← All insights
Data Engineering

Event-Driven Data Integration: Synchronizing Salesforce with Core Banking

Batch synchronisation between CRM and core banking creates a window where both systems are confidently wrong. Events close it — if ownership is explicit.

Why batch synchronisation fails here

Nightly batch reconciliation between a CRM and a core banking platform leaves a multi-hour window in which relationship managers act on stale balances and operations teams act on stale customer data. Both systems are internally consistent and jointly wrong.

Event-driven integration reduces that window to seconds, but it only works if the fundamental question is answered first: for each field, which system is the source of truth?

Establish ownership before wiring anything

Produce a field-level ownership matrix. Core banking typically owns balances, product holdings, transaction history, and regulatory identifiers. The CRM typically owns interaction history, opportunity state, preferences, and consent records. Contested fields — contact details are the classic case — need an explicit precedence rule.

Without this matrix, bidirectional synchronisation produces update loops in which each system overwrites the other, and the winner depends on message timing.

Idempotency, ordering, and replay

Every event carries a stable business key and a monotonic version. Consumers reject events whose version is not greater than the version already applied, which makes redelivery harmless. Partition by customer identifier so ordering is guaranteed for a given customer without serialising the whole stream.

Retain events long enough to rebuild downstream state from the log. When a projection is found to be wrong — and eventually one will be — replay is dramatically cheaper than a bespoke repair script, and it produces an auditable result.

Reconciliation and observability

Event-driven does not mean reconciliation-free. Run a continuous comparison over owned fields and alert on divergence beyond a defined tolerance. Regulated environments will require evidence that the two systems agree, and a passing reconciliation report is that evidence.

Instrument end-to-end lag from source commit to target visibility, not just broker lag. Business stakeholders care about when a relationship manager sees the balance, and that is the number the service-level objective should express.

Key takeaways

What to carry into your own estate

  • Publish a field-level ownership matrix before building any flow.
  • Give every event a stable business key and monotonic version.
  • Partition by customer identifier to preserve per-customer ordering.
  • Retain events long enough to rebuild downstream projections by replay.
  • Measure end-to-end lag from source commit to target visibility.

Talk this through with Talentium Technologies

Our integration architects review middleware estates for regulated, high-throughput workloads. Start with a structured survey and we will scope an architecture review.

Request a consultation

Related articles