Every AI engagement starts with a data conversation. And in almost every case, that conversation reveals a gap between what the organisation thinks their data readiness level is and where it actually is.
This is not a criticism. Data infrastructure debt is normal — it accumulates invisibly over years of prioritising feature delivery over data quality. The problem is not the debt itself. The problem is underestimating how much debt exists and how long it takes to address.
The Data Readiness Framework
We assess data readiness across four dimensions. Each dimension has three levels: Basic, Intermediate, and Advanced.
**1. Accessibility.** Can the AI system reach the data it needs?
- Basic: Data exists in silos, accessible only through export queries or manual extracts.
- Intermediate: Key data sources have APIs, but integration requires custom engineering per source.
- Advanced: Data is available through a unified access layer (data mesh, data lakehouse, or consistent API gateway) with governed access controls.
**2. Quality.** Can the AI system trust the data?
- Basic: No systematic quality monitoring. Data quality is discovered during incidents.
- Intermediate: Basic quality checks exist (null checks, range validation) but coverage is incomplete.
- Advanced: Automated quality monitoring with drift detection, lineage tracking, and alerting on quality degradation.
**3. Documentation.** Can the AI system understand the data?
- Basic: Minimal documentation. Column names are the only description.
- Intermediate: Data dictionary exists but is maintained manually and often out of date.
- Advanced: Active metadata management with automated schema documentation, business glossary, and usage patterns.
**4. Governance.** Can the AI system handle sensitive data appropriately?
- Basic: No formal data classification. Access is broad.
- Intermediate: Data is classified (PII, confidential, public) but enforcement relies on policy rather than technical controls.
- Advanced: Automated data classification, column-level access control, and auditable data usage trails.
The Reality Check
In our engagements, roughly 70% of organisations are at Basic or Intermediate across at least three of these four dimensions. That does not mean they cannot use AI — but it does mean their first AI initiative should include a parallel track of data infrastructure investment.
The most expensive mistake we see is an organisation that starts building AI on top of fragile data infrastructure, then needs to rebuild both when the infrastructure fails under production load.
The Practical Path
The good news: you don't need to reach Advanced on all four dimensions to start delivering AI value. The approach we recommend is:
1. **Pick one high-value use case** where the data is relatively clean and accessible.
2. **Build the data infrastructure for that use case only** — not an enterprise-wide data platform.
3. **Deliver value quickly** (within 8-12 weeks) to build momentum and justify further investment.
4. **Expand incrementally** — each subsequent use case improves the data infrastructure further.
This pragmatic approach avoids the two failure modes that plague most AI initiatives: "build the perfect data platform first" (which never finishes) and "build AI on top of bad data" (which never delivers).
Your data does not need to be perfect for AI to succeed. But it does need to be understood, accessible, and improving. Start measuring your readiness today, and the path forward becomes clear.