AI adoption is an organisational question

Generative AI is often introduced as a technology decision. In practice, the more difficult questions concern the organisation around the technology. Do people have the skills, access and time to use it well? Is leadership visibly committed? Are the boundaries clear enough for responsible experimentation? Can staff see where it improves real work?

These questions shaped a study of generative AI adoption in human resources departments across UK higher education. The research tested a Responsible Generative AI Adoption Model, examining organisational readiness, governance maturity, perceived benefits, perceived risks, trust, current use, changing perceptions and adoption intentions.

The central finding was clear: readiness is the foundation of adoption. Governance and risk remain essential, but policies alone do not create routine, responsible use. Adoption becomes more likely when organisations combine practical enablement with credible oversight.

What the research examined

The study used a two-phase mixed-methods design. Three focus groups brought together nine HR professionals in executive and operational roles. A survey then gathered 122 complete responses across 51 UK higher education institutions.

The quantitative analysis tested relationships across eight adoption constructs using reliability assessment, correlation, regression and mediation analysis. The qualitative work examined the practical barriers, concerns and use cases described by HR professionals. Bringing both strands together made it possible to compare what people said was important with what most strongly predicted intention and use.

The results should be read in context. The research was cross-sectional, relied on self-reported data and focused on one national sector. It identifies useful relationships, not universal causal rules. Even so, the combination of qualitative and quantitative evidence offers a practical view of why responsible AI programmes often stall.

Readiness drives use

Organisational readiness was the only significant predictor of current GenAI use in the regression model. In this study, readiness included access to suitable tools and infrastructure, leadership support, training and the practical capacity to experiment.

This distinction matters. An organisation can approve an AI policy without making AI usable. If access is uneven, training is occasional and staff have no time to learn, formal permission will not translate into capability. Readiness turns strategic intent into the conditions required for action.

The focus groups reinforced this point. Time and capacity, skills and literacy, and policy clarity were recurring concerns. Participants wanted practical guidance, trusted colleagues who could help and opportunities to learn through real work rather than abstract instruction.

Perceived value shapes intention

Improved perceptions and perceived benefits were the strongest predictors of adoption intentions. Respondents were more likely to intend to adopt GenAI when they could see its relevance and when their view of the technology had shifted positively.

That suggests a simple management principle: adoption grows through credible evidence of value. Broad claims about transformation are less persuasive than a well-chosen workflow where the benefit is visible, the risk is controlled and the result can be evaluated.

Low-risk pilots are therefore more than technical tests. They are a way to build informed confidence. A documented improvement to an administrative process, accompanied by clear human review, gives teams evidence they can assess for themselves.

Trust is built through responsible practice

Trust did not directly predict adoption intentions in the final regression model, but it played an important mediating role between readiness and intention. This points to a more nuanced conclusion: trust is not simply a belief that leaders can request. It develops when people have the capability, support and experience to use AI responsibly.

In practical terms, trust is strengthened by competent use, visible safeguards and honest communication about limitations. It is weakened when tools appear without explanation or when staff are expected to reconcile ambitious messages with unclear rules.

This makes trust a relational bridge. Readiness creates the conditions for engagement; experience allows people to judge value and limitations; governance gives that activity legitimate boundaries.

Governance must be operational

Risk and governance were prominent in the qualitative findings, yet neither directly predicted adoption in the final quantitative models. That does not make them unimportant. It suggests that their contribution is enabling and protective rather than sufficient on its own.

Governance becomes useful when it answers practical questions. What information may be entered into a tool? Which outputs require specialist review? Who owns a decision? How should an issue be reported? What evidence must be retained?

Long policies that staff cannot apply create the appearance of control without consistent practice. Clear usage guidance, defined accountability, human review and a maintained risk register are more likely to influence day-to-day decisions.

A sequence for responsible adoption

The research proposed ten actions, sequenced from organisational readiness through trust-building to embedded governance and risk management:

  • Deliver targeted, recurring GenAI training.
  • Ensure fair access to approved tools.
  • Protect time for learning and experimentation.
  • Establish volunteer champions within teams.
  • Run low-risk pilots in real workflows.
  • Recognise responsible use and share credible results.
  • Publish simple, accessible usage guidance.
  • Establish cross-functional governance across business, technology and legal roles.
  • Require human review of AI outputs.
  • Maintain a GenAI risk register with clear ownership and review cycles.

The sequence is important. Organisations need not wait for perfect maturity before learning, but experimentation should occur within clear boundaries. Training, access and protected time make responsible participation possible. Pilots and peer support then build evidence and confidence. Governance and risk practices turn those lessons into repeatable operating standards.

Four questions for leaders

Before adding another tool or policy, leaders can test the quality of their adoption environment with four questions:

  • Readiness: Do people have the access, skills, leadership support and time required to use AI well?
  • Value: Are priority use cases tied to a clear operational problem and a measurable outcome?
  • Trust: Can staff see how outputs are checked, limitations are communicated and good practice is supported?
  • Governance: Are responsibilities and boundaries clear enough to guide an actual decision at the point of use?

Weakness in any one area does not mean the programme should stop. It identifies where management attention is required. The objective is not rapid deployment for its own sake. It is a coherent system in which capability, value, trust and control reinforce one another.

The management task behind the technology

The wider implication extends beyond higher education HR. In any high-accountability environment, responsible AI adoption is likely to depend less on access to a model than on the quality of the surrounding management system. That proposition should be tested in each organisational context, but it offers a useful starting point.

Technology can expand what is possible. Readiness determines whether people can use it. Trust determines whether they will engage with it. Governance and risk management determine whether that use can be sustained responsibly.

Responsible adoption begins when these elements are treated as one organisational challenge.