The Five Dimensions of AI Data Readiness

Most AI programs are measured on the wrong thing. Teams track model accuracy and demo velocity while the real blocker sits upstream, in the data estate. 88% of enterprise AI pilots never reach production, and roughly 60% of AI initiatives are abandoned outright — almost always because the data feeding them is fragmented, ungoverned, or semantically opaque.

Readiness is not a vibe. It is a measurable property of your data estate across five dimensions, and one number — the lowest — decides how far your AI can go.

The five dimensions

We score an estate on five dimensions, each on a 0–20 scale (0–10 per domain, rolled up):

Dimension	Question it answers	Core banking metric
Context	Do systems carry business meaning, not just table names?	% of tables actively mapped to standard FIBO classes
Clarity	Can every automated decision produce an explanation trail?	BCBS 239 lineage completeness; deterministic explainability
Coverage	Is unstructured + real-time data in scope, not just tidy tables?	% of legacy unstructured assets parsed into vector space
Credibility	Is the data trusted, fresh, and de-duplicated?	Entity duplication rate; real-time pipeline latency
Capacity	Are governance and consent enforceable at query time?	PII / "Do-Not-Train" enforcement; permission-check latency

Context and Clarity are about meaning and traceability. Coverage is about scope. Credibility is about trust. Capacity is about control. An estate can be strong on one and fatal on another.

Why the lowest score is the only score that matters

The composite readiness score is not an average. It is bounded by the weakest dimension:

type Dimension = "context" | "clarity" | "coverage" | "credibility" | "capacity";

// R is the headline number; B is the dimension that caps ROI.
function readiness(scores: Record<Dimension, number>) {
  const values = Object.values(scores);
  const R = values.reduce((a, b) => a + b, 0); // 0–100 composite
  const B = Math.min(...values);               // binding constraint
  const binding = (Object.keys(scores) as Dimension[])
    .find((k) => scores[k] === B);
  return { R, binding };
}

readiness({
  context: 12, clarity: 9, coverage: 11, credibility: 7, capacity: 8,
});
// → { R: 47, binding: "credibility" }

An estate that scores 12 on Context but 7 on Credibility is a 7-readiness estate. Pouring engineering into better ontologies while duplicate counterparties and stale pipelines go unaddressed raises the average and moves nothing that matters. The binding dimension — here, Credibility — is where the next dollar belongs.

This is the single most important idea in the model: fix the binding constraint first, not the most visible or most fashionable gap.

An example, end to end

A relationship-manager copilot needs to answer "What is our total exposure to ABC Corp, and any breaking-news risk today?" That single question touches all five dimensions:

• Context — the agent must know that oracle.crm.acct and sqlsrv.loan both describe a FIBO LegalEntity, and that "exposure" means deposits + credit.
• Coverage — breaking-news risk lives in call transcripts, complaint letters, and a Reuters feed, not a table.
• Credibility — "ABC Corp" must resolve to one golden entity, not three near-duplicate profiles across silos.
• Clarity — the answer needs a citation trail regulators can audit.
• Capacity — any masked PII or "Do-Not-Train" record must be excluded at query time, not scrubbed later.

Miss any one and the copilot is either wrong, unexplainable, or non-compliant. That is why readiness is a minimum, not an average.

The path to implement

You do not need a multi-year platform rewrite to move these numbers. The sequence that works:

1. Inventory the estate — domains, sources, AI use cases, and owners — and produce a federated estate map.
2. Score all five dimensions per domain, then compute the composite and identify the binding constraint.
3. Target the binding dimension with a specific accelerator rather than a general "data quality" program.
4. Re-score monthly so capability investment is tied to a measurable delta, not a slide.

Each dimension maps to a purpose-built accelerator. Incipient's ReadinessIQ runs the scoring and produces the prioritized roadmap; GovPilot and the rest of the Lab close the specific gaps the score exposes. The next article walks through the scorecard and KPI thresholds in detail.

Readiness is the shortest path from siloed POCs to production-grade agentic AI — and it starts by finding the one dimension holding everything else back.

Score your estate with ReadinessIQ →