Enterprise AI readiness checklist

Purpose and accountability

• Is there one testable workflow and a named result owner?
• Are prohibited autonomous actions explicit?
• Is there a baseline against which value will be measured?

Data

• Are sources, owners and permitted purposes documented?
• Are personal, confidential and regulated records identified?
• Are storage, retention, deletion and update paths clear for prompts, outputs, indexes and logs?

Quality

• Does the evaluation set contain real tasks and expected outcomes?
• Does it test correctness, completeness, grounding, refusal and hostile input?
• Is there a minimum release threshold?

Architecture and access

• Is the hosting model justified by the data and risk?
• Are model use, data reading and tool execution separate permissions?
• Do agents have limits, timeouts, confirmation and a stop mechanism?
• Can the model provider be replaced without rebuilding the application?

Operations and economics

• Are model, prompt, knowledge and tool versions observable?
• Can the team see latency, error rate, cost and quality drift?
• Is there an incident owner and a safe non-AI mode?
• Does every material change trigger regression testing?
• Is cost measured per useful outcome, including integration and support?

An unanswered item is not an automatic rejection. It is an open risk that needs an owner, a date and a validation method.