Production-readiness checklist for analytics leadership
Before an analytics or AI workload enters production, establish evidence for ownership, correctness, permissions, lineage, freshness, monitoring, and recovery. Define the checks for the actual data products and consumers in scope.
| Gate | Evidence | Example acceptance criterion | Owner |
|---|---|---|---|
| Ownership | Data-product record and escalation contact | Every input has an accountable owner | Data-product owner |
| Quality | Recent completeness, accuracy, and reconciliation tests | Critical checks pass or approved exceptions are documented | Analytics lead |
| Permissions | Role-based access test results | A user cannot access a restricted source through the workload | Security and data owners |
| Lineage | Source-to-output mapping and version record | A decision metric can be traced to its approved source | Analytics engineer |
| Freshness | Service expectation and alert test | Late or missing inputs are detected within the agreed interval | Operations owner |
| Recovery | Backup, rollback, or replay exercise | The team can restore the required service and verify the outputs | Technical owner |
Choose thresholds with the accountable owners before testing. A passing average quality score does not excuse a failed access boundary or unexplained financial reconciliation. Link exceptions to decisions and expiry dates.
Pair this review with the AI launch checklist, security controls, and governance decision record.
Purpose
This checklist evaluates whether organizational data foundations are strong enough to support reliable, lawful, and secure AI use. Weak data readiness is one of the most common reasons AI initiatives stall or produce poor outcomes.
Who Should Complete It
- Chief data officer or data governance lead
- Data platform and architecture leads
- Privacy and security representatives
- Business data owners for priority domains
- AI product or analytics leads
Instructions
Respond to each item with Yes, No, In Progress, or Not Applicable. Base answers on evidence such as data catalogs, quality reports, access policies, and ownership records.
Scoring Guidance
- Yes = 2
- In Progress = 1
- No = 0
- Not Applicable = exclude
Compute section and overall percentages. Interpret using the guidance below.
Assessment Questions
Data Ownership and Accountability
- Critical data domains have named business owners. (Yes / No / In Progress / N/A)
- Stewardship roles are defined and active. (Yes / No / In Progress / N/A)
- Ownership disputes have an escalation path. (Yes / No / In Progress / N/A)
- AI use cases identify required source systems and owners. (Yes / No / In Progress / N/A)
- Accountability for data defects is clear. (Yes / No / In Progress / N/A)
- Ownership records are reviewed at least annually. (Yes / No / In Progress / N/A)
Data Quality
- Quality dimensions are defined for priority datasets. (Yes / No / In Progress / N/A)
- Completeness is measured for AI-critical fields. (Yes / No / In Progress / N/A)
- Accuracy issues are tracked to remediation. (Yes / No / In Progress / N/A)
- Timeliness and freshness requirements are documented. (Yes / No / In Progress / N/A)
- Duplicate and conflict resolution rules exist. (Yes / No / In Progress / N/A)
- Quality thresholds gate AI use where needed. (Yes / No / In Progress / N/A)
- Feedback loops exist when models reveal data defects. (Yes / No / In Progress / N/A)
Metadata and Cataloging
- A data catalog or equivalent inventory exists. (Yes / No / In Progress / N/A)
- Metadata includes definitions and lineage references. (Yes / No / In Progress / N/A)
- AI teams can discover approved datasets efficiently. (Yes / No / In Progress / N/A)
- Sensitive data is classified in metadata. (Yes / No / In Progress / N/A)
- Dataset certification or trust markers are used. (Yes / No / In Progress / N/A)
- Metadata quality itself is monitored. (Yes / No / In Progress / N/A)
Privacy and Retention
- Personal data used in AI is identified. (Yes / No / In Progress / N/A)
- Lawful basis or comparable requirements are assessed. (Yes / No / In Progress / N/A)
- Retention schedules apply to AI training and logging data. (Yes / No / In Progress / N/A)
- Deletion and suppression processes work in practice. (Yes / No / In Progress / N/A)
- Purpose limitation is enforced for AI reuse. (Yes / No / In Progress / N/A)
- Cross-border transfer risks are reviewed. (Yes / No / In Progress / N/A)
- Prompt and output logs are governed for privacy. (Yes / No / In Progress / N/A)
Access Control and Security
- Access to AI-relevant data is role-based. (Yes / No / In Progress / N/A)
- Privileged access is reviewed regularly. (Yes / No / In Progress / N/A)
- Service accounts for AI systems are least privilege. (Yes / No / In Progress / N/A)
- Sensitive fields are masked or tokenized where appropriate. (Yes / No / In Progress / N/A)
- Export controls prevent uncontrolled data exfiltration. (Yes / No / In Progress / N/A)
- Vendor data access is contractually and technically controlled. (Yes / No / In Progress / N/A)
Master Data and Architecture
- Master data for key entities is reliable enough for AI use. (Yes / No / In Progress / N/A)
- Source-of-truth systems are identified. (Yes / No / In Progress / N/A)
- Architecture supports governed analytical and AI access patterns. (Yes / No / In Progress / N/A)
- Data products or domains are reusable across use cases. (Yes / No / In Progress / N/A)
- Breaking schema changes are managed with communication. (Yes / No / In Progress / N/A)
- Technical debt in critical pipelines is visible to leaders. (Yes / No / In Progress / N/A)
Compliance and Monitoring
- Regulatory data obligations relevant to AI are mapped. (Yes / No / In Progress / N/A)
- Audit evidence for data controls can be produced. (Yes / No / In Progress / N/A)
- Data incidents have defined response processes. (Yes / No / In Progress / N/A)
- Monitoring detects unauthorized access or unusual extraction. (Yes / No / In Progress / N/A)
- Model and retrieval data sources are reproducible. (Yes / No / In Progress / N/A)
- Compliance reviews occur before high-risk AI data use. (Yes / No / In Progress / N/A)
Data Lifecycle for AI
- Ingestion standards exist for AI-bound data. (Yes / No / In Progress / N/A)
- Labeling or annotation quality is managed where used. (Yes / No / In Progress / N/A)
- Training, validation, and production data splits are controlled. (Yes / No / In Progress / N/A)
- Drift in source data is monitored for critical systems. (Yes / No / In Progress / N/A)
- Deprecated datasets are retired from AI pipelines. (Yes / No / In Progress / N/A)
- Archival and deletion do not break required audit trails improperly. (Yes / No / In Progress / N/A)
Readiness for Priority Use Cases
- Each priority AI use case has a data readiness review. (Yes / No / In Progress / N/A)
- Known gaps are reflected in roadmap sequencing. (Yes / No / In Progress / N/A)
- Business owners accept residual data risk explicitly. (Yes / No / In Progress / N/A)
- Synthetic or proxy data use is governed if applicable. (Yes / No / In Progress / N/A)
- Unstructured content sources have quality and access controls. (Yes / No / In Progress / N/A)
- Retrieval corpora are curated and permission-aware. (Yes / No / In Progress / N/A)
Operating Discipline
- Data governance forum reviews AI-related issues. (Yes / No / In Progress / N/A)
- Funding exists for remediation of AI-blocking data gaps. (Yes / No / In Progress / N/A)
- Metrics for data readiness are reported to executives. (Yes / No / In Progress / N/A)
- Data and AI teams share a common intake language. (Yes / No / In Progress / N/A)
- Third-party datasets are assessed before AI use. (Yes / No / In Progress / N/A)
- Documentation standards are followed for critical datasets. (Yes / No / In Progress / N/A)
- Emergency data access paths are controlled and logged. (Yes / No / In Progress / N/A)
- Continuous improvement actions from incidents are tracked. (Yes / No / In Progress / N/A)
- The organization can explain data lineage for material AI outputs. (Yes / No / In Progress / N/A)
Interpretation of Results
High readiness (about 80%+): Data foundations can support prioritized AI initiatives with normal control vigilance.
Moderate readiness (50% to 79%): Proceed only for use cases where required domains are strong. Fund remediation before broad scale.
Low readiness (below 50%): Restrict AI initiatives that depend on weak domains. Treat data remediation as a prerequisite workstream.
Recommended Next Actions
- Identify the lowest-scoring sections tied to current priority use cases.
- Assign remediation owners and dates.
- Block high-impact launches that depend on unresolved critical data gaps.
- Feed results into the Executive Roadmap and AI Governance intake.
- Reassess after major data platform or domain changes.
Frequently Asked Questions
Does high data maturity in one domain mean enterprise readiness?
No. Readiness is use-case specific. Strong finance data does not imply strong customer interaction data.
Should every dataset be perfect before AI work begins?
No. Prioritize critical domains and set explicit quality thresholds. Perfectionism can become avoidance.
How often should this checklist be repeated?
At least semiannually, and before major AI portfolio expansion.
Summary
This checklist provides a practical evaluation of AI data readiness across ownership, quality, metadata, privacy, access, architecture, compliance, and lifecycle controls. It helps organizations determine whether data can support AI initiatives reliably before scale commitments are made.