Data & Knowledge Readiness Audit — Quick Assessment & Remediation Guide

Welcome — What this audit helps you do

This quick audit evaluates whether your information environment is ready to produce trustworthy, actionable AI outputs. It covers seven practical domains: data inventory, provenance & lineage, labeling status, access & governance, freshness, privacy & compliance constraints, and integration complexity. For each domain you'll find assessment questions, suggested scoring, concrete remediation steps, and estimated effort buckets so leaders can prioritize work before starting a pilot.

How to use this audit

1. Answer the questions in each domain honestly (Yes/Partial/No). For partial answers, note the gaps.
2. Score each question: Yes = 2, Partial = 1, No = 0. Sum scores per domain and overall.
3. Use the readiness bands to interpret results and select the recommended remediation tasks.
4. Record outcomes and consider converting this audit into an interactive form to save responses and track progress over time.

Scoring & readiness bands

Each domain has multiple questions; total possible score depends on how many questions you answer. Normalize domain scores to a 0–100 scale (actual / possible × 100) and average domains for an overall readiness percentage.

• 80–100%: Ready for an expanded pilot with attention to minor gaps.
• 50–79%: Pilot possible but requires focused remediation (labeling, access, governance).
• 0–49%: Pause pilots. Invest in foundational data and governance work first.

Assessment sections (questions, remediation, effort)

1) Data Inventory

Questions

• Do you have a documented inventory of relevant datasets and knowledge repositories?
• Are dataset owners and primary contacts identified?
• Are datasets cataloged with basic metadata (purpose, schema, sample size, sensitivity)?

Typical remediation steps

• Create or update a dataset catalog with owners and key metadata.
• Prioritize datasets for pilot use based on size, completeness, and business value.

Estimated effort

• Low: Catalog exists; needs augmentation — days to 2 weeks.
• Medium: Partial catalog + interviews — 2–6 weeks.
• High: No catalog; discovery across systems — 1–3 months.

2) Provenance & Lineage

Questions

• Can you trace the origin and transformations applied to the data?
• Are ETL/ingestion processes documented and reproducible?

Remediation

• Document end-to-end data flows and major transformations.
• Instrument pipelines to capture lineage metadata going forward.

Estimated effort

• Low: Minor documentation gaps — days to 2 weeks.
• Medium: Add lineage capture to pipelines — 3–8 weeks.
• High: Rebuild/instrument multiple legacy pipelines — 2+ months.

3) Labeling Status (for supervised models)

Questions

• Are labels available, consistent, and documented?
• Is inter-rater agreement measured for human-labeled data?
• Is there a plan and capacity to label additional examples if needed?

Remediation

• Create label definitions, examples, and a labeling QA process.
• Estimate labeling volume and decide on in-house vs. vendor labeling.

Estimated effort

• Low: Minor label cleanup — 1–3 weeks.
• Medium: Define taxonomy + label a pilot set — 3–8 weeks.
• High: Large-scale labeling and QA program — months.

4) Access & Governance

Questions

• Are access controls, roles, and approval processes defined and enforced?
• Is there an accountable data steward or governance body for critical datasets?

Remediation

• Define stewardship roles and implement role-based access controls (RBAC).
• Create a lightweight data governance checklist for pilots.

Estimated effort

• Low: Assign stewardship and approvals for pilot datasets — 1–2 weeks.
• Medium: Implement RBAC and approval workflows — 3–6 weeks.
• High: Enterprise governance program rollout — months.

5) Freshness & Coverage

Questions

• How frequently is the data updated relative to use cases?
• Are there known gaps, seasonal biases, or missing populations?

Remediation

• Define acceptable freshness windows for each use case and monitor lateness.
• Plan targeted data collection to fill coverage gaps before scaling.

Estimated effort

• Low: Adjust ingestion cadence or use caching — days to 2 weeks.
• Medium: Add collection sources or corrective pipelines — 3–8 weeks.

6) Privacy, Compliance & Constraints

Questions

• Are sensitive attributes identified and handled appropriately?
• Are legal/regulatory constraints (GDPR, HIPAA, contract terms) documented for datasets?

Remediation

• Perform a privacy review for pilot datasets and apply minimization, anonymization or access restrictions as needed.
• Log data processing purposes and lawful bases.

Estimated effort

• Low: Apply access controls and simple masking — 1–3 weeks.
• Medium: Conduct formal DPIA or legal review — 3–8 weeks.
• High: Redesign systems to remove restricted data — months.

7) Integration Complexity

Questions

• How many systems must be integrated to serve the AI use case?
• Are there existing APIs, feeds, or extract processes that support integration?

Remediation

• Map required integrations, estimate effort, and consider intermediate data marts for pilot pace.
• Prototype a single integration to validate feasibility quickly.

Estimated effort

• Low: One well-documented API or feed — days to 2 weeks.
• Medium: Multiple systems with moderate work — 3–8 weeks.
• High: Legacy systems requiring custom adapters — months.

Example output & recommended next steps

After scoring, produce a one-page summary: overall readiness percentage, top three domain gaps, recommended remediation (with estimated effort bucket), and a proposed pilot data scope. Use that summary to decide whether to:

1. Proceed with a small, well-scoped pilot (if overall readiness ≥ 50% and gaps are manageable), or
2. Pause and invest in foundation work (if readiness < 50%).

Practical tips for leaders

• Favor a minimal dataset for the first pilot—reduce integration and privacy complexity.
• Keep ownership clear. Assign a data steward and a pilot owner responsible for outcomes.
• Measure early: set simple success metrics for both model performance and business impact.
• Plan for iteration: treat the audit as living — repeat it after remediation and before scale.

Why convert this to an interactive audit?

Making this audit interactive lets teams save responses, track improvements over time, assign actions, and roll the assessment out across sites or business units. The platform supports rendering interactive forms and storing submissions so results can feed dashboards, track remediation status, and become a reusable domain-level toolkit for other teams.

Quick checklist (one-page handout)

• Cataloged datasets with owners — Yes/Partial/No
• Lineage documented — Yes/Partial/No
• Labels available / QA in place — Yes/Partial/No
• Access controls & stewards assigned — Yes/Partial/No
• Data freshness acceptable — Yes/Partial/No
• Privacy constraints reviewed — Yes/Partial/No
• Integration feasibility validated — Yes/Partial/No

Final note

This audit is designed to be practical and action-oriented: not a long compliance exercise, but a decision tool to help you determine whether to pilot, where to focus investments, and how quickly you can scale AI responsibly. If you want, transform it into an interactive form to collect team responses and track remediation—then package the collection as a reusable readiness toolkit for other groups.