Architecture Lens

Start AI readiness with a business decision map

AI readiness is strongest when the first question is not which model to use, but which recurring decision or workflow needs better evidence, faster response, or safer automation. A useful intake separates business value, data availability, risk, user adoption, operating cost, and support ownership before prototype work starts.

  • Name the decision, user group, source systems, expected improvement, and review owner.
  • Score data sensitivity, data quality, policy exposure, workflow fit, and measurement difficulty.
  • Prefer a narrow first release with acceptance criteria and rollback options.
AI readiness path
Use case

Decision, workflow, users, and measurable value.

Data fit

Access, quality, lineage, privacy, and retention.

Risk model

Policy, harm, security, cost, and human review.

Pilot release

Thin scope, test set, telemetry, and owner.

Original InSkyto diagram informed by the NIST AI Risk Management Framework.

References

NIST AI Risk Management Framework

Delivery Pattern

Use evaluation before automation

Teams should build evaluation sets and human review habits before they automate consequential steps. This protects the organization from confusing a polished demo with a reliable business capability.

  • Create examples for normal work, edge cases, restricted data, adversarial prompts, and escalation paths.
  • Track quality, refusal behavior, latency, cost, source coverage, and user feedback.
  • Document where the AI can recommend, where it can draft, and where a human must approve.

Operating Model

Define ownership before production pressure

The first production AI workflow needs owners for data, application behavior, model evaluation, security review, user training, support, and incident handling. Without those owners, the pilot becomes a shared orphan at the exact moment adoption starts to grow.