Written by: Agata Wojtas, Chief Commercial Officer, Digital Colliers
The OCC denied Wise's application for a national trust bank charter last month, citing incompatibility with new Federal Reserve policies. That's the headline. The subtext is more useful for anyone else in the queue.
When a regulator says "incompatible," they're rarely talking about strategy decks. They're talking about whether you can produce clean, auditable data on demand about how your business actually runs. The Wise denial isn't a one-off. It's a signal that the bar for US fintech charters has moved, and the applicants who file in 2026 with a 2019-shaped data model are going to be left behind.
The real question behind a charter review
OCC and Fed reviewers don't want your pitch. They want to know if you can answer a specific operational question, in a specific timeframe, with evidence.
Things like: show me every transaction flagged as suspicious in Q3, why it was flagged, who reviewed it, and what the final disposition was. Show me the daily liquidity position for the last 90 days, reconciled to source systems. Show me the model that decides which customers get onboarded, its training data lineage, and its false positive rate over the last twelve months.
Most fintechs can produce narrative answers to those questions. Very few can produce evidence answers on the same day the examiner asks.
That gap is what a charter denial actually punishes.
Where the data model breaks
If you're a growth-stage fintech eyeing a US charter, the pattern is predictable. You've got production systems that run the business, a data warehouse that runs the dashboards, and finance and compliance teams pulling numbers by hand into spreadsheets when anyone asks a hard question.
Most mid-market finance teams still run month-end close in spreadsheets, pulling numbers across systems by hand. Close runs 8 to 10 days at typical shops, versus under 5 at the operators who've actually built the plumbing. A regulator asking for a same-day answer doesn't care which camp you're in until you fail to deliver.
The deeper problem sits in compliance. AML transaction-monitoring false-positive rates run 85 to 95% at typical mid-market banks, and every one of those alerts creates a data artefact a regulator can ask about. If your case management system doesn't tie each alert back to the rule that fired it, the analyst who cleared it, and the model version in production at the time, you don't have an audit trail. You have a story about an audit trail.
The stack a US-bound fintech needs before filing
The operators I see clearing this bar don't buy a single tool. They build a small number of connected capabilities and they build them before they file. The rough shape:
- A canonical event log for every customer, transaction, and account state change, with immutable timestamps and source system lineage.
- A model registry that records every version of every decisioning model in production, its training data, and its performance metrics over time.
- A case management layer for compliance work where every alert, disposition, and reviewer action is queryable in seconds, not extracted from a ticketing system after the fact.
- A reconciliation engine that ties the general ledger to source systems continuously, not at month-end.
- A regulatory reporting layer that composes reports from the canonical event log, so the numbers in the report match the numbers in the business.
None of this is exotic. It's just expensive to retrofit under examiner pressure, which is why the operators who wait get denied.
Narrative versus evidence
European fintechs know this shape already. DORA has been in force since January 2025, and it forces the same discipline: prove operational resilience with data, not with policy documents. The US charter regime is converging on the same test from a different direction.
There's a broader trend behind this. Around 95% of enterprise AI projects fail to reach production or ROI, and the failure mode is almost always the same. The data underneath isn't in a shape that supports the decision the model is meant to inform. Charter applications fail for the same reason. The story is fine. The evidence underneath doesn't hold up.
If you're planning a 2026 or 2027 US filing, the useful question isn't whether your application narrative is strong. It's whether an examiner asking for operational data at 9am gets a real answer by lunch. If the honest answer is no, that's the work.