Before automation, BSP relied on fragmented, manual intake channels — email, phone, and walk-ins — to manage complaints in a rapidly digitising market. These channels were slow, inconsistent, and poorly suited to the low-value, high-volume nature of modern digital harm, where most disputes involve small amounts but carry outsized consequences for household financial stability.

Data shows that over 61% of complaints involve PHP 10,000 or less, and nearly 90% fall below PHP 40,000, underscoring the vulnerability of lower-income consumers and the impracticality of manual resolution at scale. The majority of complaints were submitted online (86%), while only 5% came through bank branches — clear evidence that consumer protection had to become digital-first.

From a gender perspective, women account for 59% of complainants, with age and location distributions nearly identical between men and women. Importantly, the absence of large statistical differences does not imply equal experience of harm. Rather, gendered barriers — such as time constraints, digital confidence, and underreporting — may be masked by standardised complaint categories that flatten lived experience.

Scams represent one of the most persistent threats. Unauthorised online transactions (~16%) and fraud (~9%) together account for roughly a quarter of all complaints, with e-money wallets — used heavily by both men and women — representing 47.3% of all cases. These dynamics make scam detection and resolution a central consumer-protection priority.

Proto and BSP designed BOB as a unified, multilingual entry point for consumer protection, accessible via webchat and Messenger in English, Tagalog, and Taglish. Consumers can obtain instant regulatory guidance or file formal complaints at any time, removing time-of-day and literacy barriers that disproportionately affect women and lower-income users.

When complaints are submitted, BOB captures structured data — financial institution, product, transaction type, and amount in dispute — while applying AI models to analyse narrative text for sentiment, urgency, and scam-related language. Complaints are automatically routed to the appropriate BSP-supervised financial institution, and consumers receive a reference number to track progress directly through the AI agent. The BSP can then monitor all communication between the consumer and the financial institution through to resolution.

While men’s and women’s complaint categories were similar, language use varied subtly by gender and age. Older women were more likely to reference terms such as scams, unauthorised, locked out, and SIM, indicating access barriers and digital dependency risks. Younger women more frequently used deferential language — please, help, thank you — which could mask urgency if systems rely only on rigid categories rather than tone and context. These insights can directly inform continual improvements to BOB’s triage workflow, strengthening sentiment-aware automation and enabling the system to surface distress signals that traditional forms often miss.

The impact of this Proto deployment is best captured with the story of Kitty Elicay, a female financial consumer in Manila who suffered an 8 month credit card refund problem during the COVID-19 pandemic – she used the Central Bank's chatbot to effect a monitored resolution with her financial institution that saw the funds returned within six days.

BOB now functions as the Philippines’ most accessible national consumer-protection channel, combining automated guidance, complaint intake, and supervisory intelligence in a single SupTech platform. Since launch, automated responses have doubled, with more than half of all queries arriving outside business hours, confirming the importance of always-on consumer protection. The system now produces 100% structured complaints data, enabling BSP to analyse trends by product, issue type, value band, age, and sex—capabilities rarely available in traditional complaint systems.

Account management remains the largest category (~29%), but scam-related harm is a fast-increasing risk (~25%). Sentiment analysis also shows that older users—especially women—use stronger language of harm and distress, reinforcing the need for systems that interpret tone, not just tick-box categories.

To move from reactive to proactive supervision, Proto has released an insight advisor usecase with social media monitoring and scraping across massive datasets. By correlating complaints with social media, app-store reviews, forums, and news, the system detect emerging scam patterns, service outages, and institutional failures earlier—before harm escalates. These insights can be proactive notification or on-demand inquiries by consumer protection staff.

As digitisation expands financial services into underserved population segments, regulators need to match this increased volume of first-time consumers with protection systems designed for equal scale and complexity. Reliance upon existing manual processes risks failing to comply with consumer protection legislation and abandoning financial consumers to misconduct without recourse. Failure to protect consumers from these risks also affects confidence in financial technology, which could stifle growth of indigenous financial providers. Consumer protection is therefore closely entwined with efforts to improve inclusion and stability.

These themes are explored further in CGAP’s Inclusive Finance Frontiers podcast episode on the BOB AI agent, Can AI Revolutionize Financial Consumer Protection? The conversation frames AI not just as a regulatory tool for efficiency, but as an enabler of consumer trust, faster risk detection, and multilingual engagement at national scale.

As CGAP’s Eric Duflos notes in the podcast, responsible use of AI in supervision demands more than automation—it requires a collaborative ecosystem of authorities aligned around inclusion, transparency, and long-term accountability.

Similarly, Proto's learning from this partnership demonstrates that inclusive consumer protection is not just about access, but about interpretability for vulnerable consumer segments. While headline complaint patterns appear gender-neutral, current systems risk under-detecting gendered harm due to limited narrative capture, inconsistent data fields, and lack of feedback loops. BOB addresses these gaps by:

• Allowing consumers to describe harm in their own words
• Applying NLP to surface distress, coercion, and repeat-fraud signals
• Standardising data fields to enable reliable sex- and age-disaggregated analysis

These design choices strengthen both consumer trust and regulatory effectiveness, positioning BSP to continuously refine protections as digital finance evolves.