Automation was supposed to collapse offshore rates. Published bands have barely moved, because the work left on the floor is not the work that was automated.
The rate card did not fall. Anyone who modeled AI call center outsourcing pricing in the Philippines on the assumption that software would drag hourly rates down has watched published bands hold between roughly $10 and $18 an hour, fully loaded, while the technology absorbed a growing share of contacts. The economics did change. They changed somewhere else.
Key takeaways
- Published Philippine bands have held at about $10 to $18 an hour by role, despite heavy automation.
- Automation removes the cheapest contacts first, so the average remaining contact is harder and costs more to staff.
- The useful comparison is cost per resolved outcome, not cost per seat or per hour.
- A vetted customer service seat is published near $12 to $14 an hour fully loaded, against roughly $32 to $35 onshore.
- Buyers should stop paying for volume capacity they will not use and start paying for the judgment layer they now depend on.
This is the part of the automation story that procurement decks keep getting wrong. The savings are real, but they do not arrive as a cheaper hour. They arrive as fewer hours, bought at a slightly higher quality, doing work substantially more difficult than what the same program handled in 2023. Buyers who separate capability pricing from capacity pricing, the distinction at the center of PITON-Global’s call center outsourcing framework, tend to spot the shift a year earlier than those who do not.
What has AI actually done to Philippine rates?
It has shifted the mix rather than the price. Entry-level support roles still sit at the bottom of the published range. Quality assurance, audit and compliance-facing roles sit several dollars higher, and those are the roles automation creates demand for, because somebody has to check what the system did.
The result is a program whose blended rate rises even when every individual band stays flat. A buyer comparing this year’s invoice against a 2023 baseline sees a higher average and concludes the provider has raised prices. Usually the provider has not; the composition of the team has changed underneath a stable rate card.
Why doesn’t automation lower the hourly rate?
Because it removes the cheapest work first. The contacts a model handles unaided are short, repetitive and well documented, password resets, balance checks, delivery status. Those were always the lowest-cost minutes on the floor.
Strip them out and what remains is disputes, exceptions, distressed customers, regulated conversations and anything where the record is incomplete. That residue needs more experienced people, longer training and closer supervision. Paying less per hour for it is not a negotiating win; it is a staffing plan that will fail its first peak.
“Clients tell me they want the AI discount,” says John Maczynski, chief executive of PITON-Global, an outsourcing advisory and supplier sourcing firm. “What they are describing is a discount on work they are no longer sending. The hours that remain are the hardest hours in the operation, and those have never been cheap.”
What is the real unit of cost now?
Cost per resolved outcome, measured end to end. A seat is an input. An hour is an input. Both were adequate proxies when one person handled one contact from start to finish, and both became misleading the moment a contact could be opened by software, escalated to a person, returned to software for fulfillment and closed by a supervisor.
The arithmetic that matters takes total program cost, people, supervision, quality assurance, technology, the client-side management time nobody books, and divides it by contacts resolved without recurrence. Programs that look expensive per hour frequently look cheap on that measure, and the reverse is true often enough to be worth checking before a renewal.
Ralf Ellspermann, chief strategy officer at the firm, puts the failure mode simply: “Every buyer can tell me their rate per seat. Very few can tell me what a resolved case costs them, including the ones that came back twice. That second number is the only one that moved when AI arrived.”
How much does an offshore seat actually cost in 2027?
Published comparisons put a fully loaded vetted Philippine seat near $12 to $14 an hour, or roughly $23,000 to $27,000 a year, against about $32 to $35 an hour and $61,000 to $67,000 for an equivalent Western build. Savings in the range of 50% to 65% remain the standard planning assumption.
Those figures describe fully loaded cost, which is where quotes diverge. A rate that excludes supervision ratios, quality sampling, retraining after each model change and cover for attrition is not a lower price for the same thing; it is a price for a different thing. The gap usually appears in the fourth month, when the first model update lands and nobody has been funded to re-test the affected intents.
What should buyers refuse to pay for in 2027?
Three things, each of which looked reasonable in a 2023 agreement.
Long minimum volume commitments
Any commitment that assumes contact levels will hold flat for three years is unsafe now, because a single automated intent going live can step them down overnight.
Duplicated technology line items
Charges for platform capability the client already licenses elsewhere in its own stack, rebilled as part of the service.
Reserve headcount for a flattened peak
People held against a seasonal surge that the automation was explicitly bought to absorb.
What deserves funding instead is the judgment layer: the audit capacity that reviews what the system did, the senior agents who take the escalations, and the trainers who keep both current as the models change.
Maczynski argues the negotiation has inverted: “Five years ago the buyer pushed the rate down and the provider protected margin by thinning supervision. Now the buyer who thins supervision is the one who pays, because the supervision is what makes the automation safe to run.”
None of which makes the Philippines cheaper than it was. It makes the country’s role different, less a place to buy hours, more a place to buy the oversight that makes automated service defensible. The rate card survived. What it buys did not.
