Thest

Checklist

Production AI Checklist

A practical checklist for enterprise teams moving from GenAI pilot to production: data, security, evaluation, ops, and ownership.

Use this in workshops, security reviews, and delivery planning. Print or copy freely.

Problem and outcome

  • Business outcome is measurable and owned by a named stakeholder
  • Success criteria are defined beyond “users liked the demo”
  • Out-of-scope uses are explicit

Data and permissions

  • Data sources and sensitivity classes are inventoried
  • Access control model matches source systems
  • Retention, logging, and residency requirements are documented

Architecture

  • Vendor-neutral target architecture is written down
  • Tool and API boundaries use least privilege
  • Failure modes and fallback behavior are designed

Evaluation and release

  • Golden dataset or evaluation set exists
  • Quality, safety, latency, and cost gates are defined
  • Regression checks run before each release

Operations and handoff

  • Monitoring and alerting cover critical failures
  • Runbooks exist for incidents and model/tool changes
  • Internal owners can operate the system without the delivery team