Practical guide
AI readiness starts with enterprise data and ownership
Define the task, data, permissions, evaluation, human review and operation before selecting a model.
Page purpose
Use AI readiness starts with enterprise data and ownership to resolve a specific decision, not as a generic information list.
Use this guide as a six-gate readiness checklist for one named AI task: purpose and owner, affected decisions and impact, data authority and fitness, evaluation and failure thresholds, human workflow, and production operation. Model selection is an input after the task and gates are understood.
Record each gate as ready for a bounded pilot, conditional with named evidence, or hold. The result applies only to the stated users, data, workflow, impact, and operating boundary; it is not an organization-wide AI maturity score.
Decision trap
The ambiguity this guide is designed to remove.
- Teams start with a model demonstration before naming the decision or workflow it supports.
- Enterprise data lacks clear ownership, permission, quality, lineage, or retention treatment.
- Evaluation uses impressive examples rather than representative cases and failure conditions.
- Human review is mentioned without authority, workload, escalation, or audit design.
Comparison criteria
Comparison criteria that should be recorded.
- Is the use case ready for a bounded pilot, conditional on named evidence, or held at a specific gate?
- Which data source and purpose are approved separately for development, evaluation, and live operation?
- Which error classes trigger prevention, review, override, escalation, refusal, rollback, or incident handling?
- Which material changes reopen evaluation, data, human-workflow, security, or production acceptance?
Apply the framework
Apply the framework to a real situation.
- Complete the use-case card and separate assistance, recommendation, prioritization, generation, and automated action.
- Review every data source separately for development, evaluation, and production; assign missing evidence to an accountable owner.
- Define a non-AI baseline, error taxonomy, representative and adverse cases, thresholds, and stop conditions before model comparison.
- Walk normal, uncertain, refused, overridden, escalated, unavailable, and incident scenarios through the human workflow.
- Make the gate decision, pilot within the approved boundary, and reopen affected gates after material data, model, provider, prompt, or workflow change.
Reader output
The output the reader should produce.
- Completed use-case card naming task, users, affected parties, decision influence, owner, boundary, and prohibited uses
- Data checklist by source covering authority, approved purpose, sensitivity, provenance, quality, retention, and environment
- Evaluation matrix with baseline, representative and adverse cases, error classes, thresholds, and acceptance owner
- Human-workflow checklist covering review authority, capacity, override, escalation, refusal, fallback, and evidence capture
- Production gate card for access, security, integration, monitoring, drift, provider or model change, incidents, and retirement
Decision clarity
A clearer decision, not merely more information.
- A readiness decision whose scope, owner, evidence, conditions, and prohibited uses are explicit.
- A data-use decision separated for development, evaluation, and live operation.
- Acceptance thresholds tied to task performance, adverse cases, failure handling, and human workload.
- A pilot entry or hold decision plus the evidence required before expansion, material change, or retirement.
Misapplication
When can the framework create false confidence?
- Sensitive or restricted data may be used beyond its approved purpose.
- Evaluation may miss rare but high-impact errors or changing data conditions.
- Automation can obscure who remains accountable for the final decision.
- Model, provider, prompt, or source changes can invalidate earlier acceptance evidence.
Decision questions
Test the interpretation before applying the guide.
What readiness does a data platform establish?
It may establish access, processing, lineage, or governance capabilities for selected data. The use case still needs its own task boundary, approved purposes, data fitness, evaluation, human authority, security, integration, and operating owner.
When should model comparison begin?
Begin after the task, baseline, constraints, evaluation cases, data conditions, and operating boundary are clear enough to compare options against the same requirements.
What turns pilot evidence into a production decision?
The accountable owners compare representative evaluation, failure handling, security and data checks, human workload, integration, monitoring, incident readiness, and change controls with the approved thresholds and residual conditions.