A content writer in Manila spent months cleaning up copy that an outside PR firm had produced with AI, and teaching that system her employer’s house style. Eight months in, with permanency a month away, she lost the job. “I feel like I dug my own grave,” she told the BBC. “We were the ones who trained the artificial intelligence that replaced us.” She asked not to be named, because she signed a confidentiality agreement in exchange for severance and works in an industry small enough that speaking up would follow her. Read that as a story about the Philippines and it is affecting but distant. Read it as a story about who was buying the output, and it lands rather closer to home.
The pressure has a return address
The BBC’s account, reported from Manila, is built on interviews with several former outsourcing employees who all requested anonymity on the same terms. Around them sits an industry the Philippines spent two decades constructing on purpose. From the early 2000s, successive governments sold the country to business process outsourcing buyers as the English-speaking option beyond India, offered tax incentives, and built the infrastructure to support it. Accenture, Concentrix and Teleperformance put large campuses into Manila. The result now carries around 1.9 million jobs, turns over $40bn (£30.1bn) a year and accounts for about a tenth of the national economy.
The interesting part of the BBC’s reporting is not that AI is arriving in that industry. It is who is making it arrive. Paul Quintos of the University of the Philippines-Diliman told the BBC that managers within the sector are uneasy and would rather slow adoption down, concerned by the scale of displacement it could produce at home. What overrides them is the client. “But there’s tremendous pressure from foreign clients to adopt AI,” he said. “It’s a major cost-cutting measure.”
Those foreign clients include a great many British companies. UK financial services, telecoms, retail and utilities have run customer service, back-office processing, finance functions and marketing production out of Manila for years. If your organisation holds a BPO contract, the pressure Quintos describes is partly yours. It arrives as a procurement expectation — cheaper unit rates each renewal, plus AI now written into the service description — and it is answered several thousand miles away by decisions about who stays employed.
Strategic Reality: The automation decision in an offshore contract does not sit with the supplier alone. It is co-authored by the buyer’s price expectations, and the buyer typically never sees the workforce consequence because it lands on someone else’s payroll in someone else’s jurisdiction.
The Philippine exposure picture
| Measure | Figure | Source |
|---|---|---|
| Outsourcing employment | ~1.9 million jobs | Industry figures reported by the BBC |
| Sector revenue | $40bn (£30.1bn) a year, ~10% of GDP | Industry figures reported by the BBC |
| Workers in GenAI-exposed occupations | 12.7 million — more than a quarter of employment | ILO research brief |
| Regional ranking | Highest exposure rate among ASEAN countries with comparable data | ILO |
| Jobs in the highest exposure category | 3.6% | ILO |
| Metro Manila (National Capital Region) exposure | Roughly 40% of jobs | ILO |
| IBPAP members running AI pilots | More than two-thirds | Jack Madrid, IBPAP president, to the BBC |
| Government reskilling commitment | 300,000+ sector employees | Philippine DICT |
Exposure is not the same number as displacement
The 12.7 million figure travels well and is usually where coverage stops. The ILO’s own brief is more careful, and the care matters for anyone modelling their own workforce.
More than a quarter of Philippine employment falls into occupations carrying some generative AI exposure — the highest rate across ASEAN countries with recent comparable data. But the ILO puts only 3.6% of jobs in the top exposure band, the one it associates with elevated displacement risk. Its expectation for the rest is task automation inside occupations rather than the occupation disappearing — jobs changing shape, not vanishing. Exposure is also uneven in ways an average hides: women face roughly double the rate men do, reflecting where they are concentrated, and Metro Manila, home to the outsourcing campuses, runs at roughly 40%.
Both halves of that are load-bearing. A roughly sevenfold gap between “exposed” and “highest risk of displacement” should discipline the apocalyptic reading. A quarter of national employment concentrated by geography, gender and sector should equally discipline the complacent one. The macro number will keep looking survivable while specific districts and specific job families absorb the whole of the adjustment — which is the same reason UK payroll aggregates keep failing to show an AI jobs crisis that individual firms and towns can feel perfectly well.
Critical Context: Jack Madrid, president of IBPAP, told the BBC that over two-thirds of the association’s membership has AI pilots under way, and that “there’s a layer of jobs that are automatable or have been automated”. He also maintains that nearly everyone affected so far has been moved into another role at the same employer. Both claims can hold. Redeployment is what the exposure-not-displacement finding predicts, and it is also what the workers who were not redeployed cannot tell you about.
”AI washing” is a governance problem on the buyer’s side too
Quintos gave the BBC the sharpest concept in the piece. He notes there is a term, “AI washing”, for presenting layoffs as technology-driven when the company is in fact responding to weak demand or a slowing economy. Pinning job cuts on the technology, he argues, makes a restructuring look like progress rather than a reaction to market conditions.
That is normally read as a critique of employers dressing up cuts. It has a second edge that UK boards should feel more directly. Madrid himself notes that clients are taking longer to decide whether to offshore a function at all, which he attributes to weaker global demand and genuine uncertainty about how firms will end up deploying the technology. Meanwhile, despite the hundreds of billions being spent globally, Quintos observes that evidence of what workplace AI tools actually deliver in productivity remains thin — a point that will be familiar to anyone who has followed the productivity paradox in enterprise deployment.
Put those together and you get an uncomfortable question about your own supplier’s numbers. When a BPO partner comes to renewal with a lower headcount and an AI-attributed saving, you are being asked to accept a causal claim you cannot audit. If the reduction was actually demand-driven, you are paying for a narrative. If it was genuinely automation-driven, you have inherited a delivery model whose quality assurance now depends on a system nobody has shown you. Either way, “AI efficiencies” on a supplier’s invoice is an assertion, not a measurement — and the same failure of scrutiny that lets it through in procurement is the one that lets it through in your own internal business cases.
Reality Check: Ask your outsourcing partner one question at the next review: which specific processes are now model-assisted, and what is the measured error rate and human review step for each? A supplier genuinely running AI in production can answer it. A supplier repricing a demand-driven cut as innovation cannot.
The pattern in Lisa’s account is not a Philippine pattern
Strip away the geography and the mechanism is completely portable. A team is given AI output to edit. The editing improves the output. The improved output reduces the need for the team. Another former writer, whose name the BBC also changed, described the intermediate stage precisely: her employer promoted the technology as a way to raise output, and instead of easing the work it added to it. “We had to edit more, fact-check more because the data AI produced was inaccurate,” she said. “Technically it was more work for us.” She lost her job too.
UK content, customer support, paralegal and analyst teams are doing this exact work right now, under exactly this framing, and the intensification half of it is well documented — AI tools reliably expand workload before they reduce it. What the Philippines has that Britain does not is visibility. The employment is concentrated in identifiable buildings in identifiable districts, so the transition produces a story. In the UK the same displacement is diffuse — a few roles per employer, a graduate intake quietly not renewed, the entry-level rungs of a career ladder removed without an announcement. Diffuse harm does not generate a BBC investigation. It generates a slightly worse hiring statistic three years later.
There is a structural asymmetry worth naming too. Labour groups told the BBC that Philippine employment law covers redundancy but offers little guidance on how firms should introduce AI into a workplace or consult staff about it, and the outsourcing sector is not unionised, leaving workers with almost no formal say. The UK is not as exposed on the second point, but it is not far off on the first: nobody holds a statutory duty to retrain a worker whose role has been automated here either. Britain has consultation rights that predate the technology and no obligation written for it.
| Party | What they control | What they are exposed to |
|---|---|---|
| UK client organisations | Contract terms, price expectations, AI requirements in scope | Service quality risk they cannot inspect; reputational exposure to a supplier workforce they never see |
| Philippine outsourcing firms | Deployment pace, redeployment and retraining decisions | Client pressure on both cost and AI adoption, plus competition from India |
| Global vendors (Accenture, Concentrix, Teleperformance) | The augment-versus-replace framing and retraining budgets | Public commitments that will be measured against actual headcount |
| Philippine government | Reskilling programmes, regulatory guidance | An economic model built on foreign direct investment now being repriced |
| Outsourcing workers | Very little | Redundancy law with no AI provisions, no union, NDAs traded for severance |
What UK buyers should actually do
None of this argues for pulling work back onshore as a moral gesture. It argues for treating an offshore contract as what it now is: a place where an AI transition is being executed on your behalf, at your price point, under governance you have not asked to see.
Get the AI clause out of the marketing and into the contract. If AI is in the service description, it needs the same treatment as any other material change in delivery method — disclosure of which processes it touches, human oversight arrangements, error rates, and a notification obligation when the mix changes. Concentrix, the largest operator in the Philippines, has said its AI systems should stay under human oversight. That is a commitment worth writing down in your own agreement rather than reading in a press statement.
Separate the cost story from the automation story. Require suppliers to attribute savings to a cause. A demand-driven headcount reduction and an automation-driven one carry different risks to you: the first is a resilience question, the second is a quality-assurance question. Accepting them as a single line called efficiency means you are managing neither.
Price the transition risk, not just the transition saving. A supplier moving 1.9 million people’s worth of industry through a technology shift, in a sector with no union structure and thin statutory guidance, has elevated delivery risk during the shift regardless of where it ends up. Attrition, knowledge loss and morale are real service variables.
Apply the standard you would want applied to you. Manila has pledged reskilling for over 300,000 people in the sector, and Leandro Aguirre, of the government department covering information and communications technology, was candid with the BBC about the position: “To be completely honest, I think we’re slightly behind. We have to catch up.” If a state can say that publicly about its own workforce, a client can ask a supplier what retraining actually happened to the team that used to hold its account.
Implementation Note: These are procurement and vendor-management actions, not CSR ones. Every item above is defensible purely on service-continuity grounds, which is what makes it possible to get them into a contract renewal.
Four things that will be harder than they look
The confidentiality agreements work. Everyone the BBC spoke to needed anonymity because severance was traded for silence. That means the evidence base for what actually happens during an offshore AI transition will always be thinner than the evidence base for what firms say happens. Mitigation: treat supplier attrition and tenure data as a governance metric you request routinely, because the qualitative account will not be available to you.
Augment-and-replace are the same sentence at different time horizons. Teleperformance expects the technology to absorb routine contacts, with its people moving up into harder cases, and has said it will retrain staff. Accenture says generative AI reshapes nearly every job rather than eliminating positions, and has promised billions for AI capability and workforce training. These are credible descriptions of the intended end state. They say nothing about the composition of the workforce that arrives there. Mitigation: ask for headcount trajectory alongside the retraining commitment; the two together are informative, either alone is not.
Your leverage is real and expires. Madrid’s observation that offshoring investment decisions have slowed is a buyer’s-market signal, and buyer’s markets are when contractual terms can be changed. Once AI-inclusive delivery becomes the sector standard rather than a differentiator, the disclosure terms you could have negotiated become the terms everyone declines equally. Mitigation: raise it at the next renewal, not the one after.
The reshoring temptation solves the wrong problem. If the offshore cost advantage narrows because AI compresses routine work everywhere, some UK firms will bring functions back and discover the same automation logic waiting for them domestically, applied to their own staff, without the distance that made it comfortable. Mitigation: decide your position on internal displacement before the location question forces it. The obligation defaults to whoever made the automation decision regardless of postcode.
The strategic takeaway
Aguirre’s ambition, as he described it to the BBC, is for the Philippines to reduce its dependence on inward investment from abroad and grow a domestic AI industry that generates better-paid work. He rejects sweeping new AI legislation in favour of adapting what already exists on labour, privacy and consumer rights through regulatory guidance, on the grounds that it moves faster. His stated test is whether the benefits reach people rather than stopping at employers.
That is a reasonable ambition and a long one. In the meantime, the transition is being run by suppliers under commercial pressure from clients who have not thought of themselves as participants in it. Three things follow for UK organisations.
You are a party to this whether or not you engage with it. The cost expectation you set is an input to someone else’s automation timetable. That is true of every buyer in every supply chain; AI just shortens the interval between the expectation and the consequence.
The pattern will present itself internally next. Editing model output, correcting it, teaching it the house voice — that is the same sequence Lisa described, and UK teams are already several months into it. Whether it ends the same way is a management decision, not a technological inevitability, and it is easier to make deliberately than retrospectively.
Exposure figures are a planning input, not a forecast. More than a quarter of Philippine employment is exposed; 3.6% sits in the band that carries real displacement risk. The gap between those numbers is where every useful decision lives, and the equivalent gap exists in your own organisation. Measure it before someone else’s headline measures it for you.
Take Action: Three questions before your next outsourcing review. Which processes in your delivered service are now model-assisted, and what human review sits behind them? Can your supplier attribute its last headcount change to a specific cause? And if you ran the same exposure analysis over your own UK teams — by role family, not by department — which cluster comes out highest? The third question is the one nobody is asking you yet.
Sources and further reading
This analysis responds to “‘I feel like I dug my own grave’: The workers caught in the AI transition”, reported from Manila by the BBC, with further contributions from Jaltson Akkanath Chummar and Regine Cabato. The worker accounts, the industry employment and revenue figures, and all quotations from Jack Madrid, Paul Quintos and Leandro Aguirre are drawn from that report. The workers quoted are identified by changed names at their own request.
Exposure data is from the ILO research brief Generative AI and jobs in the Philippines: Labour market exposure and policy implications, which also supplies the 3.6% highest-exposure figure, the gender split and the regional breakdown. The methodology behind it is set out in ILO Working Paper 140, Generative AI and Jobs: A Refined Global Index of Occupational Exposure.
Related Resultsense analysis: why macro payroll data is the last place an AI jobs crisis appears, the UK’s missing AI jobs policy machinery, how AI tools intensify work before they reduce it, and the erosion of entry-level career pathways.
Resultsense covers UK AI policy and its consequences for British organisations. If AI is changing what your suppliers deliver or what your own teams do, get in touch.