Agenda item

Peter Davies introduced the report and answered the members’ questions withLeader Mary-Ann Brocklesby, Paul Jefford and Matt Gatehouse:

 

Is the council being sufficiently realistic and ambitious in its approach to AI, given that the report proposed a three-year period to ‘build the foundations’, while many organisations are moving more rapidly. Could an emphasis on caution result in an approach that appears slow or unambitious, particularly if, by the second year, continuing to focus on foundations no longer reflects the pace of change elsewhere?

 

The programme is intentionally front-loaded, with significant investment and activity taking place early in the three-year period. He explained that trusted external partners would be used initially to accelerate delivery and establish capability, after which the council and the Shared Resource Service would increasingly bring that capability in-house. This approach is designed to balance pace with control, enabling faster progress in later years once the necessary foundations and skills are embedded.

 

The council cannot afford to take no action but must also avoid fragmented or poorly governed adoption of AI. The purpose of the programme is to establish a strong ‘bedrock’ that ensures safeguards, governance, and alignment with public service values, rather than simply proceeding quickly. Unregulated adoption of AI can lead to unintended negative consequences, and that the council’s approach is intended to ensure human-centred decision-making is maintained.

 

The approach is not about delaying progress, but about enabling early and tangible benefits within a controlled framework, citing examples such as developing a ‘single view’ of a child, homelessness, or debt to support early intervention and prevention. The approach seeks to combine pace with responsibility, ensuring improvements in outcomes for residents while maintaining the council’s core principles.

 

What are we doing about prioritising and selecting from the 160 AI use cases?

The initial focus is on a small number of high?priority, shared use cases agreed across the four authorities. The ‘AI front door’ (customer-facing interactions) and ‘single view of a child’ are already identified as priority workstreams for year one. In addition, approximately three further use cases per authority will be selected for early implementation, based on collective agreement. A collaborative approach is being used so that one authority can lead on a use case, and others can adopt it, allowing faster scaling and shared benefit despite limited capacity within the Shared Resource Service.

 

How are we balancing customer-facing improvements with internal organisational benefits?

 

Both strands are being developed in parallel. The customer-facing element focuses on improving access to services, particularly through online and telephony interactions such as chatbots handling common enquiries (e.g. waste collection queries), which helps free up staff time. The internal element focuses on productivity tools, such as AI supporting document drafting, translation, redaction for information requests, and internal knowledge access (e.g. HR policy queries). This dual approach is intended to reduce administrative burden internally while improving responsiveness and access externally.

 

What do we mean by AI bias, and how are we addressing it?

 

AI bias refers to the risk of inaccurate, misleading, or unfair outputs from AI systems, particularly where they rely on large public data models. This is being mitigated by using controlled and approved systems, including ‘closed’ models that rely on internally verified council data, such as policies and procedures. Safeguards, testing, and monitoring will be built into the implementation process, and data used by the AI will be drawn from sources that have already been subject to organisational validation and review.

 

How will we ensure robust evaluation given the scale of investment?

 

Evaluation is built into the programme as an ongoing and structured process. There will be continuous monitoring and feedback as systems are developed and used, alongside formal review points such as stage gates and governance oversight through the Shared Resource Service board. This includes regular scrutiny, impact assessments, and benefit tracking to ensure that the programme is delivering improved services, building staff capability, and achieving efficiency gains. Evaluation operates both at an operational level (continuous improvement) and at strategic checkpoints (formal governance reviews).

 

How are we addressing the environmental impact of AI, particularly increased energy use?

 

The approach is to mitigate environmental impact through technology choices and responsible usage. The move to cloud infrastructure (Amazon Web Services) allows more efficient use of computing resources, including scaling usage up or down and avoiding underutilised hardware. AWS has commitments to reach net zero carbon and invests in renewable energy sources. Additionally, there is recognition within the council of the need for responsible use of AI, supported by policy and governance, to manage demand and reduce unnecessary energy consumption associated with AI tools.

 

What is the approach to reserve funding, and what does this mean for residents?

 

The funding is being drawn from the ‘invest to redesign’ reserve, which is specifically intended for transformation initiatives rather than routine IT spending. A total of £851,000 is proposed as a one-off investment, including around £200,000 to support service capacity where needed. This is set within the context of wider usable reserves of around £20 million. The purpose of this investment is to enable long-term service improvement, efficiency, and better outcomes for residents, rather than ongoing expenditure.

 

Concerning the energy implications of AI, even simple AI uses can be highly energy intensive; for example, generating a single AI image can require around 2 kilowatt hours of electricity, roughly equivalent to fully charging a smartphone. Organisational systems may default users into AI tools, which could unintentionally drive demand – should the council consider controls over AI use to manage energy consumption responsibly?

 

While there are legitimate concerns about energy use, both technology providers and the council have roles in mitigating this impact. Providers such as Amazon are already developing more efficient infrastructure and technologies to reduce the carbon footprint of AI, including improvements in computing hardware and energy use. From the council’s perspective, we would emphasise the importance of responsible usage, stating that this will be addressed through an AI policy and governance controls. This may include both guidance to users and technical safeguards (working with the Shared Resource Service) to prevent excessive or inappropriate use of high-energy AI functions.

What assurances are in place to protect confidentiality while creating a ‘single view of a child,’ and how is the tension with openness and transparency addressed?

 

Confidentiality is maintained through strict role-based access controls, meaning that staff can only access the data they are authorised to see, in line with current permissions. The system mirrors existing access rights but brings data together into a single interface, reducing the need to access multiple systems. The ‘AI foundations’ include these guardrails as a core element, ensuring that only appropriate users can view specific information.

 

The reference to openness and transparency does not mean unrestricted access; rather, it refers to making relevant information more readily available to those who legitimately need it in order to make informed decisions. Transparency operates within the confines of confidentiality, not in conflict with it. The approach ensures that while data is more usable and accessible to practitioners, it is not made broadly available beyond authorised users.

 

What consideration is being given to the potential for locating data centre infrastructure locally, for example at Sudbrook, given existing water resources, or is this unrealistic?

 

The current strategy does not involve building local authority-owned data centres, as this has already evolved towards more efficient and cost-effective cloud-based infrastructure. Moving to cloud services allows the council to make more efficient use of shared resources, avoiding underutilised hardware and reducing overall energy consumption compared to maintaining standalone infrastructure. However, the possibility of alternative or local infrastructure has not been entirely dismissed. The point about Sudbrook is recognised as relevant and worth further consideration, particularly in the context of wider regional partnerships such as the Cardiff Capital Region. Existing work on energy and infrastructure opportunities, including at Sudbrook, has been undertaken in the past, and the suggestion will be taken away for further exploration.

 

The overall approach remains focused on partnership working and collaboration with larger providers and regional initiatives to achieve economic, environmental, and operational efficiencies, rather than developing isolated local data centre capacity.

 

How will we ensure that this investment delivers a clear return, and how will success be measured and reported over time?

 

The investment is governed through a staged and controlled process, meaning funding is only released when there is confidence that proposed work will deliver measurable benefits. These benefits include both financial efficiencies and wider service improvements, such as increased productivity, better outcomes for residents, and the ability to manage more complex demand.

 

Success will be tracked through a defined benefits and value framework, with outcomes monitored during implementation and beyond. There are multiple layers of governance overseeing performance, including internal Council governance groups and the Shared Resource Service boards. This enables regular oversight of progress and impact. Scrutiny also plays an ongoing role, with further reporting expected through mechanisms such as self-assessment and review of enabling strategies, ensuring continued transparency and evaluation.

 

Do we have the flexibility to adjust or change direction if elements of the programme are not working?

 

Yes, the programme is deliberately designed to be iterative and adaptable. The use of stage gates and ongoing evaluation allows priorities to be reassessed and refined as the programme progresses. This ensures that if a particular approach is not delivering the expected outcomes, adjustments can be made before further investment is committed.

 

The overall approach is not a fixed plan but a responsive one, with continuous feedback loops built in so that learning from early implementation informs future phases and use cases.

 

How are we ensuring that staff are supported, reassured, and kept informed about the impact of AI on their roles?

 

There is a clear commitment to ongoing communication, engagement, and workforce development. The programme is aligned with the council’s ‘future focused workforce’ approach, which looks at developing the skills and capabilities needed both now and in the longer term. Staff are recognised as central to service delivery, and AI is intended to support and enhance their work rather than replace it.

 

Regular engagement will take place with staff and trade unions, reflecting obligations under the social partnership duty. Training and development will be provided to help staff adapt to new tools, and workforce planning will consider how roles might evolve over time. Changes are expected to be managed sensitively, including through natural staff turnover, allowing roles to be redesigned gradually where appropriate. The overall aim is to reduce administrative burden and enable staff to focus on higher-value, professional and interpersonal aspects of their work.

 

What safeguards are in place to prevent AI?generated risk indicators influencing safeguarding decisions inappropriately, particularly in a ‘single view of the child’?

 

AI is used only to analyse and present information, not to make decisions. It performs advanced pattern recognition on existing, trusted data sets and presents this in dashboards for practitioners. The responsibility for interpretation and decision-making always remains with qualified professionals, who use their judgement to assess the information. A ‘human in the loop’ is maintained at all stages, ensuring that AI outputs are advisory only and cannot determine safeguarding outcomes on their own.

 

Has a Data Protection Impact Assessment been completed, and will it be made available to members?

 

A Data Protection Impact Assessment has been undertaken as part of the work, and it can be shared with members. This is recognised as a key element of assurance given the sensitivity of the data involved. (ACTION)

 

How are we aligning this work with national Welsh frameworks, such as those developed through WLGA and Digital and Social Care Cymru, to avoid fragmented approaches?

 

The work is being actively aligned with national frameworks and governance structures. There is direct involvement in national data and performance discussions through the WLGA, including participation on relevant boards, and coordination through networks such as Data Cymru (now being integrated into WLGA structures).In addition, the Shared Resource Service maintains links with national structures and other local authorities, ensuring that learning and practice are shared. The approach also aligns with wider collaborative work across Wales and beyond, including developments in ‘single view of a child’ initiatives in other authorities and sectors. The intention is to contribute to, and draw from, a coordinated national approach rather than operate in isolation.

 

Chair’s Summary:

 

Thank you to the Leader and officers for their time. This is a very important report and could prove to be seminal when looked back on in future. A few points to note would be that the statement in 3.37 that ‘AI cannot and will not replace professional judgement’ might require clarification, as it would seem that AI is indeed used in this way. Similarly, in 3.38, the comment that AI is ‘not to bypass normal workforce processes’ might require clarification.