For Sponsor Banks, AI Is the Infrastructure, Risk Discipline Is the Foundation
By Amanda Swoverland, President at Hatch Bank
Simple Subscribe
Subscribe Now!
Banking has always been a business of trust delivered through process. For decades, those processes — underwriting, monitoring, complaint handling, vendor reviews — were carried out by people, in queues, on cycles measured in days and weeks.
That cadence no longer matches what customers, partners, or regulators expect. Fintech distribution has compressed the customer journey to minutes. Examiners expect issues to be identified and resolved before they cascade. Partners expect their sponsor bank to keep pace with their roadmap, not slow it down.
That is why I have stopped thinking about artificial intelligence as an optional efficiency tool. At Hatch Bank, we are building toward an operating model where AI is the backbone. It’s the layer that will let a focused team deliver institutional-grade output across a growing roster of partners. We are not there yet, and I don’t think any sponsor bank honestly is. But the direction is clear, and the work of getting there safely, pragmatically, and in step with our regulators, is one of the most important things we are doing as an institution.
Customer Expectations Have Already Moved. Bank Infrastructure Must Follow.
When a homeowner applies for financing while a contractor is still standing in their kitchen, or when a small business owner draws on a receivables-backed line of credit at the point of sale, the bank in the background has seconds (not days) to evaluate identity, fraud signals, eligibility, and disclosures.
The same is true for the unglamorous parts of the relationship: complaints, disputes, servicing inquiries, and exception handling. Customers do not distinguish between the partner’s app and the bank behind it (even though it’s disclosed). If the bank’s tooling cannot keep pace, the partner’s product cannot keep pace.
Key insight: This is where the “efficiency tool” framing falls apart. Treating AI as a way to do today’s work a little faster underestimates the gap between traditional manual cycles and what real-time embedded finance actually requires. The right framing is structural: AI is the layer that will allow a bank to ingest, normalize, and act on partner data continuously; not in monthly reporting batches. Without that capability, customer experience degrades, partner economics suffer, and risk surfaces emerge in places where the bank simply was not looking quickly enough. Getting there is a journey, not a switch. The banks that start the journey deliberately will be the ones still standing on the other side of it.
How To Experiment with AI Agents To Take Cost And Latency Out of Manual Review
Most manual review functions in a bank look the same: a queue of documents, a checklist, an analyst, and a turnaround time measured in days. Vendor due diligence is a good example. A SOC 2 review at most institutions is a multi-day exercise involving one or two specialists, often with inconsistent output depending on who picked up the file. It is also exactly the kind of repeatable, structured work where a thoughtfully designed AI agent can add real value.
We are early in this work, but we have started piloting purpose-built AI agents, configured by our own subject-matter experts, on contained, lower-risk use cases. Our information security team has prototyped a SOC 2 review agent tethered to our third-party-relationship documentation and designed to produce a fixed, examiner-aware output structure: executive risk summary, key findings, residual risk rating with rationale, follow-ups, and explicit assumptions and limitations.
Key insight: The agent is not a generalist — it is a domain-specific assistant with a defined role, defined inputs, and a defined output format, and every output is reviewed by a human specialist before it goes anywhere near a decision or a file. As one of my colleagues likes to say, “Examiners don’t care how you think; they care how the bank behaves predictably.” A well-configured agent, properly supervised, can support that predictability over time: repeatability, continuity, staff-change resilience, and audit defensibility.
Our approach to these experiments is deliberately conservative. Each new use case goes through a vetting process before it ever touches production work: we define the scope narrowly, identify the data the agent can and cannot access, set clear human-in-the-loop checkpoints, and document the controls just as we would for any other vendor or model. If an agent graduates from playbook to production, it is because we have proven on our own data and with our own people that it behaves the way we expect it to. If it doesn’t, it stays in the sandbox.
Key takeaway: The economic potential is significant. Sponsor banks at scale are often carrying compliance functions of 30 to 50+ people, in some cases driving compliance headcount costs well into the millions annually. Over time, we believe a well-designed AI-augmented operating model will let us deliver the same regulatory coverage with a leaner team. Employees will be focused on judgment, escalation, and partner relationship work, and agents handling the high-volume, repeatable layers underneath. That is the long-term thesis. The work in front of us is proving it, one use case at a time.
Why Integrated AI Layers (Not Point Tools) Will Prove Best for Risk Reduction
The temptation in any bank is to buy a point solution for every problem: one tool for fraud, another for AML, another for fair lending, another for vendor management. That approach feels prudent because each tool is purpose-built. In practice, it fragments the data, multiplies vendor risk, and leaves the bank without a coherent view of what is happening across its programs.
Our strategy (still always fluid) is focused on a hybrid build/buy model layered on top of a centralized data foundation. Internal workflows and process automation are being designed to run on established enterprise AI platforms, configured and overseen by internal specialists. Function-specific capabilities like BSA monitoring, fraud detection, and credit analytics will ride on top of specialized vertical platforms that we evaluate continuously, connected to a centralized data layer that ingests and normalizes account, transaction, customer, and application data across our third-party relationships.
Key insight: Integration is what will produce the durable risk-reduction benefit. When AI sits on top of clean, normalized, near-real-time data, a bank can move from periodic sampling to continuous monitoring; correlate signals across programs rather than chasing each in isolation; and document exactly why a decision was made, when, and based on what data. The recent wave of pre-built agents announced for KYC, AML, audit, and credit analysis from major AI providers, paired with deeper integrations to data and core providers, signals where the industry is going: AI is moving from copilot-style assistance to embedded, governed execution inside the systems of record themselves. The banks that thoughtfully position themselves on that curve (without skipping the vetting work) will have a meaningful advantage.
The Fundamentals Still Decide Who Wins
None of this works without sound banking fundamentals underneath it. Technology can enhance banking; it does not replace credit discipline, liquidity management, capital planning, or compliance maturity. Recent banking failures were not technology failures; they were fundamentals failures. AI will accelerate whatever culture and controls it is dropped into. In a disciplined bank, it compounds the bank’s edge. In a bank that has skipped the fundamentals, it accelerates the problem.
That is why I think of AI at Hatch Bank as the operating system we are building toward, deliberately and pragmatically, rather than a product we are buying. It is also why our pace is set by safe, well-vetted progress rather than headline announcements. For any sponsor bank serious about competing in embedded finance, AI is the infrastructure on which everything else will eventually ride. Optional is the wrong word. The right word is inevitable. Banks that get there responsibly will be the ones customers, partners, and regulators trust most.
