Framework Dimension 2 of 6

Data & Infrastructure

Data quality, pipelines, vector stores, and the platform readiness needed for production AI.

The Data & Infrastructure dimension measures whether your data is AI-ready — governed, catalogued, and reachable by production inference systems — or whether every use case starts with a six-month data-plumbing project.

Framework v2026.1 · Updated · machine-readable spec

Why this dimension matters

The most common failure mode in enterprise GenAI is not model quality but data access latency. Teams that ship data platform investments alongside model investments reach level 3 substantially faster than teams that treat data as a downstream concern.

Signals we look for

  • Data catalogue with lineage
  • Vector store & embeddings pipeline in production
  • Feature store or equivalent shared substrate

What this dimension looks like at each maturity level

  1. 1Aware

    Data is siloed. No shared catalogue. Access is ad-hoc and slow.

  2. 2Exploring

    A data catalogue exists but is not universally adopted. First embeddings pipeline in a pilot.

  3. 3Operational

    Governed data pipelines feed at least one production AI system. Vector store is a shared internal service.

  4. 4Integrated

    Feature store (or equivalent shared substrate) used by multiple teams. Data quality SLAs enforced.

  5. 5Autonomous

    Pipelines self-monitor for drift and freshness. Data platform team treats AI teams as first-class internal customers.

  6. 6Transformative

    Text, image, audio, and telemetry treated as first-class inputs across the data platform. Real-time embeddings on write.

Common blockers

  • Vector databases procured but not integrated into the data governance regime.
  • PII redaction handled per-app rather than at platform layer.
  • Data lineage tracked for reporting but not for AI training corpora.

How to move up a level

  • Publish a data catalogue with clear ownership per dataset.
  • Stand up a vector store as a shared internal service (not per-team infrastructure).
  • Add lineage tracking that specifically covers training-corpus provenance.
  • Move PII redaction from application code to a shared middleware.

Related on this site

Next steps