Yano.AI Technologies Inc.

Posted on Oct 5 Originally published at yanoai.tech

Digital Payments Hit 64.7% in the Philippines. The Harder Problem Is the Software Running Them

#ai #fintech #banking #philippines

Digital payments took 64.7% of all retail transactions in the Philippines in 2025, up from 57.4% a year earlier. In that same year QR Ph payments overtook debit and credit card transactions for the first time ever, logging 2.47 billion transactions (Source: BSP, 2026). The plumbing underneath Filipino commerce is now software-defined, and that changes what a bank actually has to supervise.

The Rail Is No Longer the Bottleneck

As of 31 July 2026, the transaction accounts of BSP-supervised institutions participating in InstaPay already cover 99% of all transaction accounts in the system (Source: BSP, 2026). QR Ph person-to-person participants make up 94% of that scheme's total, and the merchant side sits at 92% (Source: BSP, 2026).

Growth has not flattened. InstaPay QR posted average quarterly growth of 33.0% in volume and 31.7% in value from Q2 2025 to Q2 2026, while QR Ph posted 27.4% and 26.2% on the same basis (Source: BSP, 2026).

Cost has fallen alongside it. As of July 2026, InstaPay fees range from PHP 0.00 to PHP 35.00 (Source: BSP, 2026). That is the range at which on-store payments, transfers and merchant settlement become software line items rather than bank branch operations.

This is where the earlier open-weight models conversation turns into a Philippine question rather than a Silicon Valley one.

A Governance Framework Arrived in June

In June 2026 the BSP issued Memorandum No. M-2026-031, a guidance paper on Governance Principles for Artificial Intelligence in Financial Services (Source: BSP, 2026). It is built on five principles: Sustainability, Transparency, Accountability, Responsibility and Security, which the regulator abbreviates as STARS (Source: BSP, 2026).

Two details matter more than the acronym. The guidance extends to outsourced service providers supporting AI-related activity under a shared responsibility model, so a bank cannot outsource a model and outsource accountability along with it (Source: Baker McKenzie, 2026).

And the principles are expressly non-binding and voluntary, positioned as minimum supervisory expectations rather than prescriptive requirements (Source: Baker McKenzie, 2026). That is a deliberate choice, and it puts the burden back on each institution to produce its own framework.

Deputy Governor Lyn Javier framed the intent as encouragement rather than constraint: "We want BSFIs to take advantage of AI, especially to serve their customers, and do so while being guided by developing global standards" (Source: Asian Banking & Finance, 2026).

Open-Weight Models Just Moved the Cost Floor

On 5 October 2026, Reflection announced Beam, a sparse mixture-of-experts model with 501 billion total parameters and 23 billion active, pretrained on 23.8 trillion tokens (Source: Reflection, 2026).

The company reports Beam reaches reasoning scores comparable to GLM-5.2 while using three to four times less inference compute, and says it will release the weights under an Apache 2.0 licence later this month (Source: Reflection, 2026). Reflection's own figures describe that comparison as an approximate compute estimate rather than measured inference cost, so treat the efficiency claim as vendor-reported.

One point needs stating plainly: at the time of writing those weights had not been released. Early access runs through a waitlist, and the model is still in red-teaming (Source: Reflection, 2026).

The Philippine relevance is not that Beam exists. It is that self-hosted open weights make fraud-pattern detection, document review and customer onboarding economically viable for institutions that could not previously justify the inference spend. Control over where transaction data physically sits is not a nice-to-have when you answer to the Data Privacy Act and to a regulator that just wrote Security and Transparency into its supervision expectations.

The Identity Layer Is Being Rewired

In August 2026 the BSP told supervised institutions it wants National ID Authentication Services integrated into customer due diligence protocols (Source: BusinessWorld, 2026). The draft rules push institutions toward Tier 2 authentication, which includes e-KYC services using pre-agreed demographic data fields, and require the appropriate tier to be selected per customer, product and transaction risk (Source: BusinessWorld, 2026).

Rollout is staged rather than immediate. Phase one covers universal and commercial banks with retail services, digital banks, electronic money issuers and virtual asset service providers, three months from issuance. Phase two covers thrift, rural and cooperative banks plus payment operators with know-your-merchant functions, six months after (Source: BusinessWorld, 2026).

The substrate is already in place. As of end-October 2025, 90.29 million Filipinos - 80% of the population were registered to the National ID system, according to data from the Philippine Statistics Authority (Source: BusinessWorld, 2026). A regulator asking for biometric identity verification at onboarding is asking for something that is already built.

Public Markets Just Underwrote the Thesis

On 2 October 2026, Mynt, the parent company of GCash, priced its initial public offering at PHP 6.60 per share, roughly PHP 53 billion or about $845 million in total (Source: Reuters, 2026). That sits well below the PHP 10 ceiling the company had floated, a ceiling that implied a valuation of roughly $10.7 billion on about 66.90 billion shares outstanding after the offering (Source: The Wall Street Journal, 2026).

Pricing below your own ceiling is ordinary market behaviour rather than a verdict on the business. What it does confirm is the scale of the base: a very large share of Filipino adults already move money through a single app, which is precisely what makes national mandates like NIDAS affordable to comply with.

Three things landed inside one quarter: a payments system reaching 99% of transaction accounts, a governance acronym for the software running it, and a public market price for the country's largest wallet.

FAQ

Q: Do the STARS principles carry the force of a regulation?
A: No. M-2026-031 is guidance, expressly non-binding and voluntary, and the BSP positions it as minimum supervisory expectations rather than prescriptive rules (Source: Baker McKenzie, 2026).

Q: What happens to a bank that outsources its AI to a vendor?
A: The guidance explicitly extends to outsourced service providers under a shared responsibility model, so contractual and oversight arrangements need to reflect that split rather than assume the vendor absorbs it (Source: Baker McKenzie, 2026).

Q: Can a Philippine bank run an open-weight model today?
A: Reflection states Beam weights will be released under Apache 2.0 later in October 2026, following a red-teaming and evaluation phase, and early access is currently waitlisted (Source: Reflection, 2026).

Q: Which institutions face the National ID rules first?
A: Phase one covers universal and commercial banks with retail banking services, digital banks, electronic money issuers and virtual asset service providers, three months from issuance of the memorandum (Source: BusinessWorld, 2026).

Key Takeaway

Each of these shifts is useful alone. Together they mean the binding constraint for Philippine fintech has moved from building rails to demonstrating, in writing and under supervision, how the code behind them behaves.

The useful question is not whether your institution has an AI policy. It is whether you could hand that policy to a BSP examiner tomorrow and reconstruct what your model decided last Tuesday, on which data, under whose approval.

DEV Community