AI readiness sounds vague until you test the data properly.
Is the data actually ready?
Most organisations now have a growing list of AI ambitions including, better reporting, faster service delivery, knowledge assistants, automated compliance checks, predictive analytics and AI-supported decision-making.
But none of that works reliably unless the underlying data can support it.
A FAIR Data Assessment turns AI readiness from a broad ambition into an evidence-based test. Instead of asking whether the organisation is ‘ready for AI’, it tests whether priority datasets are sufficiently Findable, Accessible, Interoperable and Reusable to support AI, reporting and operational decisions.
What a FAIR Data Assessment is really testing
The FAIR principles were designed to improve how digital assets are structured for reuse, with a strong emphasis on machine-actionability.
That matters because AI systems depend on data they can:
- Locate
- Interpret
- Connect
- Reuse
with minimal ambiguity.
A FAIR Data Assessment applies that standard to real datasets, not theory. It answers a simple question – can this data actually support the AI or reporting outcome we are planning?
Read: What are FAIR data principles and why do they matter for AI?
Why a FAIR Data Assessment is different from a normal data review
A typical data review focuses on quality – accuracy, completeness, consistency and duplication.
That matters, but it is not enough.
AI depends on more than clean data. It depends on whether data can be:
- Found
- Accessed safely
- Connected across systems
- Reused for new purposes
This is what makes a FAIR Data Assessment more than a technical audit. It is an investment control.
For CIOs, CTOs and CDOs, it should answer four commercial questions:
- Which AI use cases can move forward?
- Which datasets need remediation first?
- Which risks require governance or compliance input?
- Where are business cases built on weak data foundations?
Catapult CX’s FAIR Data Assessment is designed around these questions, assessing data against FAIR principles and AI readiness criteria, then producing a prioritised view of gaps, risks and next steps.
The assessment framework. From discovery to roadmap
A FAIR Data Assessment should not attempt to audit the entire estate.
The focus should be on priority datasets linked to real use cases.
A practical assessment includes four stages:
- Inventory and use case alignment
Identify datasets required for a specific AI, analytics or reporting objective - Evidence gathering
Review real systems, metadata, access controls, documentation and data samples - FAIR stress test
Assess datasets against Findability, Accessibility, Interoperability and Reusability - Remediation roadmap
Define what needs fixing, in what order, to unlock the use case
The point is not to produce a maturity score. It is to give leaders a clear go, fix or pause decision before funding delivery.
What each FAIR area should examine
A strong assessment turns each FAIR principle into an evidence-based test.
Assessment areaWhat the assessment should examineEvidence to requestLeadership decision
| Assessment area | What the assessment should examine | Evidence to request | Leadership decision |
|---|---|---|---|
| Findable | Ability for people and systems to locate datasets quickly | Metadata, catalogues, ownership, naming standards | Can teams and systems reliably find the right data? |
| Accessible | Ability for authorised users and systems to access data safely and lawfully | Permissions, APIs, authentication, sharing rules | Can data be retrieved securely at scale? |
| Interoperable | Ability to use data across systems and workflows | Formats, identifiers, schemas, integration routes | Can data connect without manual effort? |
| Reusable | Ability to reuse data for AI, analytics and reporting | Lineage, documentation, lawful basis, controls | Can this data be trusted to inform decisions? |
Findable. Can the organisation locate the right data?
In many organisations, valuable data exists but cannot be reliably found.
The assessment should test:
- metadata quality
- ownership clarity
- catalogue coverage
- searchability
Weak findability means:
- AI systems cannot ground outputs
- reporting relies on informal knowledge
- governance cannot prove data sources
The output should highlight datasets that are hard to locate or poorly defined.
Accessible. Can authorised users and systems access the data safely?
Accessible does not mean open. It means controlled and usable.
This should be tested through:
- permissions and role-based access
- authentication and APIs
- data sharing agreements
- approval workflows
AI projects often fail here. A model may be viable, but if data cannot be accessed lawfully and at scale, delivery stops.
The output should show:
- what is accessible now
- what requires approval
- what cannot be used safely
Interoperable. Can the data work across systems?
Interoperability can be the barrier between isolated datasets and usable intelligence.
Data may exist and be accessible, but inconsistent in terms of:
- formats
- definitions
- identifiers
- schemas
meaning it cannot be used together. For AI, this is critical.
If datasets cannot be joined consistently, systems will produce partial or misleading outputs.
The assessment should identify where interoperability gaps create delivery risk or manual workarounds.
Reusable. Can the data support future decisions?
Data may work for its original purpose but still fail for AI.
The assessment should examine:
- provenance and lineage
- documentation and context
- lawful basis and usage restrictions
- quality controls
If data informs decisions, the organisation must be able to defend:
- where it came from
- how it was used
- whether it is reliable
This creates a defensible data-to-decision chain.
What evidence should leaders expect?
Interviews capture perception. Leaders need evidence.
A credible assessment should include:
- priority datasets reviewed
- ownership and stewardship clarity
- metadata quality findings
- access and permission constraints
- integration and format issues
- known data quality risks
- legal and compliance gaps
- legacy system constraints
- dependencies between remediation tasks
- impact on AI or reporting use cases
This is where the assessment becomes commercially useful. It turns assumptions into a structured remediation plan.
What should the final outputs include?
A FAIR Data Assessment should produce decision-ready assets, not observations.
At a minimum:
- FAIR Data Readiness Scorecard
Where data is AI-ready and where it is blocked - Gap Register
Structured list of technical, governance and data issues - Prioritised Remediation Roadmap
What to fix first, based on value and risk - Compliance and DPIA Gap Register
Legal and privacy risks affecting AI use - Executive Summary
Board-level view of risks and next steps - Actionable next steps
Clear path into remediation or delivery
What this means for regulated organisations
Regulated organisations do not just need better data. They need data that can be trusted, traced and defended.
A FAIR Data Assessment highlights where:
- ownership is unclear
- lineage is missing
- access is weak
- governance is incomplete
These are not minor issues. They are delivery blockers.
Financial services
For financial services firms, the key issue is defensibility.
AI use cases such as:
- KYC/AML
- compliance automation
- complaints analysis
- knowledge retrieval
depend on data that can be audited and traced.
A FAIR Data Assessment shows whether datasets:
- have clear ownership
- can be traced to source
- are governed correctly
- can support regulated decisions
Government and public sector
For public sector organisations, the main challenge is fragmentation.
Data is often:
- spread across legacy systems
- held in silos
- poorly documented
This makes findability and interoperability major blockers.
A FAIR Data Assessment provides:
- evidence for internal assurance
- support for governance and spend controls
- visibility of risks linked to AI deployment
Start with the data, not the model
AI readiness is not proven by buying tools or running pilots.
It is proven when data can be:
- found
- accessed safely
- connected across systems
- reused with confidence
That is what a FAIR Data Assessment tests.
If you are planning AI investment without testing the data first, you are making decisions without the full picture.
Catapult CX’s FAIR Data Assessment gives CIOs, CTOs and CDOs a clear view of which datasets are ready, where the risks sit, and what needs to change before committing to delivery.
FAQs. FAIR Data Assessments and AI readiness
What does a FAIR Data Assessment actually test?
It tests whether priority datasets are sufficiently findable, accessible, interoperable and reusable enough to support a specific AI, reporting or operational use case.
How is a FAIR Data Assessment different from a data audit?
A data audit focuses on quality and compliance across the estate. A FAIR Data Assessment focuses on specific datasets and tests whether they can support a defined outcome, producing a remediation roadmap.
How long does a FAIR Data Assessment take?
A focused assessment covering three to five datasets typically takes two to four weeks, depending on system complexity and data access.
Who should commission a FAIR Data Assessment?
Typically CIOs, CTOs, CDOs or transformation leaders preparing to fund AI or data-driven programmes.
What evidence should an organisation expect?
Evidence includes dataset reviews, ownership clarity, metadata quality, access constraints, integration issues and compliance gaps.
What happens after a FAIR Data Assessment?
The assessment produces a prioritised remediation roadmap. Organisations typically move into targeted fixes, governance improvements or phased AI delivery.
Does a FAIR Data Assessment cover compliance and privacy?
Yes. It identifies legal, privacy and governance gaps, including issues related to data protection, lawful use and regulatory requirements.
Can a FAIR Data Assessment support AI business cases?
Yes. It shows which datasets can support delivery, where risks exist and what investment is required before scaling AI.
Where should a FAIR Data Assessment start?
It should start with a small number of priority datasets linked to real AI or reporting use cases. Assessing the entire data estate rarely leads to action, whereas focused assessments produce clear decisions and faster progress.
