take a representative week of work across your finance function and break it into tasks. For each task, ask one question: does this require genuine judgment that cannot be specified in a rule, or does it have a determinable right answer given the inputs?
The second category — the processing layer — is your automation set. In most finance functions, it represents a disproportionate share of total hours. Reconciliation, first-draft report generation, variance flagging, invoice matching, expense verification: these are tasks where an AI workflow produces an output a human reviews, not produces.
The goal is not to eliminate those tasks from the function. It is to eliminate them from the daily cognitive load of your best people. When a senior FP&A analyst is no longer spending two days of every week-close on variance gathering, those days are available for the analysis and business partnering the function nominally exists to provide but rarely has time to deliver.
2. Redesign Roles Around the Judgment Layer, Not Around the Tools
The temptation when AI tools arrive is to define the human role as "the person who reviews the AI output." That produces a diminished role, not a redesigned one. The reviewer of an AI-generated report is doing a subset of the same work, faster, with less cognitive engagement.
The redesign that creates genuine value asks a different question: now that AI handles the processing, what judgment work was finance always supposed to be doing but never had capacity for? In most organisations, the answer is substantial: genuine business partnering embedded in operating divisions; scenario modelling with real depth; proactive risk identification; capital allocation advisory that influences decisions before they are made.
The finance professional of 2028 must understand AI-generated outputs well enough to interrogate their assumptions — that is a new baseline skill. But the function rewards what AI cannot replicate: contextual intelligence, relationship trust, and the ability to translate quantitative complexity into clear strategic narrative. See how organisations are redesigning roles for the agentic era across every function.
3. Build the Governance Layer Before the Regulator Asks for It
MAS FEAT principles are not bureaucratic box-ticking. They are a substantive requirement that AI-assisted decisions retain genuine human accountability — the reviewer must be capable of reviewing, not nominally present for liability purposes.
The governance layer — how AI outputs are reviewed, what override conditions exist, how errors are monitored, who is accountable when an AI-assisted decision causes harm — must be designed at deployment time, not retrofitted after something goes wrong. For each AI-augmented process, document who reviews the output, what information they must engage with before a decision is made, what the escalation path is if the output appears anomalous, and how the decision is logged for audit. This creates the accountability trail that makes both regulatory inspection and continuous improvement possible.
For implementation support translating FEAT requirements into operational workflow design, Freemansland works with Singapore organisations on AI strategy and implementation.
4. Invest in the Signal Skills — Now, Not Later
The finance professionals most valuable in a redesigned function share three capabilities: the ability to interrogate AI-generated outputs rather than accept them; the ability to translate quantitative analysis into strategic narrative non-finance executives can act on; and the relationship depth to function as a genuine business partner rather than a reporting service.
The first is new. The second and third are old skills the processing layer never had time to develop. SkillsFuture, WSG's Career Conversion Programmes, and e2i's industry partnerships provide funded pathways to build all three — but they require proactive use. The finance leaders who close the redesign gap fastest are already mapping their team's skill profile against this target and investing in the gap now, while the transition still allows deliberate planning rather than forced adaptation.
Close-up of a finance professional's hand annotating a printed graph in a quiet meeting room, warm incandescent light, charcoal and cream tones, shallow depth of field, professional and contemplative atmosphere, no text or logos
The Close: What the Winners Look Like
The finance functions that navigate this transition well will be smaller in headcount than five years ago — not because of indiscriminate cuts, but because the processing layer no longer requires the same number of people. The headcount reduction will be a byproduct of redesign, not its goal.
They will be faster. The week-close cycle will shrink not because standards drop, but because manual work is automated and remaining human effort concentrates on interpretation and decision-support rather than data assembly.
They will be more valuable to the business — providing forward-looking analysis, scenario intelligence, and business-partnering capacity that influences decisions before they are made, not after. That is a more powerful function. It is also a more defensible one, because the value it creates is not the kind AI can straightforwardly replicate.
And they will have done the governance work. In Singapore's regulated environment, under FEAT principles that are only going to become more specific, the finance functions that built genuine accountability structures around their AI workflows will have a compliance advantage that compounds. The audit questions are coming. The functions with answers ready — because they designed governance before they needed to explain it — will be in an entirely different position.
The Insights conversation about AI and finance is still early, but the direction is not ambiguous. The processing layer is going to AI. The judgment layer is going to the humans who have invested in being excellent at it. The organisations that understand this and redesign accordingly will not simply survive the transition. They will emerge from it with a finance function that is, for the first time in a generation, doing what it was always meant to do.
The work is not waiting for better AI. The AI is already here. The work is the redesign.
Explore how the same redesign logic is playing out across professional functions in our Insights section — including the rise of the AI orchestrator reshaping senior finance roles and what Fortune 500 AI workforce moves mean for Singapore businesses.