BIAN in Practice, Part 4: Connect Information Without Losing Its Authority
Connect authoritative records, business events and query-only replicated views while preserving ownership, freshness and recovery.
Perspectives · BIAN in Practice · Part 4
The lending team says a loan has been established. The customer’s dashboard still shows an application in progress. Reporting has not received the new record. Three views differ, and a support agent needs to explain what happened.
The architecture question is not simply which database is fastest. It is which record establishes each business fact, why another view may be behind, and who can resolve a discrepancy.
Part 3 translated a BIAN reference interaction into a local implementation contract. Part 4 follows the information after that interaction: from the responsible business service to events, query-only replicated stores and customer-facing views. The lending example below is a proposed local design, not a mandatory BIAN deployment pattern.
Begin with the responsibility behind the information
BIAN’s Service Landscape organises Service Domains as reference capabilities. Its information architecture connects a Service Domain’s Control Record with the business information it governs. These are useful starting points for discussing responsibility; they should not be mistaken for an instruction to create one physical database, application or microservice for every domain. Keep the selected reference release and the local mapping explicit. [1] [2]
In the fictional bank, the origination service owns the approved offer version and acceptance evidence. The authorised lending service and core establish the operational loan record. The payment service owns the disbursement workflow, with relevant posting and rail records establishing its financial outcome. A customer profile can bring these facts together, but displaying them does not transfer authority over them.
A single customer identifier helps connect records. It does not mean every connected service may change every customer attribute. Define authority by business fact and action: who changes an address, who records a consent decision, who approves an offer, and who can correct an accounting entry.
Separate a command, a fact and a read
| Interaction | Example in the lending journey | Responsibility to preserve |
|---|---|---|
| Command | Establish a loan using the approved offer version. | The receiving service checks authority, preconditions and duplicate protection before applying the action. |
| Event | Notify permitted consumers that loan establishment was confirmed. | The publisher states a fact with a defined meaning; each consumer owns its authorised reaction. |
| Authoritative query | Recover the establishment outcome using the original request reference. | The service returns the state its records can establish, with appropriate access controls. |
| Replicated query | Read a customer’s loan summary from the ODS. | The query service explains the data’s freshness and uses it only for permitted read purposes. |
An event called “LoanEstablished” should not ambiguously mean “request received” to one team and “funds credited” to another. Define the boundary of the fact. If a downstream service uses that event to initiate a separate action, it still needs its own authority, preconditions and repeat controls.
Keep event identifiers, business references and correlation references distinct. An event ID identifies a notification; a loan ID identifies a record; an establishment request ID identifies the instruction. Their relationship lets support teams trace one journey without exposing whole customer payloads.
Use replicated data to protect source-system capacity
An ODS brings replicated data from different sources into a common data store to support reads. Replication engines populate and maintain the store; the ODS itself is query-only. Moving suitable enquiries and reporting reads away from the core can reduce the load placed on transaction-processing systems.
The ODS does not establish a loan, approve a payment, reverse a debit or execute a recovery action. Any surrounding application that initiates such a task must call the authorised source service through its controlled interface. Directly editing a replicated row would change a copy, not legitimately correct the underlying business transaction.
Separating command and query models can help align different workload needs. Microsoft’s CQRS guidance describes that distinction and the possibility of lag when read data is maintained separately. It is an implementation pattern the bank may choose, rather than a condition of using BIAN. [3]
Agree freshness for each decision
Not every read needs the same freshness. A trend report may tolerate an agreed delay; confirmation of a just-submitted loan instruction may require an authoritative outcome query. The acceptable age belongs to the use case and its consequences.
For each view, record the originating service, source reference, relevant business timestamp, replication or ingestion time, and the transformation version. Where a technical watermark or sequence is available, expose it to operations. A recent ingestion timestamp alone cannot prove that an upstream source had already supplied all changes.
A customer summary assembled from several sources may contain values current at different moments. Label that limitation rather than presenting the assembly as a single consistent snapshot. If a decision requires coordinated evidence, define how the responsible service obtains it.
Set behaviour for stale or missing data: show the known state with its update time, defer a decision, or use a controlled authoritative query where capacity and access allow. Do not turn every stale read into uncontrolled polling of the core; that can defeat the resilience benefit of the replicated model.
Design events for real delivery conditions
Publish a business fact from confirmed state. A change-data-capture feed can propagate database changes, but individual row updates are not automatically complete business events. A mapper may need to combine records and interpret transaction boundaries before it can describe an established loan accurately.
Define event version, source, business reference, permitted payload, ordering scope and replay policy. A common envelope such as CloudEvents can standardise event metadata. It does not define the bank’s business meaning, guarantee delivery order or authorise the consumer’s next action. [4]
Consumers should handle repeated and late events without duplicating effects or moving a record backwards. Define how an amended fact or authorised correction is represented. Publishing history again to rebuild a read model must not accidentally resend customer messages or initiate disbursements. Separate projection rebuilding from operational side effects and verify that separation.
Resolve the mismatch without inventing another transaction
Return to the opening example. The original establishment reference resolves to a confirmed core loan record. The dashboard projection is behind, and the reporting feed has not applied that change.
The bank should preserve the confirmed establishment outcome, investigate propagation and update the read models through their governed mechanisms. It should not establish the loan again because a copy is missing. Nor should it describe the loan as disbursed until the separate payment evidence supports that statement.
Reconciliation should connect the instruction, authoritative record, emitted event and relevant projection. Give breaks a named owner, age, escalation route and correction process. A Service Domain mapping is useful when it makes that responsibility clearer during an incident.
Carry security and lifecycle controls into every copy
A reporting copy still contains banking data. Apply purpose-based access, row and field restrictions, masking where appropriate, retention and audit controls to replicated stores and event consumers. Restrict who can subscribe to which event types; a broad topic is not permission to receive every customer attribute.
Document how corrections, access changes and applicable retention or deletion decisions propagate into caches, projections and retained events. Where a projection is rebuilt, ensure it cannot reintroduce information that should no longer be exposed. Data minimisation reduces the number of copies that need coordinated handling.
Our perspective
BIAN gives teams a shared language for responsibility. The bank must turn that language into explicit rules for authoritative facts, permitted actions, freshness and recovery. A useful architecture review follows one loan from approval to source record to replicated view, then deliberately delays or repeats an event.
If the teams can explain which fact is confirmed, which view is behind and who resolves the gap, the information architecture is doing practical work. In Part 5, we will examine how to use these boundaries in a phased modernisation roadmap.
Continue with Learning
The loan exists. Why is the customer summary still pending? Distinguish confirmed source records, delayed projections and the authority to act.
Sources & further reading
- BIAN — Service Landscape.
- BIAN — Introducing the Architecture Practitioner Certification, information-architecture overview; and official BIAN artefact repository.
- Microsoft Azure Architecture Center — CQRS pattern (implementation reference).
- CloudEvents — Common event description (optional implementation reference).
Put this perspective into practice.
Explore the concepts, make a decision and test what changes when the situation changes.
Topics: Banking Architecture, BIAN in Practice, Inside the Banking Backbone, Perspective