Event driven architecture with AI

Plain language meaning

Event driven architecture with AI explains how banks can trigger AI support from business events such as onboarding, payment acceptance, posting, alert creation, case update, status change and reconciliation break while keeping event meaning stable and auditable.

This topic is about banking event design and AI triggers. It is not about loosely broadcasting every database change and calling the result an architecture.

In a real bank, AI belongs inside a governed operating model. It may recommend, rank, classify, summarise, detect weak signals or prepare evidence, but the bank still owns the customer outcome, the regulatory outcome, the accounting record, the operational closure and the audit trail. That distinction is what separates useful banking AI from uncontrolled automation.

Where it sits in the banking AI journey

This card belongs to Banking AI Architecture. The working flow is Business event, Event contract, AI trigger and context, Downstream decision support, and Audit and feedback.

Read the flow as a controlled banking journey. Each stage needs a business purpose, a source system, a data contract, a decision boundary, an exception route, an accountable owner, a fallback path and retained evidence. If any of those pieces is missing, the architecture may still run, but it is not safe enough for critical banking use.

Banking data and evidence

The important data points are event name, business timestamp, entity ID, event payload, correlation ID, trigger reason, consumer action, and feedback label. These data items matter because they can influence payment execution, risk treatment, customer communication, operational priority, compliance disposition, ledger accuracy, investigation quality and management reporting.

The evidence pack should include event catalogue, schema version, trigger log, consumer decision, feedback event, retention proof, and monitoring report. A strong bank can replay the journey from source event to AI output, control decision, human action, final status and monitoring result. A weak bank only knows that a component returned an answer.

Controls that make AI adoption safe

The core controls are event taxonomy, contract ownership, semantic versioning, consumer entitlement, duplicate handling, event retention, and impact monitoring. These controls prevent AI from drifting outside approved bank policy, scheme rules, compliance obligations, privacy expectations, model-risk governance, operational resilience requirements and customer-service commitments.

The safest design is usually not the design where AI touches everything. The safest design is the design where AI touches the right decision-support points, receives the minimum necessary context, records its version and reasoning, exposes uncertainty, allows fallback and leaves enough evidence for operations, risk, technology, compliance and audit to agree on what happened.

Architecture and resilience lens

Banking architecture must handle more than a successful happy path. It must handle degraded AI services, stale features, late events, duplicate messages, unavailable vendors, broken schemas, false positives, false negatives, manual overrides, privacy restrictions, customer complaints, regulatory review and post-incident learning.

That is why the practical architecture must show channels, source systems, data platforms, integration contracts, feature generation, model serving, decision workflows, case tools, audit stores, monitoring dashboards and fallback operation as one joined control environment. AI is one capability inside that environment, not the owner of the bank.

Regulatory and governance lens

Federal Reserve SR 26-2, dated 17 April 2026, gives revised model-risk guidance for traditional models and non-generative AI models used by banking organisations, with a risk-based approach to validation, monitoring and governance.

NIST AI RMF 1.0 uses Govern, Map, Measure and Manage functions for AI risk management, and NIST AI 600-1 adds generative-AI risk actions for source grounding, content provenance, data protection, security and human oversight.

CPMI's February 2026 updated harmonised ISO 20022 data requirements explain why structured, consistent and interoperable payment data matters for faster, cheaper, more transparent and more reliable cross-border payments.

CPMI-IOSCO Principles for Financial Market Infrastructures cover payment, clearing and settlement systems, including governance, comprehensive risk management, liquidity risk, settlement finality and operational reliability.

FFIEC Architecture, Infrastructure and Operations guidance expects bank technology environments to be governed, resilient, secure, monitored, documented and aligned to business risk.

FFIEC BSA/AML examination guidance expects suspicious activity monitoring systems to be risk-based, explainable by management, periodically reviewed and independently validated where appropriate.

OFAC's Framework for Compliance Commitments describes sanctions compliance programme components including management commitment, risk assessment, internal controls, testing and auditing, and training.

Basel operational resilience principles expect banks to identify critical operations, manage dependencies, test disruption scenarios, learn from incidents and maintain continuity of important banking services.

Diagram walkthrough

Read the diagram from left to right as Business event, Event contract, AI trigger and context, Downstream decision support, and Audit and feedback. The diagram is a control map, not decoration. It shows how banking data, AI support, policy control, human accountability and evidence should connect.

Use it as a 30-minute study method. For each box, ask which system creates the data, which rule or model acts on it, what can go wrong, who can override it, what customer or regulatory impact exists and what evidence proves the final state.

Most important mistake to avoid

The common failure is designing events around technical convenience instead of clear banking meaning, which makes AI outputs hard to explain and downstream controls hard to reconcile.

The correction is disciplined design. Keep AI bounded to the topic, connect it to banking purpose, define the control owner, preserve traceability, test the fallback path and make the final bank outcome explainable from retained evidence rather than from memory or assumptions.

Events as facts, scores as later interpretations

An accepted payment instruction, a screening hold, a model score, a policy decision and a scheme status are separate events. Each has an identifier, timestamp, source and meaning. The model consumes facts available at its decision cutoff; it cannot use later investigation results as if they were prior inputs. An event-driven design can publish score and policy actions for downstream systems, but consumers must not treat a score event as a final payment state.

Test an event delivered twice, out of order or after the decision deadline. Use business IDs and idempotency rules to preserve one intended action while retaining technical history for incident review. A replay after source correction should create a marked restated view, not overwrite the original model request. Monitor queue lag and missing events alongside model outcomes. If scoring is unavailable, a policy owner defines what event records the fallback and how operations receives it. This keeps the architecture causally honest and auditable.

A two-clock event sequence

A channel publishes a transfer at 10:00; a feature update arrives at 10:02; the hub requests a score at 10:01. The score cannot include that update even though its underlying payment event has an earlier business time. At 10:03, a duplicate submission arrives, and at 11:00 an analyst confirms that the first attempt was legitimate. The event log records all four facts with separate event and availability times. A historic back-test reconstructs the feature seen at 10:01, while a current investigation can use the later information without overwriting the original decision.

An event contract defines producer, schema, business key, correlation key, time convention, correction relationship and retention. Consumers version their interpretation. If an upstream system changes an accepted status into a generic processed status, model features and policy actions may shift despite healthy message delivery. Run semantic acceptance cases across services after source changes. An alert should distinguish queue lag from a fall in genuine customer activity, and an incident response should find affected decision IDs through reverse lineage.

The event owner should also define when a compensating event is needed. If a hold is lifted, publish a release decision linked to the original hold; do not delete the hold event. If a payment is returned after settlement, retain both states with their financial and customer consequences. A model retrained later may use the return as an outcome under a documented horizon, but a replay of the original score must stop before that later event.

Banking practice note: banking purpose

For event driven architecture with ai, banking purpose is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from event name to event catalogue. Then ask which control from event taxonomy proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

AI can assist by ranking cases, classifying defects, detecting unusual patterns, retrieving policy, summarising evidence, recommending next actions and warning about weak signals. The bank should not allow a generated explanation, a high confidence score or a fast API response to replace deterministic rules, compliance disposition, accounting proof, customer communication discipline or accountable human review.

A strong implementation records the source event, event time, data contract, model or prompt version, feature values, request and response payload boundaries, score or generated output, threshold, policy decision, user action, fallback status, monitoring result and closure evidence. That record lets operations, architecture, risk, compliance, audit and management speak from the same facts.

Banking practice note: source lineage

For event driven architecture with ai, source lineage is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from business timestamp to schema version. Then ask which control from contract ownership proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: data minimisation

For event driven architecture with ai, data minimisation is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from entity ID to trigger log. Then ask which control from semantic versioning proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: customer impact

For event driven architecture with ai, customer impact is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from event payload to consumer decision. Then ask which control from consumer entitlement proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: regulatory impact

For event driven architecture with ai, regulatory impact is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from correlation ID to feedback event. Then ask which control from duplicate handling proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: event meaning

For event driven architecture with ai, event meaning is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from trigger reason to retention proof. Then ask which control from event retention proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: API contract

For event driven architecture with ai, API contract is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from consumer action to monitoring report. Then ask which control from impact monitoring proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: feature freshness

For event driven architecture with ai, feature freshness is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from feedback label to event catalogue. Then ask which control from event taxonomy proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: model version

For event driven architecture with ai, model version is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from event name to schema version. Then ask which control from contract ownership proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: decision threshold

For event driven architecture with ai, decision threshold is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from business timestamp to trigger log. Then ask which control from semantic versioning proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: human authority

For event driven architecture with ai, human authority is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from entity ID to consumer decision. Then ask which control from consumer entitlement proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: exception handling

For event driven architecture with ai, exception handling is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from event payload to feedback event. Then ask which control from duplicate handling proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: fallback design

For event driven architecture with ai, fallback design is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from correlation ID to retention proof. Then ask which control from event retention proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: operational resilience

For event driven architecture with ai, operational resilience is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from trigger reason to monitoring report. Then ask which control from impact monitoring proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: fraud control

For event driven architecture with ai, fraud control is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from consumer action to event catalogue. Then ask which control from event taxonomy proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: sanctions control

For event driven architecture with ai, sanctions control is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from feedback label to schema version. Then ask which control from contract ownership proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: AML linkage

For event driven architecture with ai, AML linkage is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from event name to trigger log. Then ask which control from semantic versioning proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: privacy and security

For event driven architecture with ai, privacy and security is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from business timestamp to consumer decision. Then ask which control from consumer entitlement proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: audit logging

For event driven architecture with ai, audit logging is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from entity ID to feedback event. Then ask which control from duplicate handling proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: reconciliation

For event driven architecture with ai, reconciliation is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from event payload to retention proof. Then ask which control from event retention proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: status communication

For event driven architecture with ai, status communication is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from correlation ID to monitoring report. Then ask which control from impact monitoring proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: case ownership

For event driven architecture with ai, case ownership is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from trigger reason to event catalogue. Then ask which control from event taxonomy proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: monitoring cadence

For event driven architecture with ai, monitoring cadence is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from consumer action to schema version. Then ask which control from contract ownership proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: incident learning

For event driven architecture with ai, incident learning is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from feedback label to trigger log. Then ask which control from semantic versioning proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: vendor dependency

For event driven architecture with ai, vendor dependency is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from event name to consumer decision. Then ask which control from consumer entitlement proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: schema control

For event driven architecture with ai, schema control is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from business timestamp to feedback event. Then ask which control from duplicate handling proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: duplicate handling

For event driven architecture with ai, duplicate handling is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from entity ID to retention proof. Then ask which control from event retention proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: latency risk

For event driven architecture with ai, latency risk is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from event payload to monitoring report. Then ask which control from impact monitoring proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: management reporting

For event driven architecture with ai, management reporting is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from correlation ID to event catalogue. Then ask which control from event taxonomy proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: closure evidence

For event driven architecture with ai, closure evidence is not a side point. It determines whether the bank can connect the AI or ML output to a real banking process, a permitted use of data, a controlled decision and a defensible outcome. Study the topic as banking architecture first and model architecture second.

Trace one item from trigger reason to schema version. Then ask which control from contract ownership proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: banking purpose

Trace one item from consumer action to trigger log. Then ask which control from semantic versioning proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: source lineage

Trace one item from feedback label to consumer decision. Then ask which control from consumer entitlement proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: data minimisation

Trace one item from event name to feedback event. Then ask which control from duplicate handling proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: customer impact

Trace one item from business timestamp to retention proof. Then ask which control from event retention proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: regulatory impact

Trace one item from entity ID to monitoring report. Then ask which control from impact monitoring proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: event meaning

Trace one item from event payload to event catalogue. Then ask which control from event taxonomy proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: API contract

Trace one item from correlation ID to schema version. Then ask which control from contract ownership proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: feature freshness

Trace one item from trigger reason to trigger log. Then ask which control from semantic versioning proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: model version

Trace one item from consumer action to consumer decision. Then ask which control from consumer entitlement proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: decision threshold

Trace one item from feedback label to feedback event. Then ask which control from duplicate handling proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: human authority

Trace one item from event name to retention proof. Then ask which control from event retention proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: exception handling

Trace one item from business timestamp to monitoring report. Then ask which control from impact monitoring proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: fallback design

Trace one item from entity ID to event catalogue. Then ask which control from event taxonomy proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: operational resilience

Trace one item from event payload to schema version. Then ask which control from contract ownership proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Banking practice note: fraud control

Trace one item from correlation ID to trigger log. Then ask which control from semantic versioning proves the item was valid, timely, authorised, relevant and retained. If the trace depends on manual guessing, the design is not yet bank-grade.

Primary sources for further study

• Apache Kafka documentation: event streams and processing guarantees
• BIS research: ML anomaly detection in high-value payment systems