Responsible AI Adoption: A Practical Framework for Public Service Leaders

In a previous article, I explored how public service leaders should understand AI, the opportunities and risks it presents and also the danger of using it simply to make existing arrangements more efficient.

In this article I argue that you should start with the change people need from public services, and then make AI earn its place in delivering it. Plenty of people are saying this – these are just some ideas on how you might actually do it.

The biggest challenge is not simply understanding AI (you can put the work in and achieve that) or identifying potential applications (we’re all being inundated with ideas and quick fixes). Some will improve productivity. Others may reduce admin or improve access to information. All might be worth pursuing.

However, the biggest challenge is how to think about the AI opportunity in the context of the desperate need for broader public service reform. How do leaders match AI investment with wider ambitions for prevention, integrated working and financial sustainability? How do they make sure that improvements in individual processes translate into a different experience for residents?

As I have been thinking about responsible AI adoption in public services, it’s become clear to me that AI doesn’t actually need a new framework of its own – although many will offer one!

This technology needs to show how it can support and accelerate already well established thinking, particularly around place-based, relational working and on how we can create an “abundance” of the things we know people need. This will require a new mindset - but it certainly doesn’t require us to discard the thinking and progress which has been made over recent years.

The work we do on place-based, relational  working has developed around three interconnected environments  - Enabling, Delivery and Sense-making. This approach has been developed with Mark Smtih and others over a number of years as a general framework to consider reform – but it also provides a practical way for public service leaders to assess where AI might add value.

Importantly, it starts with the public service challenge rather than the technology.

Start with a reform ambition, not a list of AI use cases

Imagine a council that wants to improve support for families experiencing multiple and interconnected difficulties. The council is experiencing rising demand for statutory services but also a shared ambition with partners to strengthen early help. This is a situation many will recognise.

Meanwhile, teams across the organisation are independently exploring AI…children's services are considering tools to support case notes. Customer services are looking at automated interactions. Corporate teams are experimenting with document production and information retrieval.

Each initiative will probably have merit - but they are more likely to be focused on becoming more efficient at what already happens.

A different starting point would be to establish a shared ambition for AI – something like - We will use AI to help families receive support earlier and prevent escalation - and reduce unnecessary administrative activity in service of that.

This is just an example - but an ambition like this changes the role of AI and is where the three environments framework becomes useful. It allows leaders to examine the AI opportunity from three connected perspectives: the conditions required to make change possible; the changes needed in service delivery; and the evidence and learning needed to understand whether the new approach is working.

Leaders and teams need to work across all three throughout any reform process.

Let’s look at each in turn, starting with the enabling environment.

Enabling: Create the conditions to make different decisions

Making decisions that cut across existing organisational and service boundaries - and then sharing resources (funding, data etc.) across those same boundaries – is hard!

Returning to our example, earlier help for families who need the most support usually requires closer working between children's and adults’ services, schools, housing, NHS and voluntary organisations - maybe even police and the justice system.

A promising idea might require one organisation to spend while another receives much of the benefit in terms of immediate savings or reduced future demand. Professionals may recognise opportunities to achieve better results for people (and reduce overall demand) but lack the authority or evidence to challenge established processes.

In the context of Responsible AI Adoption, the Enabling Environment part of the framework is about how you can use the AI opportunity to reduce the thickness of the service silo walls and soften the barriers being experienced by public service teams.

Can AI be used to better track resources so Finance Directors are more likely to approve innovative ways of working? Can it be used to provide more confidence around information and data sharing so the default answer is not always “no” - or a very protracted/unsatisfying “yes with caveats”?

In practice, this means bringing the relevant leaders and partners together around a shared reform ambition, rather than sitting back and letting each organisation develop its own independent AI priorities. It means identifying the decisions that need to be made collectively, the resources that could be used differently and the accountability arrangements needed to support change. I’m under no illusions that this is easily achieved!

There are also important Enabling questions about where AI should and should not be used. In sensitive people services, leaders need to establish clear boundaries around professional judgement, safeguarding, data use and the authority given to AI systems. These safeguards should be designed around the service model and the people it supports, rather than treated as a separate technical compliance exercise.

It is also worth a comment on leaders Enabling in a sustainable way. Right now, the pricing of AI products is not reflective of the cost of providing the service. The companies are in the “grow your customer base” phase and are practically giving it away. As with all rapid growth tech innovations, the plan is to make it readily available and get people hooked and reliant - then jack up the prices. We need to be aware of that when modelling the use of AI tool, especially if the plan is to make them a critical part of future public service delivery.

Delivery: Redesign the work before selecting the technology

The Delivery Environment takes the reform ambition into the reality of how services operate. This is where detailed knowledge of people services becomes essential.

A conventional technology-discovery exercise might simply examine the tasks undertaken by staff and identify those that could be automated.

A reform-led approach goes further. It takes the space created by the Enabling environment, examines how people experience support, how work moves between professionals and organisations, where time and resources are consumed and which activities genuinely contribute to better outcomes.

Define what is important. Eliminate what is not.

The resources which can be deployed are limited so avoid wasting time automating the unimportant as it will just create more work in the long run - even if parts of it are automated and handled by AI. This is an opportunity to ruthlessly create an abundance of things that matter, of things that are important.

Do not create a false sense of security and the comfort of action by automating the unimportant!

Considering the experience of a family who have repeatedly sought or been offered help or whose needs have escalated into statutory services and how public services best supports them. AI might assist with summarising information, retrieve policies or legislation, help with case notes, reduce administration duplication - these are all the obvious things. Taking it further, it might even play the role of an intelligent assistant who is across data and information in a way that a human couldn’t possibly be, suggesting actions/ people to check in with (as they may be approaching crisis), preparing teams with information and ideas for conversations. However far you take it, the technology must be selected to support a redesigned model, not the other way around.

That distinction is particularly important when considering how to convert efficiency gains into real benefits. Suppose a tool releases four hours of administrative time per professional each week. That tells us something about productivity but it doesn’t tell us what has changed for families. A reform programme should make an explicit decision about the intended use of released capacity. It might allow professionals to spend more time with families. It might reduce reliance on temporary staffing. Or, it might contribute to cashable savings.

The practical output of this Delivery environment should be more than a prioritised list of AI applications. It should be a redesigned way of working, clear changes to roles and processes, an explanation of where AI contributes, and a credible plan for embedding and evaluating the changes.

Sense-making: Build the ability to know what is working

The third environment is where I think public service leaders have the opportunity to be more ambitious.

A council introducing AI will understandably want to measure its performance - How much time has been saved? How accurately does the tool operate? How many staff are using it? What does it cost?

These measures are useful, particularly when some are skeptical about the whole notion of AI.

A reform programme needs to understand something different: Is the new way of working (not just the AI bit) genuinely improving people's lives and changing the pressures on the wider system?

In our council family-support example, that might involve examining whether families receive help earlier, whether they experience fewer handoffs, whether professionals have more time for meaningful support and whether needs are escalating less frequently. But it should also start to look at whether the family has had meaningful life improvements.

It also involves looking at the financial consequences. Has the service avoided additional costs or redirected resources? Are costs and demand shifting elsewhere?

These questions cannot be answered through an AI supplier's performance dashboard. They require a broader approach to evidence.

The Sense-Making environment brings together quantitative information, lived experience, frontline professional judgement, evidence about outcomes and understanding of demand across organisational boundaries.

AI enhanced operational data could show changes in referral patterns and professional workloads. Families should be asked whether support feels more coordinated and useful. Practitioners could identify barriers that are not visible in the data. Financial analysis could examine whether demand and expenditure are changing over time.

Perhaps the new arrangements have reduced duplication but not improved access to early support. Perhaps administrative time has been released but workload expectations have prevented it from being used differently. Perhaps demand has reduced in one service but increased elsewhere.

These findings are not simply measures of success or failure. They are information that should shape the next decisions about service delivery. This is what makes sense-making an ongoing reform capability rather than a periodic evaluation exercise.

The practical output should be an agreed way of understanding outcomes, demand, experience and financial impact, together with a mechanism for using that learning to change decisions.

A different way to make the business case

This approach also changes how councils should think about investment.

Some of the benefits of AI may be immediate and measurable, such as reduced administration or lower spending on specific activities. Others may involve improved quality, greater professional capacity or better experiences for residents. Some of  these important benefits, particularly those associated with prevention, may take longer to emerge and may be distributed across several organisations.

That creates a challenge for investment decisions. A council may be reluctant to fund an intervention if the financial return is uncertain or largely benefits another organisation.

Those responsible for creating the Enabling Environment must recognise this early and make the case that this is not simply about adopting an AI tool which will makes things more efficient - it is about changing how public services operate, with AI playing a defined and proportionate role.

The next step for public service leaders

The debate about responsible AI adoption is important, and the guidance available to public service leaders is improving.

But the next challenge is to move from principles to practice. We believe the Enabling, Delivery and Sense-Making environments offer a way to do that.

For a council considering its next phase of AI investment, we would suggest starting with one significant reform ambition.

It might be one of the Prime Minister’s top priorities such as tackling homelessness – or one of your own such as improving early help for families, reducing avoidable demand in adult social care, strengthening neighbourhood working or simplifying the experience of residents who interact with multiple services.

Use the three environments to examine what needs to change, what AI could make possible and how the benefits will be realised and understood.

Start with the change people need from public services, and make AI earn its place in delivering it.

If you are interested in discussing any of this, I would love to hear from you – andrew@mutualventures.co.uk