HomeIn May 2023, on an earnings call that lasted barely an hour, IBM's chief executive said something that travelled further than almost anything the company shipped that year. Arvind Krishna told Bloomberg that IBM expected to pause or slow hiring for around 7,800 back-office roles — a chunk of them in human resources — because he believed artificial intelligence could handle that work over roughly the next five years.
No product launch. No keynote. Just a number — 7,800 — attached to the most quietly radical idea in modern management: that a category of white-collar jobs at one of the world's most established technology companies might simply stop being hired for, because software could now do enough of the task.
The headline wrote itself, and it was wrong in the way most viral headlines are wrong. "IBM to replace 7,800 jobs with AI" raced across feeds and group chats and slide decks. HR professionals read it on their commute and felt the floor tilt. Investors read it and reached for a calculator. Founders read it and wondered, quietly, whether they could do the same.
But the real story is not the number. The real story is what IBM did around the number — the redesign, the internal AI it had already built, the roles it kept hiring for even as it slowed others. A hiring pause is not a massacre. It is a decision about where human judgment is worth paying for — and where it isn't.
That distinction is the whole game. Get it right and you build a leaner, more valuable company without burning the trust and knowledge that took years to accumulate. Get it wrong — confuse "the task can be automated" with "the person can be removed" — and you discover, often too late, that you cut the one thing the dashboard never measured.
For Singapore companies sitting on large back offices, shared-service centres and HR functions, IBM's move is not a curiosity from across the world. It is a preview — and the kind of signal this Insights series exists to decode for the local market. The question is no longer whether the routine tier of your back office can be automated. It can. The question is whether you redesign before you reduce — or after it's too late to matter.
What IBM actually did
Let's be precise, because precision is exactly what the headline destroyed.
In that May 2023 conversation, Krishna described a forward-looking decision, not a completed purge. IBM had a large global workforce, and within it a significant population of non-customer-facing, back-office roles — the administrative engine room of a big company. HR sat squarely in the frame. His estimate, reported as around 7,800 roles, was the slice he expected AI and automation to absorb over roughly five years. The mechanism he described was, in plain terms, hiring slower for those functions and letting natural attrition do the rest — people retire, people move on, and the company simply doesn't backfill every seat because the work behind it has changed.
That is a profoundly different action from announcing layoffs. A hiring pause through attrition is a bet on a trajectory; a layoff is a cut to the present. Treating them as the same thing is the first and most common error in this entire conversation.
The machine was already running
What made IBM's statement credible — and what most coverage skipped — is that IBM wasn't speculating about a future tool. It was describing the logical extension of something it had already built and deployed internally: AskHR.
AskHR is IBM's own AI-powered HR assistant. When an employee needs to check leave, ask a payroll question, request a standard letter, understand a benefit, or navigate a policy, they ask AskHR instead of opening a ticket to a human in a shared-service centre. IBM has publicly said the system now resolves a large share of routine HR queries automatically — the high-volume, repetitive, rules-based requests that used to consume the bulk of an HR helpdesk's day.
Sit with what that means operationally. The classic HR back office runs on volume: thousands of near-identical questions, most of which have a single correct answer buried in a policy document. That is precisely the shape of work a language model handles well — bounded, repetitive, document-grounded. AskHR didn't make HR people smarter; it ate the part of their day that never needed a person in the first place.
So when Krishna pointed at 7,800 roles, he wasn't gesturing at a hope. He was extrapolating from a system already absorbing the routine tier in production.
The part the headline buried
Here is the nuance that turns a scary story into a useful one. IBM did not describe a company shrinking. It described a company reallocating.
In the same breath as the back-office pause, IBM made clear it kept hiring — aggressively — in client-facing, technical, software and AI roles. The headcount wasn't simply leaving the building; a meaningful share of it was being redirected toward work that creates revenue and competitive advantage. The framing IBM has consistently used is that automating the routine freed staff for higher-value work rather than freeing them out the door.
IBM didn't announce that AI replaces people. It announced that AI replaces tasks — and then used the savings to hire more people where judgment actually pays.
This is the detail that should reframe the whole debate. The company most cited as "cutting 7,800 jobs to AI" was, at the same time, expanding the parts of its workforce that AI cannot do. The story isn't subtraction. It's a portfolio shift — out of low-judgment administration, into high-judgment, client-facing and technical work.
And — to stay honest about the limits of what we know — every figure here should be read as reported and approximate. The 7,800 is an estimate over a five-year horizon, stated on a call, not a line item in an audited filing. The "large share" of HR queries AskHR resolves is IBM's own characterisation. The direction of travel is clear and credible; the decimal points are not. Anyone who quotes these as hard, precise facts is doing the same thing the original headline did — mistaking a vivid number for a verified one.
Why IBM, and why HR, first
It's worth asking why this signal came from IBM specifically, and why HR was the named target. The answer tells you something useful about which companies and which functions are exposed first.
IBM is not a startup chasing a narrative; it is a century-old enterprise with a deep, mature back office and a long institutional memory of automation. When a company like that says a category of administrative work is becoming automatable, it carries a different weight than a founder's tweet. The signal is more credible precisely because the source is conservative. Established firms with large shared-service operations are, counter-intuitively, among the most exposed — not because their people are less capable, but because decades of scale produced exactly the kind of standardised, high-volume, rules-based work that AI absorbs most easily.
And HR was the lead example for a reason that generalises everywhere. HR is the function where the gap between routine volume and human value is widest. A vast share of HR's daily workload is repetitive administration — leave, payroll, policy, letters, records — while the part that genuinely matters, the part you'd never automate, is a smaller slice of judgment-heavy, relationship-heavy, legally-sensitive work. That bimodal shape is what makes HR the cleanest illustration of the whole thesis: a function where you can automate enormous volume while protecting the much smaller, much more important core. If you want to understand where AI hits your own organisation first, look for the function with that exact profile.
A precise editorial still life of a corporate back-office function being reorganised, papers and workflow nodes flowing from a dense cluster into ordered streams
The misread
Almost everyone who reacted to IBM's announcement made the same mistake, and it is worth naming because it is the mistake that destroys value in real companies every quarter.
The misread is this: "AI can do the job, so the person goes."
It feels logical. It is also wrong, and it's wrong at the level of definition. A job is not a task. A job is a bundle of tasks, glued together by judgment, context, relationships and accountability. When you look closely at almost any HR role, you find a bimodal distribution of work inside it. Some of it is high-volume and rules-based — answer the leave question, generate the letter, update the record. And some of it is irreducibly human — sit with an employee in distress, navigate a harassment complaint, redesign a team after a reorganisation, coach a first-time manager through firing someone for the first time.
AI is extraordinary at the first kind. It is close to useless — and frequently dangerous — at the second.
So when a headline says "AI can do an HR person's job," it is quietly committing a category error. AI can do part of the bundle, often a large part by volume. But volume is not value. The 70% of tasks that are routine might be only 30% of the value the role actually creates — and the 30% that's hard is where the trust, the retention, the legal exposure and the culture actually live.
Why the misread is expensive
The danger isn't just intellectual sloppiness. It's that the misread leads straight to a specific, costly decision: cut the headcount, keep the volume, and assume the AI will absorb the gap.
What happens next is predictable. The routine tier flows to the machine and works fine. Then the hard cases arrive — and there's no longer anyone with the experience, the context or the authority to handle them well. The complaint escalates. The good employee, badly handled, leaves. The institutional knowledge that lived in the person you let go walks out with them and never comes back. None of this shows up in the efficiency dashboard that justified the cut. It shows up later, in attrition, in legal cost, in the slow erosion of trust.
This is exactly the trap a company like Klarna walked into and had to reverse — automating support so aggressively that it later rehired humans to protect quality. We unpacked that whole episode in Klarna Replaced 700 Agents With AI — Could a Singapore Bank?, and the lesson rhymes perfectly with IBM's: the routine tier automates beautifully; the last, hardest tier is where the value you can't see actually lives.
IBM's quiet genius was that it didn't make this mistake. It automated the tasks, redirected the people, and kept hiring where judgment pays. The companies that copy the headline instead of the strategy are the ones that get hurt.
The second misread: that the timeline is instant
There's a quieter version of the error worth naming, because it cuts the other way. Some leaders read "AI can do this" and panic into cutting now. Others read the same words and dismiss the whole thing as years-away hype they can ignore. Both are wrong, and IBM's own framing is the corrective.
Krishna's estimate was explicitly a five-year horizon, achieved largely through attrition — not a switch flipped overnight. That tells you the realistic shape of this transition: it is gradual, cumulative and survivable if you start now, and brutal only for the companies that ignore it until a competitor's operating leverage forces their hand. The mistake isn't moving too slowly or too fast in the abstract. The mistake is not mapping the work at all — and then being forced into a reactive, headline-driven cut when the pressure finally arrives. The companies that map early get to redesign on their own timeline. The ones that wait get to react on someone else's.
Redesign, not replacement
If "replace the person" is the wrong instinct, what's the right one? The answer is a discipline, not a vibe. Before you touch a single role, you decompose the work into three buckets — and you treat each one differently.
Bucket one: machines are simply better
Some work is high-volume, rules-based, available-records-driven and emotionally neutral. Machines are not just cheaper here; they are genuinely better — faster, more consistent, available at 3am, fluent in English, Mandarin, Malay and Tamil at once, and immune to the fatigue that makes a tired human give a wrong answer at the end of a long shift.
In HR, this bucket is large and obvious: answering "how much annual leave do I have left," generating a standard employment letter, processing a routine leave application, surfacing the right clause of the policy handbook, handling tier-one onboarding paperwork, doing a first-pass scan of inbound resumes against hard criteria. This is the AskHR bucket. Automate it without guilt. Leaving humans to do it isn't kind — it's a waste of a person.
Bucket two: humans are simply better
Some work degrades the moment you hand it to a machine, because its entire value is the human in it. The employee who is being made redundant and needs to be told with dignity. The grievance that is half fact and half feeling. The high-performer quietly deciding whether to leave. The reorganisation that will reshape three teams and a dozen careers.
These are not "complex tickets." They are acts of judgment, trust and accountability — and the cost of getting them wrong is measured in lawsuits, resignations and reputational damage, not in handle time. AI can assist here (draft a first version, summarise a case file, surface relevant precedent) but it must never own the outcome. The human isn't a fallback for when the AI fails. The human is the point.
Bucket three: better together
This is the bucket most companies miss entirely, and it's where the real upside lives. A lot of work isn't best done by the machine or the human alone — it's best done by a human amplified by a machine.
Picture the redesigned HR business partner. The AI drafts the policy summary, pulls the case history, flags the legal risk, prepares the data pack and writes the first version of the difficult email. The human reads the room, makes the call, owns the relationship and carries the accountability. The output is faster and better than either could manage alone — and crucially, the human is now spending their day on the 30% of work that actually creates value, because the machine cleared the 70% that didn't.
The winners don't ask "which jobs can AI do?" They ask "which tasks, in which bucket, and what does the human role become once we move them?" That single reframe is the difference between a company that gets leaner and one that just gets emptier.
This is why redesign before you reduce is not a slogan — it's a sequence. You map the buckets first. You rebuild the human role around buckets two and three. Then, and only then, do you know how much headcount the work actually requires. Reverse the order — cut first, hope the AI catches the rest — and you're not redesigning. You're gambling with your institutional memory.
The hard part, honestly, isn't the AI. It's the change. Decomposing roles, rewriting job descriptions, retraining people into higher-value work and getting them to trust the new tools is a people-and-process problem long before it's a technology one — which is precisely the process and change-management work our sister company Freemansland Creatives exists to do. The model is clean. The execution is where companies fall down.
A conceptual triptych showing three modes of work — automated routine, human judgment, and human-and-machine collaboration — as three distinct flowing panels
Could a Singapore corporate back-office or HR player do the same?
Now bring it home. Could a large Singapore employer — a bank, an insurer, a telco, a government-linked company, a regional shared-service centre headquartered here — run the IBM playbook?
Technically, yes. Substantially. And several already are. Singapore is one of the most digitally mature economies on earth, with a dense concentration of regional HQs, shared-service operations and back-office hubs that exist precisely because they process high volumes of routine, rules-based work. That work — payroll runs, leave administration, HR helpdesk queries, onboarding paperwork, compliance documentation — is the textbook AskHR bucket. The automation is not the hard part.
But "could they" and "should they, the same way" are different questions, and Singapore's specific context changes the answer in three important ways.
One: the back office here is bigger than people think
A meaningful share of Singapore's white-collar employment sits in exactly the functions IBM targeted: finance shared services, HR shared services, procurement operations, regional administration for multinationals. These are good jobs held by capable people — and a large fraction of the day-to-day work inside them is automatable. That is not a comfortable sentence, but it is a true one, and pretending otherwise serves no one. The exposure is real.
What's also true is that Singapore has, uniquely, built national machinery to handle exactly this transition — which we'll come to. The combination matters: high exposure, but also high institutional support for redesign. That's a very different starting position from most markets.
Two: trust is the product, and the regulator is watching
In Singapore's flagship sectors — banking, insurance, healthcare, public service — trust is not a feature; it is the entire value proposition. A customer hands a bank their life savings or an insurer their family's security on the strength of believing a competent, accountable human stands behind the system. Automate the visible, emotional, high-stakes moments badly and you don't save cost — you spend down the one asset that took decades to build.
This is why the redesign logic bites harder here, not softer. In a trust-sensitive, regulated market, the cost of getting bucket two wrong — the complex case, the distressed customer, the disputed claim — is amplified. The Singapore version of the IBM playbook therefore leans even more heavily on keeping humans firmly in control of judgment-heavy work, with AI as the amplifier underneath.
Three: the redesign is already underway in the open
This isn't hypothetical. Singapore's banks have been among the most public globally about reshaping roles around AI — a story we examine in detail in DBS and the 4,000 Roles: What AI Really Means for Singapore's Banks. And the pattern that matters there is the same one IBM modelled and Amazon is now running at scale across its org: the headline lands on the roles being reduced, but the substance is in the roles being created and redesigned alongside them — the AI specialists, the data and platform people, the reshaped front-line jobs that now run on better tools.
The shared-service centre question
There's a specific Singapore wrinkle worth naming. The country hosts a large number of regional shared-service and global-business-services centres — back offices that multinationals consolidated here precisely to process routine finance, HR and procurement work for the whole region at scale. These centres were built on the economics of human throughput: many capable people processing high volumes efficiently.
AI changes that economic logic at its foundation. The very work these centres were built to do is the work most exposed to automation. That's not a reason for alarm; it's a reason for early, deliberate redesign. The centres that move first — automating the routine tier and moving their people into the higher-value oversight, exception-handling, analytics and client-relationship work that AI can't do — will not just survive, they'll become more valuable regional assets than the throughput model ever allowed. The ones that wait risk being out-competed by a leaner centre next door, or quietly consolidated away. For Singapore's GBS sector specifically, redesign isn't optional — it's the difference between leading the region's next chapter and being written out of it.
So the honest verdict for a Singapore player is this: yes, you can run the IBM playbook — but the prize goes to the version that redesigns, reskills and keeps trust intact, not the version that just cuts. The technology is a commodity you can buy. The redesign — and the change-management muscle to make it stick — is the part competitors can't copy off your invoice. Building the AI capability itself is what our sister firm Freemansland exists to do; building the organisation around it is the harder, more durable win.
The Singapore context
Here is where Singapore stops being just another market and becomes, arguably, the best place in the world to do this well. Because the national system is explicitly designed to push companies toward redesign and reskilling — not away from people.
Tripartism: the quiet superpower
Singapore runs on tripartism — the structured collaboration between government, employers and unions (through the National Trades Union Congress). It is unglamorous and easy to overlook, and it is one of the country's genuine competitive advantages in the AI transition. In many economies, automation becomes a zero-sum fight between management and labour. In Singapore, the institutional default is to sit all three parties at one table and ask: how do we transform this work without discarding the people doing it?
That changes corporate behaviour at the root. A Singapore employer contemplating a back-office automation programme isn't operating in a vacuum where the only lever is headcount. It's operating inside a system that actively expects — and funds — redesign as the first move.
The redesign machinery: WSG, e2i, SkillsFuture
Singapore has built, over years, a stack of national programmes whose entire purpose is to help companies do exactly what IBM did — but humanely.
Workforce Singapore (WSG) runs job-redesign support and Career Conversion Programmes (CCPs) — schemes that help employers reshape roles around technology and reskill existing staff into the higher-value work that remains, rather than releasing them. The logic is a precise institutional mirror of the three-bucket model: automate the routine, move the human up the value chain, and have the state co-fund the transition.
e2i (the Employment and Employability Institute), NTUC's arm, sits on the worker side of the same effort — helping individuals reskill and move into new roles, and helping companies that are transforming do so with their people rather than around them.
SkillsFuture underwrites the lifelong-learning layer beneath all of it — the national assumption that a worker whose tasks are automated should be retrained, not written off.
Put together, the message to any Singapore company is unusually clear: the country will help you fund the redesign. It will not help you fund the lazy cut. The incentives are deliberately tilted toward keeping and upgrading your people.
The regulatory expectation
In financial services specifically, there is a further layer. The Monetary Authority of Singapore (MAS) has been consistent that as institutions adopt AI, they remain fully accountable for fairness, ethics, accountability and transparency in how those systems are used — including in decisions that affect people. The practical translation for HR and back-office automation is direct: you cannot outsource accountability to a model. A human must remain answerable for consequential decisions. That expectation, by design, pushes regulated firms toward the "better together" bucket — AI assists, humans decide and own the outcome — rather than toward removing the human from the loop.
The net effect of all this is a national environment that is almost custom-built to reward the IBM strategy done right and to penalise the headline version done wrong. Singapore doesn't just permit redesign-before-reduce. It subsidises it. A company here that still chooses to cut first and ask questions later isn't just taking a people risk — it's leaving national funding, and a competitive advantage, on the table.
A calm, authoritative editorial scene evoking institutional collaboration — three distinct streams of activity converging toward a single shared structure
The operator's playbook
Strategy is only worth the actions it produces. So here is the concrete sequence a Singapore business should run — not someday, but this quarter. Five moves, in order.
1. Audit the work, not the org chart
Before you touch a single role, map the tasks. Take your back-office and HR functions and decompose them into discrete tasks, then sort every task into one of the three buckets: machine-better, human-better, better-together. Be ruthless and be honest. You will find that the routine tier is bigger than your managers admit and the judgment tier is more valuable than your dashboard shows. You cannot redesign a role you have never actually mapped — and almost no company has. This audit, done properly, is the single highest-leverage thing on this list.
2. Automate the routine tier first — and prove it internally
Start where the value is clearest and the risk is lowest: the bucket-one work. Stand up an internal AI assistant for the highest-volume, most repetitive queries — the Singapore equivalent of AskHR. Crucially, deploy it internally before you ever point it at a customer. Your own employees are the safest, most forgiving testbed. You learn the failure modes, build the guardrails and earn organisational trust on low-stakes ground — then you extend it outward. IBM didn't start by automating client trust; it started with its own HR helpdesk.
3. Redesign the human roles deliberately — don't let them drift
Once the machine absorbs the routine tier, the human roles must be actively rebuilt — not left to quietly hollow out. Rewrite the job descriptions around bucket two and bucket three: judgment, relationships, complex cases, and human-plus-machine collaboration. Decide explicitly what the HR business partner, the payroll lead, the onboarding specialist becomes. A role that loses 60% of its tasks and gains no new definition doesn't get more valuable — it gets more fragile. This is the step companies skip, and it's the one that decides whether redesign creates value or just anxiety.
4. Reskill with the system, not against it
This is where Singapore hands you an advantage most markets don't have: use it. Engage Workforce Singapore's job-redesign and Career Conversion Programmes, work with e2i, tap SkillsFuture to fund the retraining that moves your affected staff into the higher-value roles you just designed. Bring your people along the curve deliberately. The reskilling is not a cost centre to minimise — it's co-funded, and it's the mechanism that converts a threatening transition into a loyalty-building one. Companies that treat their staff as partners in the redesign keep the institutional knowledge that the cut-first crowd throws away.
5. Keep humans on the hook for consequential decisions
Finally, set a hard line and never cross it: AI assists, humans decide and own anything that materially affects a person or carries legal, ethical or reputational weight. Hiring and firing, grievances, performance management, claims, disputes — a competent, accountable human stays answerable for the outcome. This isn't only an MAS-aligned posture for regulated firms; it's the discipline that protects you from the quiet, dashboard-invisible damage that sinks the cut-first approach. Make the human-in-the-loop a design principle, not an afterthought.
Run these five in order and you get the IBM outcome without the IBM headline: a leaner, more productive organisation that kept its trust, its knowledge and its people's loyalty — and got the country to help pay for the transition.
The number that should move
Let's close where the investors are looking — the income statement. Because underneath all of this sits a financial story that's bigger than any HR programme, and it's the reason this matters to anyone holding equity.
The metric that should move is operating leverage — and its most legible expression, revenue per employee.
For a service business, the entire IBM-style redesign is, in financial terms, an attack on the relationship between headcount and output. The routine tier — the bucket-one work — is essentially a fixed cost that scaled with volume. Every additional thousand HR queries used to require additional people. Automate that tier and you break the link between growth and headcount for a meaningful slice of the operation. Volume can rise without the cost base rising with it. That is operating leverage in its purest form, and it flows straight to the margin line.
Revenue per employee is the simplest way to watch it happen. If you can hold or grow output while the denominator stays flat — or shifts toward higher-value, revenue-generating roles — every dollar of new revenue carries more margin than the last. That is precisely the portfolio shift IBM described: out of low-leverage administration, into client-facing and technical work that creates revenue. The company didn't just get cheaper. It changed the shape of its economics.
But here is the honest, calibrated caveat — and it's the one that separates a real analyst from a hype merchant. The leverage only materialises if the redesign is done well. Cut headcount without redesigning, and you don't get clean operating leverage — you get a temporary cost saving followed by a hidden liability: churned customers, legal exposure, lost knowledge, a brittle organisation. That cost lands on the income statement too. It just arrives later, under a different name, and it's harder to reverse than it was to cause.
The story investors should want isn't "we cut 7,800 roles." It's "we broke the link between our growth and our headcount — without breaking the trust that makes our revenue durable." One of those is a number. The other is a moat.
So when a Singapore company announces an AI workforce programme, the question a sharp investor asks isn't "how many roles did you cut?" It's "show me the redesign — which tasks moved, what the human roles became, how you held trust, and where the freed capacity was redeployed." The first question measures a one-time saving. The second measures whether you built something durable. Assume nothing about the future from a headline number; the durability is in the design, and the design is what compounds.
IBM gave the market a number. The companies that win will give the market a redesign. AI doesn't replace people — it replaces tasks. The winners redesign the work; they don't just cut headcount. Redesign before you reduce — and in Singapore, the country will help you pay for it.