AI readiness starts before the model
An AI initiative becomes enterprise architecture work when it needs reliable context, accountable decisions and a safe operating model.
Before selecting the model or agent framework, prove that the business task, data permission, control boundary, owner and evidence of success are defined.
1. Define the task in business terms
“Deploy an AI assistant” is not an outcome. “Reduce the time required to triage supplier delivery exceptions without compromising approval controls” is a testable business problem.
2. Test readiness through six lenses
| Lens | Architecture question | Typical evidence |
|---|---|---|
| Business capability | Which decision or task should improve? | Owner, process map, baseline and acceptance criteria |
| Data context | What information may the system use? | Source systems, lineage, access rules and data quality |
| Integration | Which systems will AI read or change? | Approved APIs, identity propagation and transaction boundaries |
| Risk & governance | What may go wrong and who accepts the risk? | Risk register, controls, auditability and escalation |
| Human accountability | Who reviews, overrides or approves action? | Decision-rights matrix and fallback process |
| Operations | How is behavior monitored and changed? | Evaluation, incident response, versioning and monitoring |
3. Choose the minimum safe architecture
A retrieval-supported assistant, a deterministic rule and an autonomous agent have different control demands. Do not choose autonomy where a narrower assistive pattern meets the business need with lower irreversible risk.
4. Establish evidence before scaling
Measure task accuracy against a representative test set, exception rates, reviewer effort, process cycle time and incident handling. Separate pilot results from production readiness.
Failure mode: declaring enterprise AI readiness from the availability of an LLM, vector store and cloud subscription alone.
Reference framework
NIST AI Risk Management Framework. Adapt governance and control design to the applicable enterprise, sector and law.