The Philippines' EBET Act in 2026: How BPO Firms Fund AI Reskilling

The law most BPO operators still haven't used

Republic Act 12063, the Enterprise-Based Education and Training Framework Act, did something unusual for Philippine skills legislation. It didn't create a new agency. It rationalised the delivery modalities that already existed, raised the tax incentive for participating enterprises, and put money behind training that happens inside companies rather than in classrooms adjacent to them. TESDA and DOLE signed the implementing rules, so the framework is operational rather than aspirational.

The framework covers three modes. General EBET, which is on-the-job training toward NC II certification. Apprenticeships, aimed at NC III and above. And straightforward upskilling of existing workers. TESDA has requested at least P4 billion to support a minimum of 105,000 EBET scholars, and IBPAP member firms have already registered 85 enterprise-based courses.

Related reading: How Philippine IT-BPM Firms Are Reskilling for GenAI in 2026 · Malaysia's HRD Corp Levy in 2026: Stop Donating Your Training Budget · Vietnam's Semiconductor Workforce Plan in 2026: 50,000 Engineers and the Partnership Gap.

Why the timing is not a coincidence

The Philippine IT-BPM sector is on track to close 2026 at around USD 42.3 billion in revenue with roughly 1.96 million employees, against a roadmap that targeted USD 42 billion and nearly two million jobs. Hitting the number while the growth target itself was revised downward on AI tells you what's happening underneath: the sector is holding revenue with a workforce that is no longer growing the way it did.

The Bangko Sentral's own reading is that generative AI will augment rather than displace roles in the near term, while making reskilling urgent. The government separately committed about USD 25 million toward shielding BPO jobs from AI disruption. Read those two facts together and the policy position is clear enough: nobody expects mass layoffs this year, and everybody expects the entry-level seat to change shape.

That's the seat EBET is designed for. A voice agent whose call handling is now half-automated is not redundant, but the job they were hired into no longer exists in the form they were hired into it. Retraining them inside the enterprise, against a national certification, with a tax incentive attached, is a cheaper answer than backfilling the role in eighteen months at a higher rate.

Registering a course: what it costs you in practice

The 85 courses already registered by IBPAP members are the proof that this is doable, and talking to people who've been through it, the process is administratively heavier than a normal internal programme and lighter than a TESDA accreditation from scratch.

What you'll need to produce: a curriculum mapped to competency standards, named trainers with verifiable qualifications, an assessment design that a third party can audit, and records infrastructure that survives inspection. Most large BPOs already have three of those four, and the records piece is usually solved better than they expect, because client audits have forced good documentation habits for years. The one that catches people is assessment. Internal training in the sector has historically been measured on completion and quality-monitoring scores, and neither of those is an assessment in the sense the framework means.

My advice, and it goes against how most L&D teams sequence this work: design the assessment before you design the course. If you can define exactly what a trained agent must demonstrate, and how you'd prove it to an auditor, the curriculum writes itself and the registration goes smoothly. Do it the usual way round and you'll build a good course that fails on evidence.

Does any of this survive contact with a client contract?

This is the question that decides whether an EBET programme happens, and almost nobody raises it in the planning meeting.

Philippine BPO work runs on client contracts with billing models attached, and a trained agent is not automatically a more profitable agent. If your contract bills per seat per hour, moving a tenured agent from a queue into an exception-handling role reduces billable hours and improves nothing on your side of the ledger. If it bills per resolved case or carries an outcome component, the same move pays for itself quickly.

So the reskilling conversation is really a commercial conversation, and it belongs with the account team before it belongs with L&D. The operators handling this well are renegotiating billing structure and workforce composition in the same discussion, using AI-driven productivity as the argument for moving toward outcome pricing. The ones handling it badly are training people into roles their contracts don't pay for, then wondering why operations resists deployment.

There's a second contractual wrinkle. Many client agreements specify staffing profiles, certifications, and sometimes named training programmes. Registering a new internal course under the EBET framework can require client sign-off, which takes weeks and occasionally uncovers a clause nobody has read since 2019. Check this early. A programme that's fully approved by TESDA and prohibited by your largest client is a bad quarter.

Which roles to put through which mode

Duration estimates assume training runs alongside production work rather than instead of it, which is how every Philippine operator I know of actually schedules this. Pulling agents fully offline compresses the timeline by roughly half and costs more than the saved weeks are worth, unless a client transition has already freed the capacity.

The row that returns the most is the second one. Tenured agents already know the client, the systems and the escalation paths, and that context is expensive to rebuild. Moving them into exception handling and AI output review protects institutional knowledge you'd otherwise pay a recruiter to replace with someone who doesn't have it.

The skills gap behind the skills gap

Everyone reports the same shortage: GenAI and data analytics expertise. Fair enough. The shortage that will hurt more, and gets discussed less, is the supervisory layer.

Ask any operations director what happens when an AI-assisted workflow produces a wrong answer at three in the morning. The agent flags it. The team lead decides whether to override, escalate, or accept. That decision is a judgement call about model reliability, and almost no Philippine team leads have been trained to make it, because the role was defined around adherence and schedule management. Cebu and Clark operators are running into this faster than Metro Manila ones, partly because their client mix skews toward accounts that automated early.

If you register one apprenticeship programme this year, make it the team lead one. It's less marketable internally than an AI course for everybody and it's where the operational risk actually sits.

What should that programme teach? Less than you'd think, and more concretely. A team lead does not need to understand transformer architecture. They need to know which categories of task the deployed model is reliably bad at, what a plausible-sounding wrong answer looks like in their specific account, what the escalation threshold is, and who carries the decision when they're unsure. Four things, taught against real transcripts from their own queues, assessed by putting them in front of cases where the model was wrong and seeing whether they catch it.

That's a programme you can build in six weeks from material you already own. Most operators instead buy a generic AI-literacy course for supervisors, which teaches vocabulary and changes no decisions.

What to watch through the rest of 2026

Three things. Whether TESDA's P4 billion request survives the budget process intact, because 105,000 scholar slots at full funding and at half funding are different markets. Whether the incentive is claimed at scale or stays concentrated among the largest IBPAP members who have compliance teams to handle it. And whether the mid-tier operators, the 500 to 3,000-seat firms, get in at all.

That last one is my worry. The framework rewards organisations with the administrative capacity to register courses and maintain audit-grade records, which describes the top twenty firms in the sector and not much else. If you run a mid-tier operation, the move is to register through an industry association or a shared-services arrangement rather than trying to build the compliance function yourself. Waiting for it to get simpler is how the last set of incentives passed the mid-market by.

The sector spent two decades competing on cost and English proficiency, and both of those advantages are being compressed by the same technology at the same time. What replaces them is the thing the Philippines has always been quietly good at and never marketed well: handling the case that doesn't fit the script. EBET is a funding mechanism for building exactly that capability at scale, which is a better use of it than another round of certifications nobody asked for.