We Put AI in the Hands of Public Sector Leaders

Before we get into what happened on the day, it is worth setting out the challenges that came up in conversation, because they are not unique to the people in the room. We hear the same things regularly across the public sector.

Most organisations are not short of enthusiasm for AI. The pressure to do more with less, to reduce demand on services and to improve the experience for residents and citizens means that AI feels like an obvious place to look. The problem is turning that enthusiasm into something real.

A few things consistently get in the way. The first is that AI can feel abstract, and it is worth being clear about what kind of AI we are talking about. Many attendees were already using tools like Microsoft Copilot, which helps with tasks like drafting emails or summarising documents. What we were demonstrating is different: agentic AI, which does not just assist with a task but carries out multi-step processes on your behalf. The distinction matters, because the productivity gains are in a different order of magnitude. The second challenge is governance. There is a lot of uncertainty about what is permissible, how to manage risk, and what good oversight looks like. The third is budget. Organisations can see clearly that this technology would help with their resource pressures, but finding the upfront investment to get started is genuinely difficult, particularly in the current funding environment.

There is also the question of shadow AI. Staff across the public sector are already using AI tools, often without any central awareness or oversight. That is not a criticism of those staff members. It reflects the fact that the tools are genuinely useful, and that formal adoption has not kept pace with the reality on the ground. The organisations that acknowledge this and put sensible governance in place are in a much stronger position than those that try to ignore it.

Rather than recommending tools we have not used ourselves, we have built and deployed AI agents inside our own business first, using IBM watsonx Orchestrate as the platform. Our innovation lead and Client Engineering Manager talked through use cases that are live in our own operations. They are not pilots or proof of concepts, they are things we rely on, which means we understand both the value they deliver, and the effort required to get them to that point.

We were open about the fact that getting each use case into production took time and iteration. The platform is easy to work with and does not require developers or extensive technical resources to build agents, but identifying the right use cases, structuring the knowledge base correctly and embedding tools into existing workflows all take careful thought. We shared what we learned along the way because it gives a more realistic picture of what AI adoption involves than a polished case study would.

Our operations team used to spend time every morning reviewing overnight backup job results and investigating failures. We built an agentic AI system that does that work instead. Around 28 agents run overnight, check the results of our backup jobs, cross-reference documentation and produce a summary with recommended actions. The team still make the decisions, but they start the day with the analysis already done rather than having to do it themselves.

For public sector organisations managing complex IT infrastructure, this kind of automated monitoring can significantly reduce the burden on IT teams and reduce the risk of issues being missed. You can read more about how we approach data resilience on our website.

Our Security team deal with large volumes of alerts, and a significant proportion of them turn out to be false positives. Deciding which ones need urgent attention takes time and expertise. We built an AI-assisted triage system that analyses incoming alerts, checks them against a database of known false positives and produces a verdict with a confidence score before an analyst opens the ticket.

The analyst still investigates and closes the ticket. But they start with context rather than having to build it from scratch, which speeds things up considerably. This kind of AI-assisted approach sits alongside our broader managed detection and response capability.

IBM i systems underpin some of the most critical workloads in the public sector, including payroll and benefits processing. Institutional knowledge about these environments tends to sit with a small number of specialists, which creates real operational risk when those people are unavailable or move on. We built an AI assistant for our ProCare We built an AI assistant for our ProCare managed service that connects to job failure data and can investigate issues or analyse a whole fleet in plain English, without any data leaving the controlled environment. By encoding that specialist knowledge into the system, it becomes accessible to the wider team rather than sitting with one or two individuals.

Phishing is still one of the most common entry points for cyber incidents. According to the UK government’s 2024 Cyber Security Breaches Survey, phishing was the most common attack type identified by organisations, affecting 84% of those that reported a breach or attack. The volume of phishing emails that need to be reviewed puts real pressure on security teams. and the volume of reported phishing emails that need to be reviewed puts real pressure on security teams. We built an automated pipeline that picks up phishing reports, scores the indicators of compromise, and either blocks the sender automatically or flags it for review. This happens in seconds rather than the minutes or hours it would take to do manually. It is a practical example of what good exposure management looks like with AI built into the workflow.

The thread running through all four of these use cases is the same. AI handles the initial analysis. People still make the decisions. That balance matters, particularly in the public sector where accountability and governance requirements are high.

Lab 1: Citizens services agent

In this lab, attendees used an AI agent to handle resident queries about council tax. The agent interpreted questions, retrieved relevant information and drafted responses, working through the kind of queries that contact centre staff deal with every day.

Council tax is one of the highest-volume contact areas for most local authorities. A significant proportion of queries are routine and predictable: payment deadlines, direct debit queries, change of address, eligibility for discounts. These are exactly the kinds of queries where an AI agent can handle a large volume of demand, give residents a faster response and free up staff to focus on more complex cases.

The same principle applies across other areas of local government work. Housing enquiries, planning application status, waste collection queries and benefits eligibility checks all follow a similar pattern: high volume, relatively predictable, well suited to an AI-assisted approach. Find out more about how we work with local government organisations.

Lab 2: Compliance agent

This lab focused on compliance and policy work. Attendees used an AI agent to navigate a body of complex policy documentation, ask specific questions in plain English and surface the relevant guidance.

Compliance is a significant overhead in the public sector. Procurement rules, data protection requirements, safeguarding frameworks and audit obligations all generate large amounts of documentation that staff need to be able to navigate accurately. Getting it wrong is costly. But keeping pace with it all when teams are already stretched is genuinely difficult.

An AI agent can hold and query a large body of documentation consistently. It does not interpret the policy differently depending on who is asking or how tired they are at the end of the day. For organisations dealing with CQC inspections, information governance audits or procurement compliance checks, that consistency has real value.

This is one of the clearest examples of what our Managed AI service does in practice: it takes domain knowledge and makes it accessible to everyone in the organisation, not just the people who have spent years building it up.

Lab 3: Digital worker

The third lab demonstrated a digital worker handling a multi-step administrative process end to end. Rather than just answering questions, the agent gathered inputs, processed them according to defined rules and produced an output without manual intervention at each stage.

This is where the productivity potential is most significant. The public sector is full of processes that follow predictable steps but consume considerable staff time. Benefits processing, planning application intake, HR onboarding, referral management in health and social care. These processes are not complex in the sense of requiring professional judgement at every stage, but they are time-consuming, and the cumulative burden across a large organisation is substantial.

A digital worker does not replace the person at the end of the process who reviews and approves the output. It handles the structured steps that do not require human judgement, consistently and at volume. For healthcare organisations in particular, reducing the administrative load on clinical staff is one of the most direct ways to free up time for patient care.

The consistent response across all three labs was that people could see where this would work in their own organisations. That is the shift we were aiming for: from AI as an abstract concept to AI as something genuinely useful and within reach.

We built our Managed AI service because we kept having the same conversations with organisations that were interested in AI but were not sure how to get started without taking on significant risk or resource commitment.

The service is built on IBM watsonx Orchestrate and is designed for organisations that want to move quickly but responsibly. It uses pre-built agents and low-code tooling so you do not need a large technical team to get started. Governance and security are built in from the start, which matters in the public sector where data handling requirements are strict and oversight is non-negotiable.

We also bring our own experience. Every use case we demonstrated at the event is something we have run inside our own business. We know what the challenges are because we have worked through them ourselves. That means we can help organisations avoid the mistakes and build on what works.

If you are at the early stage of thinking about AI, we can help you identify the right starting point and build a business case. If you have already started but are struggling to move from experimentation to something operational, we can help with that too. Explore our Managed AI service page to find out more or book a meeting with one of our experts.