Every bank says data is a strategic asset. Most balance sheets do not reflect it. The work of treating data as an asset is unglamorous, and that is exactly why it is the moat.
Principle one: govern at the source
Governance is not a downstream cleanup project. It is a contract written at the system of record. When ownership, lineage, and quality SLAs are defined where the data is born, every downstream agent inherits them for free.
Principle two: model the domain, not the report
If your data model reflects last year's regulatory report, you will rebuild it for next year's. Model the underlying business — client, mandate, position, instruction — and let reports be projections, not the source of truth.
Principle three: instrument everything
An agentic system without observability is a black box you cannot defend in a regulator meeting. Treat every read, every inference, every override as telemetry. The cost is small. The optionality is enormous.