PROCLOUDLABS

Best practices · Self-check

Is your data ready for AI?

The 40 questions we use to judge AI readiness: data, governance, use cases, infrastructure, security, tooling, people and policy.

Self-check

How to use it

AI is only as good as the structure beneath it. These are the 40 questions we use to judge whether a company's data, systems and team can support AI, grouped into eight dimensions. Tick each one you could show evidence for today.

A dimension with fewer than three ticks is usually a blocker: fix it before you fund a use case that depends on it.

01

Data foundation and quality

  • The data sources your use cases need are documented and accessible
  • A unified view of your core customer, account or product exists across systems
  • Data quality is measured: completeness, freshness and lineage
  • Duplicate and conflicting records are measured and can be fixed
  • Labelled or structured data exists to build and evaluate each use case

02

Data governance and privacy

  • Data owners and stewards are named for the systems in scope
  • Personal and sensitive data is classified
  • A policy says what data may train, fine-tune or feed a model
  • Retention, residency and consent for AI use are documented
  • Data lineage can be traced from source to model

03

Use-case value

  • A backlog of candidate use cases is written down
  • Each use case ties to a measurable outcome and a value estimate
  • Feasibility has been tested against the data you actually have
  • Each use case truly needs AI, rather than rules or reporting
  • A shortlist of three to five use cases is agreed

04

Integration and infrastructure

  • There's a clear way for AI to read and write your data, through APIs or events
  • Environments to build and run AI are available
  • Latency and scale needs are understood
  • The basics of running models are in place: versioning, deployment, monitoring
  • A target integration pattern is picked for each use case

05

Security and model risk

  • Access to data and models is controlled
  • Secrets and personal data can't leak into prompts and tools
  • Prompt-injection and output risks have been assessed
  • Model and vendor risk has been reviewed
  • Unsanctioned AI use is identified and contained

06

Tooling and vendor strategy

  • You've decided when to build, buy or fine-tune
  • Platforms and models are shortlisted against your requirements
  • Running costs are modelled
  • Lock-in and exit risk are assessed
  • You could switch models or vendors if you had to

07

Skills and operating model

  • Someone clearly owns AI
  • The skills are in place, or there's a hiring or partner plan
  • Someone is responsible for maintaining models after launch
  • There's a plan for change management and adoption
  • Budget to build and run is secured

08

Responsible AI and policy

  • An acceptable-use policy for AI exists and people know it
  • A human stays in the loop where decisions carry risk
  • Quality and bias are evaluated and monitored
  • There are clear explanation and escalation paths
  • Regulatory exposure is assessed and tracked