Wholesale pilots and deposit-token experiments are moving from whiteboards into limited production. The constraint is not the ledger. It is whether the token, the cash, and the customer still reconcile to a regulated deposit record — with the same AML, sanctions, and audit trail a core already owes. Treat tokenization as a settlement fabric on top of a governed data layer, not as a second bank.
A token is still a deposit if a bank issued it
The industry conversation keeps sliding toward “crypto rails.” That is the wrong frame for a chartered bank. A tokenized deposit is a representation of a liability the institution already has. If the on-chain record and the core record diverge, you do not have innovation. You have a reconciling nightmare and a potential run on meaning.
Wholesale CBDC experiments, commercial-bank deposit tokens, and DvP settlement on shared ledgers all hit the same requirement: the movement of value must be explainable in the same language as today’s books — customer, amount, purpose, sanctions result, and who was allowed to move it.
Do not rebuild the bank. Attach a settlement layer.
Big-bang core replacement to “go on-chain” will miss the window. The workable pattern is the same one we use for embedded banking APIs: keep the licensed core as the system of record, expose a controlled settlement interface, and let a narrow orchestration layer mint, burn, or transfer representations under policy.
Applied AI can score wallet risk, detect abnormal mint/burn patterns, and assemble the travel-rule package. It should not be the ledger. Agentic orchestration can execute a DvP versus a counterparty once both sides have passed the same controls that already apply to a wire.
Examiners will ask the boring questions first
Who is the customer. Where is the money. What happens if the ledger is unavailable. Can you reverse or compensate. Can you produce the trail for this transfer the way you would for ISO 20022. Answer those on the Foundational Data Layer before you advertise finality. Tokenization rewards the banks that already know how to move a transaction through data, models, and a supervised action — not the ones that bought a new vocabulary.
