Legacy to Leadership: AI Scaling for Philippine ROI

A study on AI use in the Philippines claims that 92% of surveyed organisations have launched AI initiatives. Yet 65% remain marooned in proof-of-concept purgatory, unable to convert promising pilots into production value. That number is not a surprise. What is striking is that every executive in a room full of CIOs, CISOs, and heads of innovation knew it by heart.

That room — a roundtable titled “Legacy to Leadership: AI Scaling for Philippine ROI” convened in Manila on 18 March 2026 by FutureCIO in partnership with Boomi — gathered technology leaders from across the Philippine industry: banking, healthcare, logistics, telecoms, professional services, and government.

The agenda was clear and honest: stop talking about AI ambition and start confronting the execution gap. The conversation that followed was one of the most candid assessments of enterprise AI realism this side of the ASEAN region.

The data imperative

Ram Tallavajhala, innovation and strategy architect for Asia-Pacific at Boomi, opened with the central diagnosis. “Anyone can build an AI agent very quickly,” he said. “But the most important bit that actually powers that AI agent and its decisions is the data.”

His framing — Boomi as a “data activation company” — set the tone for a recurring theme: that agentic AI’s real bottleneck is not model capability but the integration and data foundation that feeds it.

Tallavajhala sketched the architecture separating pilots from production: an integration layer, data security, human-in-the-loop governance, context memory, and — critically — a foundation that scales.

“Picking the greatest of the large language models will not get you there,” he said, “because you will still need your basics, the foundations, and the ability to activate the data that’s there on your organisation’s networks.”

Dexter Tan, head of digital and technology innovations at Development Bank of the Philippines (DBP), was equally direct: “Data governance is a real eye-opener for some users. Without that data — how you regulate and control and protect it — everything falls apart.”

For DBP, a government institution serving legacy-trained staff, the challenge was not technology selection but re-engineering culture. “The expectations of people, that technology will automate the same process — we have to teach them the right approach,” he said.

Where modernisation starts

Before any AI agent runs, someone must pick where to begin. The roundtable surfaced a pragmatic taxonomy of entry points—and one clear favourite.

Glenn William S. Alcala, technology consulting partner and CISO at Reyes Tacandong & Co., called it the “cash trap.” “If you’re running logistics, delivery receipts are taking 15 days before a sales invoice is issued, and you cannot collect because the invoice hasn’t been issued yet — you’re trapping cash that should have been in your bank earning interest,” he said.

“In some cases, our clients have a cash trap of 300 to 500 million pesos. Those are usually due to process inefficiency.” Target the cash drain first, automate it, and the ROI case writes itself.

One roundtable delegate offered a counterpoint on healthcare. With each acquired hospital running a different patient information system, the starting question was not which AI to deploy but how to standardise foundational data.

“The classification starts with foundational systems — what’s broken and how can we start managing the rest? Then, because it’s healthcare, which processes give the most impact to the patient experience,” he explained.

One delegate from the BPO sector suggested that the starting point is often client-driven. “Sometimes it’s a client request, sometimes it’s an internal offering we want to scale — but we always look at scalability,” she said. As a concrete example: “We don’t need to hire Spanish-speaking agents anymore, because there are now AI tools we can utilise for that.”

The implication for Philippine BPOs is that the long-term competitiveness of labour costs is structural. AI does not merely automate tasks; it redraws the cost curve.

Escaping the pilot trap

The session’s sharpest debate centred on why so many AI initiatives stall at the proof-of-concept stage. The answers were instructive precisely because they varied so sharply.

Chin Sing, VP and head of business intelligence and solutions at 2Go Group, described a deliberate escape from the trap.

“There’s really no challenge for us before production deployment, because number one is executive sponsorship — the direction from the owners themselves. It’s a clear path for us. So, it’s just a matter of choosing which use cases to prioritise.” Chin Sing

2Go began its AI journey three years ago; the primary friction at that point was technology selection, not organisational will. Once the board set direction, the sense of urgency and the results followed.

Francis Pugeda, head of IT emerging technologies at Globe Telecom, offered a methodology that has since become a benchmark. “We go to a proof of value, not a proof of concept,” he said. “What we want to prove is that it provides value to us — not that the concept works.”

Globe’s validation sequence runs customer validation, solution validation, and business validation before a single line of development is written. “We still fail… but we fail without starting development.” A strict two-month window to demonstrate business benefit ensures dead ends stay cheap.

Albert Aribon, head of IT at MacroAsia Corporation, favoured the same logic: “Start small, low cost—focus on projects with high impact but low risk. So, if it fails, you can still pick yourself up and get back to it. And if it succeeds, you scale gradually.”

Another delegate pushed back: “Low cost must be high risk in reality. This AI thing, we want to make sure we’re covered from all angles.”

His preferred anchor for decisions: the spreadsheet. “At the end of the day, the fit for AI will be a fit for the culture, a fit for the employees and the workforce.”

A banking executive, taking a security perspective on the discussion, reframed the concept entirely: “There’s really no pilot because when you do a pilot, that’s already at a smaller scale, which means you’re simply testing specific objectives so that when you go to full production, you’re really ready.”

Meanwhile, a delegate from the healthcare sector located the pilot trap’s real root cause elsewhere. “Seldom do I see technology initiatives where we run pilots because the technology isn’t available. It’s most likely because the technology is available, but our users do not want to use it.”

Execution, he argued, requires strong change management from the outset and not as an afterthought once the pilot stalls.

The weight of legacy

The conversation took its most candid turn when the room addressed the single factor most slowing AI adoption. A common challenge for many businesses looking to transform their operations to enable faster response to change is the burden of legacy systems and processes. One delegate acknowledged this matter-of-factly:

“Migrating from legacy to new technology is a challenge not only of capability but of mindset. And with a very challenging telecoms market, the CFO will always ask for immediate ROI.”

The cost of replacing legacy infrastructure at scale, he noted, runs into billions of pesos, with monetisation timelines measured in decades, not quarters.

Mar Apuhin explained that many AI programs fall short not because of a lack of ambition, but because of weak data foundations. Structured data drives reliability, while unstructured data provides context. Enterprises need both—properly governed—to translate AI investment into measurable business value.

One delegate from the banking sector described how their organisation navigated this: define the key risk indicators first, then the KPIs, then the solution. “After implementing agentic AI for code refactoring and for gating specific deployments, we would already realise the solution once the KRIs go down, once the KPIs work.”

To this, another delegate, also from the banking sector, chimed in, holding a firm line on human oversight throughout: “There should always be human interaction in the loop when we are processing information through AI. We expect to be more efficient, but the quality of our outputs must not be lost.”

“It’s hard to say the data will always be ready; you need governance. A lot of times, people don’t want to take ownership of how data will be used. That impacts all the other initiatives,” he said resolutely.

Boomi’s Tallavajhala synthesised the dynamic with forensic precision. “The pilot trap, at least in my view, comes from the fact that it’s not about delivering the AI use cases, as those can be very easily and quickly delivered.

He surmised that the drag comes from the fact that if the data is not ready, it covers those agentic use cases.

“That’s where the trap comes from,” he continued. “You set up the agentic layer, but the layer below is not cooperating with multiple databases holding data that doesn’t talk to each other, wasn’t built to talk to each other, with a shiny layer on top that’s not able to get the data up.”

Why it matters for the region

The Manila roundtable is more than a Philippine story. Every friction point named in that room — legacy debt, talent scarcity, CFO pressure for 90-day returns, data sovereignty under the National Privacy Commission — echoes across ASEAN from Vietnam to Indonesia. What the conversation demonstrated is that a path through them exists, and that Philippine technology leaders are actively mapping it.

That path does not run through model selection, vendor announcements, or executive mandates alone. It runs through integration architecture, data governance as a board-level concern, proof of value over proof of concept, and a culture that treats failure as a design feature rather than a career risk.

Critically, this architectural shift does not necessitate a multi-year, multi-billion-peso overhaul from the outset. Tallavajhala maintained that a “minimum viable integration layer,” consisting of reliable core connections, API management for accountability, and audit logging, can be achieved in just 6 to 8 weeks.

“The strategic discipline here is resisting the temptation to connect everything before deploying anything,” he said. “A narrow, solid foundation outperforms a broad, fragile one every time.” Ram Tallavajhala

However, the technical scaffolding is only half the battle. In the enterprise landscape of 2026, the “human architecture” remains the more formidable hurdle. Systems fail when incentives are misaligned or when sponsors endorse programs they do not truly own. Tallavajhala advocated for a specific sequencing: get the humans aligned through small, visible wins first, then build the foundation they will actually use.

The country that figures this out first at scale, across sectors, will not merely lead in AI deployment. It will have built the blueprint that the rest of the region needs. Success, therefore, belongs to the pragmatists who refuse to be blinded by the hype.

“The organisations that move from pilot to production are not those with the most ambitious AI roadmaps. They are the ones who built the integration backbone first, defined the value case precisely, and refused to let a shiny interface layer substitute for governed, clean, and ready data beneath it,” concluded Tallavajhala.