HomeThe night the algorithm filed the report
A regional bank's compliance team in Singapore used to spend roughly three days a month producing its regulatory breach report. Two analysts worked through the transaction logs, cross-referenced the watchlists, pulled the flagged items, classified each by severity, and wrote up the narrative. It was meticulous, important and — by both analysts' honest assessment — relentlessly tedious. The work required attention. It did not require wisdom.
In mid-2025 the bank deployed an AI monitoring system. The same report now runs overnight. By the time the analysts arrive at their desks, the system has screened every transaction, cross-referenced every required list, classified every flag by severity, drafted the narrative summary, and produced a first-pass filing. The three days became one morning. The volume of transactions the system monitors is fifteen times what the team could cover manually, at a fraction of the cost.
The naive reading is that the bank now needs fewer compliance analysts. The bank reached the opposite conclusion. It needed different ones — and it needed fewer of the wrong kind far less urgently than it needed more of the right kind. The right kind is the Reviewing Human: the professional who no longer screens transactions but who audits the AI's screens, catches the edge case the model misclassified, exercises the judgment the regulator will one day ask about, and owns the accountability that no algorithm can hold.
This is the most consequential redesign happening inside Singapore's legal and compliance functions right now, and most organisations are getting it wrong. They are either clinging to the old job description — pretending the machine is a minor productivity tool and barely changing the role — or they are cutting the function to the bone on the logic that the AI does the work. Both moves are expensive. The redesign that works is the one that neither shrinks nor stagnates: it elevates.
A compliance professional in a Singapore financial district office reviewing AI-generated monitoring output on dual screens, ambient city lights visible through floor-to-ceiling glass, cinematic shallow depth of field in navy and warm amber tones
The structural problem most leaders misread
Before AI arrived, the legal and compliance function had a dirty secret it rarely advertised: the majority of its work was volume, not judgment. Transaction monitoring was a needle-in-a-haystack exercise at human scale on a haystack growing faster than headcount could. Contract review involved reading the same clause in a hundred different agreements and checking it against the same playbook each time. AML screening, KYC document checks, sanctions list lookups — all variations on the same structural task: comparing specific information against a known standard, at high volume, repeatedly.
The compliance officer doing this work was not using judgment on most of it. She was using thoroughness, attention and time. The judgment — the call on a genuinely novel regulatory question, the assessment of the unusual transaction pattern that defied easy classification, the advice to the board on an emerging risk — was a small fraction of her week, buried under the volume work that consumed most of it.
AI is a scalpel aimed at precisely that volume layer. A modern compliance AI system monitors every transaction in real time against every applicable rule, flags anomalies in seconds, screens documents against regulatory requirements, generates first-pass reports, and surfaces the items that actually need human attention. It is not doing the judgment work. It is doing the preparation that used to consume the time that judgment work requires.
This is the structural shift most organisations misread. They see the AI compressing hours and do the arithmetic of substitution: fewer hours means fewer people. That arithmetic is wrong, because it treats the function as a fixed quantum of task-hours when it is actually a capability with a constrained surface area. The constraint was never the headcount. The constraint was the human attention available for judgment — and the machine just removed the thing absorbing most of that attention.
The Reviewing Human is not a diminished version of the old compliance officer. She is the liberated version — doing, finally, the part of the job that was always the point.
Why Singapore's stakes are specific
Global trends matter here, but the compliance redesign plays out differently inside Singapore's particular texture.
The Monetary Authority of Singapore's FEAT principles — Fairness, Ethics, Accountability, Transparency — represent the most mature public articulation of responsible AI use in a regulated financial context in this region. They are not decoration. They encode a view that is specific and demanding: every consequential AI decision must be explainable, auditable, and owned by a human who can be held accountable. That is a design specification, not a compliance checkbox. It tells a bank, an insurer, or a financial advisory firm exactly what the Reviewing Human has to do and why she cannot be replaced by the same system she is reviewing.
The stakes run beyond finance. Singapore's legal sector, its corporate secretariat functions, its listed-company governance — all carry accountability structures the AI cannot absorb. The company secretary who used to draft board minutes by hand can now review an AI-drafted set in minutes, but she still signs off, and the signature is still her professional accountability. Speed of production does not transfer accountability. That is the principle most organisations get wrong, and the one Singapore's regulators are least likely to forgive.
The national context sharpens the case. The World Economic Forum's Future of Jobs projections point to approximately 170 million new roles and 92 million displaced globally by 2030 — a net gain of roughly 78 million. The reason the number is positive is that roles are being redesigned faster than they are being deleted by organisations that understand what AI actually does. Microsoft's 2026 Work Trend Index named the gap precisely: a "redesign gap" where productivity gains from AI are outpacing the organisational redesign needed to capture them. Legal and compliance functions sit squarely in that gap.
The Singapore-specific advantage is the institutional infrastructure built to close it. Workforce Singapore, e2i and SkillsFuture run Career Conversion Programmes and Jobs Redesign support — not to manage redundancy, but to fund the reskilling that turns a transaction-monitor into a Reviewing Human. Singapore's tripartite model, which has navigated every major economic disruption without the social fractures other economies absorbed, is purpose-built for this moment. A compliance function that frames its AI transition as reskilling rather than restructuring does not just look better — it moves with the whole system, and the whole system moves faster.
DBS, OCBC and UOB have all been transparent about augmenting rather than replacing their workforces with AI. The signal from the institutions Singapore's financial sector trusts most is consistent: the answer is not fewer compliance professionals. It is compliance professionals who work differently. For an organisation navigating both the technology deployment and the governance questions — grant pathways, regulatory posture, workforce redesign — that is precisely the brief that FMC Collective exists to support.
A tripartite scene combining Singapore's financial district skyline, a modern compliance team working alongside AI dashboards, and a regulatory document review, rendered in deep navy with warm gold accents and shallow focus
What the Reviewing Human actually does
The title matters because it encodes the job change. The old compliance analyst found the problems. The Reviewing Human validates that the AI found the right problems, questions what it missed, and owns the call on what happens next.
The Reviewing Human exercises calibrated scepticism. She knows the AI is confident — these systems are almost always confident — and she knows confidence is not accuracy. She knows the edge cases: the transaction that looks clean on every individual rule but smells wrong when you hold the pattern up to the light. The contract clause that passes the playbook check but conflicts with a regulatory guidance issued three weeks ago. She is not auditing the document; she is auditing the AI's reading of the document, which is a harder and more valuable task.
The Reviewing Human owns the accountability the AI cannot hold. There is always a point in the chain where a human must sign. MAS expects it. The Law Society expects it. A court will expect it. AI can prepare, draft, flag and summarise, but it cannot accept professional liability, it cannot be questioned by a regulator, and it cannot exercise the judgment that distinguishes a technical breach from a material one. Every consequential compliance decision needs a human in the chain who chose it, understood it, and can defend it.
The Reviewing Human feeds the system. An AI compliance tool gets better through the corrections and edge-case judgments that experienced professionals feed back into it — the "this flag was a false positive because of this specific context" knowledge that makes the model genuinely useful for this organisation's work in this regulatory environment. When professionals believe the AI exists to replace them, they stop feeding it. They route around it and wait for it to fail. The replacement framing actively sabotages the flywheel. A redesign framing produces the opposite dynamic: the professionals become the AI's best trainers.
The Reviewing Human is not the person the AI assists. She is the person who decides whether the AI was right — and who answers when it was wrong.
This is also the pattern that the Fortune 500's AI workforce moves have been quietly pointing toward. The firms that found durable operating leverage from AI are not the ones that deployed the most tools; they are the ones that redesigned the human role most deliberately around the tool.
The playbook: four moves to run now
1. Map tasks before you touch the org chart
Spend two weeks mapping what your team actually does. Pull a representative month of work and tag every task: routine versus judgment-heavy, volume-driven versus novel. You will almost certainly find that more than half the time is consumed by genuinely routine work: monitoring runs, document checks, report drafts, standard-item classification. That is your automation surface. Start from the task map, not the org chart. Start from the org chart and you will cut the wrong things in the wrong order.
2. Deploy AI against the routine — loudly to your team, invisibly to regulators
Route the volume work to the machine without apology. But communicate this to your team before the tool arrives: this clears the grind so you can do the work that actually requires you. The adoption dynamic you need is professionals who engage with the AI, feed it corrections, and own its output. Regulators should experience only the output: faster turnaround, greater coverage, better-drafted reports. The compliance posture improves; the machinery behind it is invisible.
3. Redesign the role upward — and pay accordingly
Once the routine layer migrates, rewrite the job description around the Reviewing Human's actual work: validation, edge-case judgment, regulatory interpretation, AI oversight, stakeholder accountability. This is a harder, more senior job. If you automate 60 per cent of a compliance officer's tasks and leave the salary and title unchanged, you have manufactured resentment. Redesign the role genuinely and you have built your most productive and most defensible compliance operation. Freemansland helps organisations navigate exactly this redesign in practice — from the AI implementation to the workforce reconfiguration that captures the leverage.
4. Keep the human in the loop at the accountability layer
Every consequential compliance decision — any escalation, any matter that could carry regulatory consequence, any report that goes to a board or regulator — must have a named human who reviewed it, understood it, and owns it. This is the design constraint that separates defensible AI from delegated liability. Build the workflow so the AI prepares and the human decides, with a clear audit trail of who owned the call. MAS's FEAT principles are pointing at exactly this model — and organisations that build it now will not be caught scrambling when the regulation is explicit.
A close-up of a compliance professional's hands annotating an AI-generated regulatory report, a Singapore financial district visible softly out of focus through a window, deep charcoal and warm accent tones, premium editorial quality
The investor's close: accountability as a moat
For anyone allocating capital to, or governing the risk function of, a regulated business in Singapore, the argument cashes out on the income statement.
Legal and compliance functions have historically been pure cost centres — necessary but not obviously generative. AI changes the economics without changing the accountability structure, and that asymmetry is the opportunity. When AI handles the volume, the cost of compliance coverage drops — dramatically. The same team can monitor fifteen times the transaction volume, review contracts across a far larger portfolio, and produce reports that are more thorough and more timely than any manual process. That is not a headcount saving; it is a step-change in coverage quality at lower marginal cost.
But the risk calculus only holds if the Reviewing Human is genuinely there. An organisation that deploys AI for compliance coverage and then thins the judgment layer has not reduced its risk — it has transferred it. The AI screens fifteen times the volume, and the attenuated human layer misses the edge case the model was not trained for. The fine, when it arrives, is not smaller because the AI screened everything; it may be larger, because the regulator will ask how the oversight layer was structured.
The question is not "does this firm use AI in its compliance function?" Every serious regulated firm soon will. The question is "does this firm have a Reviewing Human in the chain — and is that person genuinely empowered to catch what the machine misses?"
This is where the broader redesign of AI-era organisations cashes out in a regulated context. AI replaces tasks, not roles. The accountability that cannot be automated must be held by someone who is genuinely accountable. In legal and compliance, the regulation makes this explicit in ways that other functions can sidestep. That explicitness is not a burden. It is a forcing function for good design — and the organisations that treat it that way will end up with compliance functions that are cheaper, more thorough, more defensible, and more trusted than the ones they are replacing.
The Reviewing Human is not the consolation prize for functions the AI did not fully automate. She is the point of the whole redesign — the professional whose judgment, finally freed from the grind of routine volume work, is what makes the AI safe to use at all. The machine earns its keep by clearing the grind. She earns hers by catching what the machine misses, owning what the machine cannot, and signing off on the outcome that matters.
For more on how the AI workforce shift is being decoded across Singapore's functions and sectors, explore our full Insights series.