Invisible Banking: The Art of Seamless Financial Services

In this article:

• How Bank of Punjab used AI to lend at a scale that no human underwriting team could ever reach, and why the real challenge wasn't the model, it was earning trust in it.
• Bank of America's Erica resolves 98% of client inquiries, not because the AI is smarter, but because of where they chose to put it.
• The Synapse collapse of 2024 exposed a structural flaw in embedded finance that the industry still hasn't fully reckoned with: when three parties share a customer, who is actually accountable?
• The banks that win the next five years won't be the ones that disappear into the background fastest. They'll be the ones that remain trusted while doing so.

This week at Money20/20 Asia in Bangkok, professionals from across the financial industry gathered to discuss what it means to build banking infrastructure that actually works for customers. Among the sessions that cut closest to the strategic core was a panel titled Invisible Banking: The Art of Seamless Financial Services, moderated by Annabelle Lin, Co-Founder and Chief Revenue Officer at Nextvestment. What the three panelists revealed was less about technology and more about trust: what it actually takes to make banking feel effortless while keeping everything that matters intact behind the scenes. The answers were more candid, and more urgent, than most industry panels allow.

‍

Three Institutions. One Shared Conviction.

‍

Arthit Sriumporn, EVP of Digital Channels at Siam Commercial Bank, has led one of Thailand's most ambitious branch-to-digital transformations. Waqas Javed, Divisional Head of Data Analytics and AI at Bank of Punjab, is rebuilding what credit access looks like for millions of underserved borrowers across Pakistan. And Terence Tan, COO of Global Payments Solutions for APAC at Bank of America, operates at the infrastructure layer powering some of the region's most complex institutional clients. Three different institutions, three different mandates - and one shared conviction: that the hardest part of invisible banking has never been the technology.

‍

"Banking is necessary. Banks are not." Bill Gates said that in 1994. Thirty years later, the industry is finally building toward it, and the hardest part isn't the technology. -Terence Tan, COO of Global Payments Solutions APAC, Bank of America

‍

‍

The Simplest Customer Experience Hides the Hardest Internal Work

Every panelist agreed on one thing immediately: invisible banking is not simple banking. If anything, the simpler it feels to the customer, the more complexity had to be absorbed inside the institution first.

Arthit Sriumporn described SCB's decade-long journey from being one of Thailand's most branch-heavy banks to a digital-first institution. But the biggest obstacle wasn't mobile app development or legacy system migration. It was people.

‍

"The most difficult part was change management within the bank. We had to move people from transacting staff into skilled staff, where they could sell, where they could help clients fulfill their needs instead of just processing transactions." - Arthit Sriumporn, Siam Commercial Bank

‍

The observation cuts to something the industry talks around but rarely confronts directly: digital transformation is an HR challenge as much as a technology one. Closing branches displaces people. Reskilling takes time and investment. And institutions that fail to manage this transition internally will see it surface as friction externally, in the product, in the service, in the brand.

‍

Waqas Javed echoed the point from a different angle. The biggest misconception in invisible banking, he argued, is that it is a technology problem. It isn't. The technology exists. What differentiates institutions is whether they build it around the customer and whether the staff delivering it actually understands what changed.

‍

‍

AI's Real Value Isn't Automation. It's Scale With Fairness

Bank of Punjab's story is one of the most compelling AI-in-banking use cases in the region. Pakistan has 70% adult banking penetration but only 2.5% access to formal lending. The gap isn't appetite. It's data. Traditional credit scoring fails thin-file customers who have never borrowed formally before.

‍

Bank of Punjab's answer: build AI that assesses creditworthiness without requiring a credit history. Their models pull from alternative data, satellite imagery for agricultural lending, behavioral signals, psychometric evaluations, and they delivered lending to over one million farmers within a four-month window.

‍

‍

"You can't achieve that kind of scale without really relying on tools like AI. But the challenge isn't the model accuracy. It's trust in the model." -Waqas Javed, Bank of Punjab

‍

‍

This is a distinction the industry tends to collapse. A model that performs well in training doesn't automatically earn customer trust at the point of decision. For a first-time borrower, particularly someone who has spent their life outside the formal financial system, a rejection needs to be explainable. The AI needs to be legible, not just accurate.

‍

Javed's team built for explainability from the start. The customer needs to understand why they were declined. The model needs to be contestable. And the institution needs to be accountable for its outputs even when those outputs are generated by an algorithm rather than a loan officer.

‍

Arthit framed a related challenge for mature digital banks: personalization at scale. SCB is moving toward what he called "segment of one," hyper-personalized offers driven by behavioral data. But he was equally candid about the risk that comes with it. The same data architecture that enables personalization creates an attack surface. Data breaches don't just cost money. They cost the trust that took decades to build.

‍

‍

Erica, Adoption, and the Deployment Mistake Most Banks Still Make

Terence Tan offered a masterclass in AI adoption through the lens of Bank of America's Erica, the virtual assistant that today resolves 98% of retail client inquiries through the mobile app, and is used by over 90% of the bank's internal workforce.

‍

The insight behind Erica's success wasn't the model. It was where they put it.

‍

"More often, AI is deployed as a dashboard or a UI that is outside of the user's action steps. That reduces the probability of adoption. Where we see AI being successful is when it's embedded in actions the user would already take." - Terence Tan, Bank of America

‍

This is the deployment mistake most institutions are still making. AI tools get built and then placed adjacent to workflows rather than inside them. A recommendation engine that lives one tab away is a tool no one will check. An insight that appears at the moment of decision, during a payment flow, inside a lending application, within a cash management platform, is infrastructure.

‍

Bank of America also made an unusual sequencing choice: they rolled Erica out to retail clients first, then to internal staff. The result was that internal adoption was built on a product already proven at scale rather than a prototype. Technology help desk calls dropped 50%.

‍

‍

The Risk That Invisible Banking Makes Invisible: Accountability

The session's most pointed exchange came when Annabelle Lin pushed the panel on the risks the industry isn't talking about loudly enough. Javed's answer was direct: invisible banking is convenient for banks, but if the trust layer isn't visible to customers, the product fails at the human level. Especially for newly included populations using formal financial services for the first time, every transaction confirmation matters. Every status update builds or erodes confidence.

‍

But it was Terence Tan who named the structural risk: the tripartite accountability gap.

‍

‍

We cannot talk about embedded finance risk without mentioning what happened to Synapse in 2024."

‍

For those who missed it: Synapse was a middleware fintech that served as the ledger controller between banks and the fintechs offering consumer-facing financial products. When Synapse went bankrupt in 2024, a $200M+ discrepancy emerged between its ledger and actual bank balances. Over 10 million retail users were locked out of funds. Some never recovered their money.

‍

The problem wasn't malice. It was architecture. One party in a three-party chain held all the ledger visibility. No one else could reconcile in real time. When that party failed, the chain collapsed and the customers at the end of it had no recourse.

‍

The FDIC has since introduced rules requiring banks in tripartite arrangements to maintain ledger visibility. But Tan's broader point stands: the industry needs clearer frameworks for who is accountable for what in embedded finance arrangements, before the next failure, not after.

‍

The parallel for institutional banking is equally sharp. When financial services are embedded into ERP systems, marketplace platforms, and corporate treasury tools, the bank's controls need to move upstream, into the architecture, not the interface. If controls only exist at the UI layer, they can be exploited. If they're buried in the orchestration layer, they hold.

‍

‍

Who Wins and Who Loses in the Next Five Years

The closing exchange was sharp. Asked which institution type has the most to lose and gain as invisible banking becomes the norm, the panel gave three answers that pointed in the same direction. Arthit: "Banks have the most to lose if they don't change. The ecosystem around banks has the most to gain if they do it well." Javed added a note of realism that often gets lost in transformation narratives: 60% of AI-based projects fail. Institutions need to get comfortable with that, keep experimenting, and keep pivoting. The risk isn't failure. The risk is stopping after failure. Tan closed with the clearest frame: the institutions that gain the most won't be the ones that disappear into the background. They'll be the ones that remain trusted while doing so.

‍

Strategic Takeaways

• Invisible banking requires visible trust. The less friction the customer feels, the more deliberate the institution must be about where accountability surfaces.
• AI's value is scale and inclusion, not automation for its own sake. The models that matter are the ones that bring new customers into the financial system fairly.
• Deployment context determines adoption. AI embedded in existing user flows outperforms AI deployed as standalone dashboards. Every time.
• Tripartite accountability is an unresolved structural risk. Post-Synapse, every institution in an embedded finance arrangement needs real-time ledger visibility and clear contractual accountability.
• The change management problem is still underestimated. Technology can be deployed faster than cultures can shift. The gap between the two is where transformation stalls.

‍