Welsh Government's AI trial: How trade union partnership shaped the introduction of AI tools across government

The problem

Governments across the world are exploring how AI tools can help their staff work more efficiently. But introducing AI into a workplace raises questions that go beyond the technology itself. Which staff get access first? How will the organisation know whether the tools are working? What happens to people whose roles change as a result? How are concerns about fairness, privacy, and inclusion addressed? And who has a say in how these decisions are made?

These questions are not hypothetical. When a government introduces a new technology that affects how thousands of staff members do their jobs, the process of introduction matters as much as the technology itself. If staff feel the tool has been imposed without consultation, or that decisions about its use were made without their input, trust is damaged — and trust is difficult to rebuild. If certain groups of staff are excluded from early access, or if the tool does not work well for people with different needs, the rollout risks deepening existing inequalities rather than reducing them.

The Welsh Government faced these questions when it decided to trial Microsoft 365 Copilot, an AI assistant built into the productivity tools its staff already use. The question was not just whether Copilot could improve productivity, but how to introduce it in a way that was transparent, inclusive, and consistent with the Welsh Government's legal obligations under the Social Partnership and Public Procurement (Wales) Act — legislation that requires public bodies in Wales to work in partnership with trade unions when making decisions that affect the workforce.

The solution

The Welsh Government's Digital, Data and Technology directorate launched its trial of Microsoft 365 Copilot in September 2023. What made the approach distinctive was not the technology- Copilot is a commercially available product- but the process the government built around it, grounded in partnership with trade unions from the outset.

Trade unions were involved before the trial began
Trade union representatives were consulted and given access to Copilot ahead of the wider trial. They were among the first cohort of users, meaning they could experience the tool themselves rather than relying on secondhand accounts of what it did. They were also given access to a dedicated Microsoft Teams channel, which provided up to date visibility of user feedback, common questions, training materials, and technical updates. This meant union representatives could see how the tool was being used across the organisation and raise concerns as they arose, rather than waiting for a formal review.

Inclusion was built into how licences were allocated
The Early Access Program deployed 300 Copilot licences. Alongside trade union representatives, licences were specifically provided to neurodivergent staff and users with workplace adjustments. This was a deliberate decision to ensure that accessibility was accounted for from the start.

When the trial scaled up between September and December 2024, with nearly 1,500 staff joining the UK's largest cross-government Copilot trial, trade unions were involved in licence allocation. This meant decisions about who received access were made transparently and with union input.

Continuous feedback is shared openly
Monthly surveys and feedback sessions created an ongoing pipeline of information throughout the trial. Trade union representatives received access to survey data, which allowed them to participate in interpreting the findings and shaping what happened next.

The Chief Digital Officer and Deputy Directors met with union representatives to ensure they were updated on plans. Trade unions representatives were also consulted in the development of an Equality Impact Assessment for the trial, with user feedback incorporated into the final analysis.

Alignment with legislation
The approach was designed to align with the Social Partnership and Public Procurement (Wales) Act, which places a legal duty on Welsh public bodies to seek consensus with trade unions and to promote fair work. The trial treated this not as a compliance requirement but as a framework that shaped how every stage of the rollout was managed.

The impact

1. Users reported gains in productivity and job satisfaction
Staff who participated in the trial reported improvements in both productivity and satisfaction with their work. The Welsh Government describes the results as “striking”. The evidence is drawn from monthly surveys conducted throughout the trial period.

2. Nearly 1,500 staff participated in the scaled-up trial
The trial grew from 300 licences in the Early Access Program to nearly 1,500 participants when the Welsh Government joined the UK's largest cross-government Copilot trial. This scale means the findings reflect a broad range of roles, teams, and working patterns across the organisation.

3. Trade unions endorsed the process as a model for future digital projects
Trade unions highlighted the transparent, inclusive approach and described it as a model for managing future technology projects. For an organisation introducing AI tools to its workforce, securing the representation of staff is a meaningful outcome, indicating that the process was perceived as fair by the people whose working lives it affected.

4. Neurodivergent users and staff requiring workplace adjustments were included from the start
Reserving licences specifically for these groups in the first cohort of users meant the trial generated early insight into how the tool worked for people with different needs.

Lessons learned

• The process of introducing AI mattered as much as the technology itself. The Welsh Government trialled a commercially available product that many organisations are adopting. What distinguished this project was not what the tool did but how it was introduced. Involving trade unions before the trial started, giving them access to the tool, sharing feedback data openly, and including them in decisions about licence allocation built trust in a way that a top-down rollout may not have achieved.
• Giving trade unions real access and real data changed the nature of the conversation. Union representatives were not briefed after decisions had been made: they experienced the tool, saw the feedback, and participated in interpreting the results. This meant discussions about the trial were grounded in shared evidence rather than competing assumptions.
• Reserving early access for staff with the greatest needs was a specific, practical inclusion measure. The Welsh Government deliberately included neurodivergent staff and those requiring workplace adjustments in the first cohort. This generated early evidence of how the tool worked for users with different needs.

This case study was written with assistance from artificial intelligence.