Chapter 8. Data, Infrastructure, Skills, and Organisational Readiness

8.1 Data quality is foundational

AI outputs depend heavily on the information available to the system. Useful data should be relevant, accurate enough for the task, current, representative, lawfully obtained, consistently defined, and appropriately protected. Poor data can produce polished but misleading results.

Before using data, establish ownership, purpose, permitted use, retention, access, provenance, quality checks, and correction procedures. Keep a record of transformations. For high-impact uses, assess whether particular groups or cases are missing or systematically misrepresented.

8.2 Infrastructure and integration

AI adoption may require reliable connectivity, identity and access management, secure endpoints, storage, monitoring, integration with existing applications, backup and recovery, and support. Some workloads run through cloud services; others may use local or private infrastructure. The best choice depends on sensitivity, latency, scale, cost, skills, and control requirements.

An organisation should understand where data are processed, whether they are used for provider training, how long they are retained, what subcontractors are involved, and how access is controlled. Product terms and settings vary, so assumptions should be verified before use.

8.3 Skills people need

AI literacy includes understanding basic concepts, recognising limitations, prompting clearly, checking sources, protecting information, and identifying when to escalate. Technical teams may also need data engineering, model evaluation, cybersecurity, software integration, and monitoring skills. Leaders need procurement, governance, change management, and value measurement.

Not everyone needs to become a machine-learning engineer. Most organisations need a mix of domain experts, informed users, technical specialists, risk and legal advisers, and accountable decision-makers. Training should be role-specific and linked to real workflows.

8.4 Organisational readiness assessment

Assess readiness across strategy, leadership, people, data, technology, process, risk, and measurement. A company with a clear business problem but poor data quality may need to improve its records before adopting AI. An institution with capable staff but no policy may need governance before allowing broad use.

Readiness is not a one-time score. It should be reviewed as systems, staff, regulation, and business needs change. Start small, learn from evidence, and scale only when the controls and benefits are demonstrated.

8.5 Build, buy, or partner

Buying an established service may be faster and easier than building a model. Building offers greater control but requires data, expertise, infrastructure, maintenance, and ongoing evaluation. Partnering can provide specialist capability, but responsibilities and data access must be clear.

Compare total cost of ownership rather than licence price alone. Include integration, training, oversight, testing, support, security, change management, and exit costs. Procurement should ask for evidence, not rely solely on marketing claims.