MA Modern Architecture
Reusable patternsAI in the enterprise

Architecture Patterns

Practical patterns for safely introducing AI into enterprise workflows without losing control, auditability or operational ownership.

Enterprise AI Mail Processing Pattern

Use AI to interpret inbound email and attachments, enrich context, support validation and route outcomes through governed integration.

Mailbox >> AI Interpretation >> Enrichment >> Human Validation >> API Mediation >> System of Record
  • Mailbox segregation and least privilege access.
  • Confidence scoring and exception handling.
  • Human approval before regulated or customer-impacting action.
  • Audit trail from intake to downstream submission.

API Mediation Pattern for AI Platforms

External AI capabilities should not become directly coupled to systems of record.

AI Platform >> Enterprise API Layer >> Systems of Record
  • Centralised authentication and throttling.
  • Transformation and policy enforcement.
  • Monitoring, traceability and future reuse.
  • Reduced vendor lock-in and operational debt.

Governed AI Assistant Pattern

Enterprise assistants need identity, data boundaries and escalation routes, not just a chat interface.

User >> Portal >> AI Orchestration >> Knowledge / Tools >> Human Escalation >> Monitoring
  • Identity-aware access to enterprise knowledge.
  • Controlled tool invocation.
  • Prompt and retrieval boundaries.
  • Operational monitoring and feedback loops.

Human Escalation Boundary Pattern

Classify AI activity by risk so automation, recommendation and human decision-making remain clearly separated.

  • Green: summarisation, extraction and low-risk assistance.
  • Amber: recommendations and prioritisation requiring review.
  • Red: regulatory, financial or legal decisions requiring human ownership.