Building an AI Data Readiness Scorecard

A readiness score is only useful if it is repeatable, defensible, and tied to concrete engineering work. This is the assessment we run — five steps that produce a composite score, a KPI matrix with hard thresholds, and a remediation backlog a CIO can fund.

The five-step assessment

Step	What happens	Output
1 · Inventory	Catalog data domains, sources, AI use cases, and owners	Federated estate map
2 · Score 5 dimensions	Rate Context, Clarity, Coverage, Credibility, Capacity (0–10) per domain	Dimension scores
3 · Find binding constraint	Composite = min across dimensions; the lowest blocks AI value	Binding dimension
4 · Set targets & gates	Define KPI thresholds, HITL escalation rules, SLA budgets per gate	Target model
5 · Track & re-score	Monthly composite, heat-map, blocker backlog; re-baseline quarterly	Readiness dashboard

The output of Step 1 is not a spreadsheet of table names — it is a map that already carries ownership and use-case context, so every later score is attributable to a domain and a person.

The scorecard matrix — banking KPIs and thresholds

Scores are not opinions. Each dimension has a measurable KPI and a target threshold. For capital markets and banking:

Dimension	KPI	Target threshold	Where it's measured
Context	Tables actively mapped to FIBO classes (e.g. FxFwdContract, DepositAccount)	≥ 90%	Oracle on-prem, ADLS Gen2
Clarity	BCBS 239 lineage; deterministic explanation trail	100% audit-ready	SQL Server ↔ Databricks joins
Coverage	Unstructured assets (PDFs, 10-Ks, agreements) parsed into metadata-enriched vector space	≥ 85% parsed	AWS S3, document archives
Credibility	Entity duplication rate; real-time ingestion lag	< 1% dup; sub-second lag	Oracle cores, Databricks pipelines
Capacity	Consent-signal propagation; PII / Do-Not-Train enforcement	< 10 ms check; 100% enforced	Full federated estate

These thresholds are deliberately hard numbers. "Better lineage" is not a target; 100% audit-ready traceability across SQL Server ↔ Databricks joins is. A target you cannot fail is not a target.

The metrics tracked every cycle

Beyond the dimension scores, four operational KPIs make the program measurable over time:

• Composite readiness score — the 0–100 weighted minimum of the five dimensions.
• Time-to-production — days from idea to a deployed AI agent.
• Adoption rate — % of AI tools in active use vs. shadow IT.
• Security / compliance incidents — prompt-injection, PII leak, audit failures — tracked as a prevention rate.
• Cost per AI outcome — dollars per resolved query, report, or decision.

The executive dashboard

The scorecard rolls up into four questions a CIO and CDO can answer on one screen:

1. How ready are we now? — composite readiness gauge with a 30/90-day trend.
2. Where is the risk? — a heat-map by domain × source (Oracle, SQL Server, Databricks, docs, streams).
3. What blocks scale? — a control-gap panel: missing lineage, legal basis, ownership.
4. What does it cost? — a remediation backlog ranked by risk reduction per engineering week.

That last ranking is what turns a score into a plan. The backlog is sorted so the highest risk-reduction-per-week item — almost always the binding constraint — is at the top.

From score to funded roadmap

A score without a roadmap is theater. Each low dimension maps to a named workstream and a specific accelerator:

Context   low → FIBO ontology mapping         → Schema-Miner Pro
Clarity   low → lineage + metadata            → LineageLens
Coverage  low → unstructured + federation     → Lakehouse Federation
Credibility low → entity resolution + CDC     → golden-record service
Capacity  low → consent & PII guardrails      → GovPilot

ReadinessIQ runs the full five-step assessment and generates the prioritized acceleration roadmap in days, not quarters. The goal is never a perfect score — it is the shortest defensible path from where you are to production AI.

Request a ReadinessIQ assessment →