The claim is everywhere: agentic AI levels the playing field. A fifteen-hundred-seat provider in Ortigas can deploy the same models as a company with a quarter of a million employees, so scale stops mattering.
Half of that is true. The other half decides whether a company is still trading in three years, because agentic AI erodes some advantages of scale and sharply amplifies others, and the two effects do not cancel out evenly across the provider base. A frank assessment of agentic AI across the Philippine outsourcing sector.
The short version, before the detail:
- Technology cost has stopped being a barrier to entry.
- Capital for a failed attempt has become a much higher one.
- Interaction volume now compounds into an oversight advantage that smaller operators cannot buy.
- Four specific conditions decide the outcome, and they only work together.
What follows separates the two, because most of the advice on this question does not.
What scale no longer buys
Three advantages that genuinely have eroded. These are real, and a smaller provider should press them.
A technology budget as a moat
Building conversational automation in 2019 meant a platform team, a long integration programme and a seven-figure commitment — a barrier mid-sized providers could not clear, which protected the large operators.
Not any more. Capable models are available on consumption pricing, integration surfaces are standardised, and a competent team of four can stand up a working deployment in weeks. The moat drained.
Depth of training content
Large providers held vast libraries of historical interactions and the instructional design teams to turn them into training material. A smaller provider could not match the depth.
Much of that advantage has been commoditised by models that arrive already fluent in general customer-service reasoning. What remains valuable is narrow and specific: your client’s policies, your client’s exceptions, your client’s edge cases. A fifteen-hundred-seat provider running one vertical deeply can hold that material as well as anybody.
Follow-the-sun coverage
Multi-region footprints were a genuine selling point, and expensive to replicate. Automated handling of the overnight routine tier has reduced, though not removed, the need for a human in every time zone. The premium a global footprint commands for coverage alone has narrowed.
What scale now buys that it did not before
Now the other half, which gets considerably less airtime.
Capital for a bet that may not pay
Deploying agentic AI properly is not a licence fee. It is a multi-year programme of integration, measurement, retraining, governance and rework, much of which produces nothing for several quarters and some of which fails outright.
A large operator can run five such programmes, lose three, and remain healthy. A smaller provider usually gets one attempt, funded out of margin it needs for something else.
This is not theoretical. IBPAP chief operating officer Celeste Ilagan, speaking on ANC’s The Boardroom in early November 2026, said plainly that some small and medium firms cannot afford the AI investment or secure client funding for it, that some have cut staff, and that some have probably closed. She described larger players acquiring smaller ones as a positive sign of confidence rather than as a market failure. That is the association’s own chief operating officer, describing consolidation as it happens.
Data volume
Models improve with evaluation data, and evaluation data is a function of interaction volume. A provider handling fifty million contacts a year can detect a failure mode that appears in one case in ten thousand. A provider handling two million cannot — it will meet that failure mode roughly two hundred times before it has enough signal to see a pattern.
That gap widens as systems take on more of the work, because the oversight burden scales with automated volume and spotting rare errors is precisely what oversight means.
Surviving the transition year
The period in which a provider is carrying the cost of both models — the people and the automation — is the dangerous one. Large operators absorb it. Smaller ones have to finance it, usually while a client is simultaneously demanding the savings the automation was supposed to produce.
“The small providers that get through this are the ones that picked a lane early,” said John Maczynski, chief executive of PITON-Global, an independent sourcing practice that assesses Philippine BPO suppliers for overseas clients. “The ones in trouble are the generalists trying to be a smaller version of a large provider, because that is the one position where scale beats you on every axis at once.”
Where a smaller provider genuinely wins
Four conditions. They are not a menu — they travel together, and a provider meeting two of them is not in the position this section describes.
1. Vertical depth, not vertical labelling
Not “we serve financial services”, but a provider whose entire floor understands collections regulation, or claims adjudication, or pharmacy benefits, at a level a generalist cannot staff. Agentic AI raises the value of that depth, because the scarce input shifts from headcount to the judgement that grounds and corrects the system.
2. A client size the giants handle badly
A programme of eighty seats is a rounding error inside a global operator and gets attention accordingly. The same programme at a smaller provider gets the chief operating officer’s phone number. That asymmetry has grown, because demanding queues need senior attention and large accounts consume it first.
3. Speed of change
A smaller provider can revise a process in a fortnight where a large one takes a quarter. When the work mix is moving continuously, the ability to re-cut the operation quickly is worth more than it was when the mix was stable.
4. Measurement it is willing to show you
A provider small enough to show a client what its automation actually gets wrong, including the figures that embarrass it, has something to trade that scale does not supply.
“We shortlist smaller providers regularly, and almost never on price,” Maczynski said. “It is because they will show us what the automation gets wrong. The large ones show us a capability deck and a reference, and those tell you much less than they used to.”
The answer, with its conditions attached
So: can they compete? Yes, on a narrower front than the levelling narrative implies, and only where the four conditions hold together.
The claim that technology has equalised the field is a half-truth that has encouraged a number of mid-sized Philippine operators to attempt a generalist transformation they could not fund. Ilagan’s remarks suggest the cost of that error is already visible in the numbers.
The accurate version is less comfortable and more useful. Agentic AI has removed the technology barrier and raised the capital one.
If a provider has genuine depth in a single domain, clients who matter to it, and the nerve to show its own error rates, then yes — it can take work from operators fifty times its size, and increasingly does. If it has two of those three, it is competing on price against companies that can afford to lose money for longer. And if it has none of them, the levelled playing field is a field it has been invited onto by opponents who chose the ground.
