In July 2026 the Philippine outsourcing industry association cut its 2028 roadmap targets from $59 billion and 2.5 million workers to $43-50 billion and 1.85-2.14 million. Employment is still growing: 1.9 million workers at the end of 2025, up 80,000 on the year. The story isn’t a collapse — it’s a widening gap between revenue and headcount.
On 5 August 2026, TaskUs — one of the large outsourcing providers with a heavy footprint in the Philippines — reported its quarter. Revenue of $309 million, up 5% year on year, full-year guidance raised. And one figure that rarely makes headlines: headcount of roughly 63,200 as of 30 June, down 1.9% from three months earlier. The company attributed the decline to automation gains.
Those gains have a specific technical shape. They are automatic speech recognition systems that transcribe the customer-agent conversation in real time, language models trained on archives of past calls that surface a suggested reply while the agent is still listening, and speech synthesis engines that handle the opening of the call, identity verification and the most repetitive requests before a human ever joins the line. Together they are artificial intelligence applied to a process that for twenty years was the opposite of automation: a human in Quezon City talking to a human in Ohio. The economic point is not that these systems replace the agent. It is that they shorten the call, remove the after-call work — the manual note-taking that follows every conversation — and reduce the number of operators needed to handle the same volume.
What was actually cut
In July 2026 the IT and Business Process Association of the Philippines, the trade body known as IBPAP, completed the mid-term review of its 2028 Roadmap. The original document, launched in 2022, set two targets for the end of 2028: $59 billion in revenue and 2.5 million workers. The revised targets are $43.3-50.5 billion and 1.85-2.14 million workers, whom the association now describes as “AI-enabled”. IBPAP’s president and chief executive, Madrid, presented the new projections as a more honest assessment of where the industry is heading.
It is worth separating what happened from what did not, because the difference is the whole story. The number of workers was not cut. The forecast of how many there will be in two years was. 2025 closed with more than $40 billion in service exports and 1.9 million workers: two billion dollars and eighty thousand jobs more than the year before. Employment grew 4%, revenue 5%, against estimated global market growth of around 3%. For 2026 IBPAP projects revenue of $42.3 billion (+5%) and employment up 2.6%, at roughly 1.94 million — with the caveat that the figure has been adjusted more than once in the association’s own communications, and that thirty or forty thousand jobs separate the various versions.
So: the Philippine outsourcing sector is still growing, growing faster than the global average, and still hiring. Anyone who read headlines about a million and a half workers at risk was reading a figure five years out of date, attached to a forecast that was wrong anyway.
The gap
The interesting number is a ratio, not a total. In 2026 the sector will produce roughly $21,800 of revenue per worker. The 2028 target is around $23,600. That is $1,800 more per person wearing a headset, in two years, in a business where margins are historically thin and the price per agent-hour is the first variable clients negotiate down.
This is where artificial intelligence enters the Philippine economy, and not through the door everyone is watching. The mechanism is not “synthetic voice takes the agent’s seat”. It is that outsourcing contracts have historically been priced by volume: so many seats, so many hours, so many minutes of talk time. If an automated system cuts average handle time by 15%, a provider billing by the minute loses 15% of its revenue without having done anything wrong. To avoid losing it, the provider has to change the pricing model — move to per-resolution, per-outcome or project-based rates — or sell the client something a language model cannot do on its own.
In practice, that something looks a lot like human work that costs more: supervising the cases the system fails to close, annotating and correcting outputs to retrain the models, handling exceptions, running quality control over what the machine said. That is why IBPAP attached the adjective “AI-enabled” to its employment target. It is not promising two million phone operators. It is promising two million people working alongside an automated system. Those are different jobs, with different entry requirements.
Where it bites first: the entry level
On 19 August 2026, Dominic Vincent D. Ligot of the Philippine AI Business Association described the AI threat to the sector as “significant but manageable”, specifying that the immediate risk is not the disappearance of the industry but the automation of routine tasks and fewer entry-level positions.
That is the single most important sentence in the whole affair, and it explains why the “BPO collapse” debate is looking in the wrong place. A sector employing 1.9 million people and adding eighty thousand a year does not vanish in two years. But a sector that thins out its first-rung positions changes its social function within one generation.
In the Philippines the call centre has played a role for twenty years that has no exact equivalent in Europe or North America: a first formal job, in English, with a contract and health coverage, accessible to a recent school leaver or a graduate in a field with no immediate opening, in a country where the historical alternative is emigration — the overseas worker remittance economy. The repetitive tasks — identity verification, password resets, shipment tracking — were not merely low-value labour. They were the door. You entered there, and within two or three years you moved up to supervisor, team leader, or a back-office role in finance or healthcare administration.
Those tasks are precisely the ones a model trained on past conversations closes without human intervention. People already inside the sector stay, and probably earn more: revenue per worker rising by $1,800 has to come, at least in part, from higher skills and therefore higher pay. The person who would have entered in 2027 finds fewer doors, and doors that ask for more.
What still cannot be claimed
Two cautions. First: there is currently no public data on how many minutes of conversation in the Philippines are actually handled by synthetic voices. The available evidence concerns slowing hiring forecasts and provider statements about their own efficiency gains, not volumes. Second: even when a company attributes a headcount decline to automation, the causality cannot be isolated. At TaskUs, the same quarter also reflects the performance of its largest client on a legacy line of business. Losing a contract and automating it produce the same number on the balance sheet.
What remains is the detail that captures the moment best: within three weeks, IBPAP revised its own 2026 employment figures more than once, flagging them as subject to review. That is not sloppiness. It is what happens when the variable a trade association has always been able to forecast — how many seats it takes to bill a billion dollars — stops being a constant and becomes an unknown.