How to get AI right in government

Image by Alexandra Koch via Pixabay

During a Global Government Forum webinar, a panel of experts discussed how to drive value through AI deployment, what the public sector can learn from the private sector when it comes to externally-facing AI, and how digital labour is ‘making humans more human’

The UK government is focused on using artificial intelligence and workplace transformation to make government more efficient – and many organisations are now entering a pivotal phase of accelerated change.

So, during a Global Government Forum webinar supported by knowledge partner Salesforce, experts from the Government Legal Department, the Met Office, UK Government Investments and Salesforce shared their experiences and advice for driving value with AI – and how digital labour is placing greater emphasis on the human characteristics that tech could never replace.   

Resisting the sugar rush

David Lewis, deputy director of the AI programme division and strategic innovation within the Government Legal Department’s COO group, began by setting out the benefits and challenges of the legal profession’s application of AI. 

He said AI is being used cautiously in the department, starting with trials of tools that assist with day-to-day tasks such as summarisation and drawing out points and action priorities from meetings, which his team report has resulted in time savings.

The plan now is to focus on bigger use cases which is where Lewis expects to see the benefits of scaling generative AI, but as he noted this requires greater consideration in terms of responsible use. 

“If you’re looking at something new and you’re looking at something that you’re going to use at scale, perhaps for tasks that carry more complexity, more risk, then that’s going to take time to get right,” he said.

“There’s a great temptation with AI to get a bit of a sugar rush and to want to move forward quickly… but actually, to do it properly, I think it needs to be more systematic.”

What’s crucial is to make sure the technology in question has the right capabilities and that data security is fully baked in from the start – and that thought has been given to the effects of AI on the way people work.

“For those of us who work on AI projects, part of the job is to manage expectations,” he said.  

Implementing externally-facing AI

Francois Zimmermann, field CTO for EMEA at Salesforce, reflected on AI use in the public and private sectors and the takeaways for government.  

He said he doesn’t see a wide maturity gap between the private and public sectors when it comes to internal AI use cases such as the summarisation tools Lewis described and “human in the loop” tasks.

Where there’s a difference between sectors, he noted, is in the use of AI in external channels that “face off to a customer or a citizen or a regulated firm”.

He gave the example of a life sciences company which is using data to support patients’ use of personal diabetes devices through “proactive nudges” that aim to keep them engaged in their care pathway – an experience Zimmermann said wouldn’t be possible without AI.  

The private sector is also using AI in support channels, something made easier for private sector companies because often they have already established “co-tasking” rules between their internal specialists and the people running their call centres abroad.

“The way they’re thinking about external-facing engagement is very different to the way we’ve thought about it for the public sector… The leading firms are the ones thinking about the agent-led experiences they could offer – that’s the ‘true north’.”

However, he said such uses require a more structured approach. Guidelines need to be drawn up that dictate what work can be passed from people to digital labour and back again and what controls need to be implemented to ensure accuracy and foster trust.

“When I define an external-facing agent, I need to have a really clear definition of exactly what the role is. For each role, I need to know what data you can retrieve and I need to know what actions you can execute. Once you’ve started that design thinking, then you can start to have a list of jobs to be done by these digital agents on external channels,” Zimmermann said.

Is AI enhancing human capability?

Reflecting on her conversations with clients, Emily Hill, Agentforce & agentic AI lead for public sector EMEA at Salesforce, expanded on her colleague’s points about the delineation between what people do and what AI is tasked to do.

The buzz around AI means public sector organisations feel they need to be doing something with it but they don’t necessarily always know what yet.

It is therefore important that organisations think carefully about where it’s going to have impact and where it’s going to add value, Hill emphasised.

Often this comes down to understanding which jobs are better done by people and which are better done by digital labour.

“We need to look at the skillsets of the AI tools that we have available and the skillsets of our humans” and align who does what accordingly, she said. “That alleviates a lot of concern… because people seem to have this conception that AI is going to come in ‘Terminator style’ and take all our jobs and take over the world, whereas a lot of the customers I speak to say that actually AI is making humans more human.”

Building relationships and connection, and understanding the nuances of the way people speak and the terminology they use and their sentiment, are among the important human characteristics that AI “can never replace”, Hill said.

So, transformation comes not when we “crowbar AI into our process… but when we step back and say ‘let’s redesign our process around the skillsets we have available’”.

What Salesforce has been seeing, she continued, is the effective implementation of AI for repetitive, summarisation-type tasks that “opens up the doors for humans to deliver more complex tasks, to uplift their skillsets, to have more time to think and make decisions and to do the things that make them more human”.

The time of transformation

Edward Steele is IT fellow for data science at the Met Office, the UK’s national meteorological service, which is one of the organisations at the forefront of public sector AI deployment. He was next to address the webinar’s audience and explained how the organisation had accelerated its use of AI over the last five years and how its scientists are more confident as a result. 

The Met Office has always been a big data organisation and had long done “some permutation of machine learning or AI, although not necessarily under that label”.

However, there was a “transformational moment” round 2020 when there was a convergence of unprecedented volumes of data and increasing availability of powerful compute to enable scientists to make sense of it.

The organisation formulated its data science framework in early 2022 and, by the end of that year, the deep learning technologies coming out of big tech companies were enabling “really disruptive innovation” in the weather space, Steele said.

The Met Office then “pressed the accelerator”, launching a programme of AI for weather prediction so it could explore complementary approaches to its traditional physics-based weather models using deep learning. It partnered with the Alan Turing Institute to advance some of this work.

A lot of this work has now been “consolidated” into AI4Everyone, an organisation-wide change programme that fosters a culture of AI innovation, and focuses on how AI and machine learning is employed within its national capability space – and how what the weather will do is presented across its products and services. 

“It’s very much a journey we’re still on,” said Steele. “What has been encouraging has been seeing the transition in our staff from very sceptical – where we’re looking to break 60 years of numerical weather prediction dogma and how we can produce some of these forecasts – to curious, and then from curious to confident through the use of some of these tools.”

Defining the ‘art of the possible’

Shehroze Junejo, executive director and chief data officer of UK Government Investments (UKGI) within HM Treasury, elaborated on Steele’s point about culture, which he said “defines your ability to innovate”.

“There is often a reluctance to move away from the norm, from what you’re comfortable with. There is a concern that you might be replaced. And then there is a concern around trust and whether you can actually lean on [AI] to help you,” he said.

On top of that, “we have to recognise that AI is effectively a field of study” that spans digital and data and technologies that he highlighted the majority of people still find challenging to navigate.

His approach at UK Government Investments, therefore, is to start with three key things.

First is governance and management, “not designed in a way to stifle innovation” but to empower people to know exactly how to use AI and “how to move forward with it”.

The second is around skills and capability. Echoing Hill’s point, he argued that expertise is more critical than it’s ever been. “The ability to think critically, the ability to problem solve, the ability to apply your knowledge and experience is absolutely essential. That’s not been devalued, it has appreciated in value, because in actual fact, when you are able to do things [with AI] at lightning speed, you also need someone who is able to then ingest that and assess whether or not they agree with it.

“So critical skills and the ‘human-in-the-loop’ does stay, at least for the foreseeable future.”

Third, “the element we all need to do is define the art of the possible”. The way Junejo has defined this within UKGI is ‘driving change through innovation’ and identifying how you can operationalise different tools that will generate efficiencies.

To identify where AI is best applied, an in-depth understanding of the organisation is important. “I challenge anyone who is in any DDaT function to sit in the organisation they’re in and be able to define what everyone does in five simple steps. Because if you can do that, that means you’ve internalised the business, and you know where you can generate most value,” he said.

Sharing government AI use cases

Junejo talked through some of the ways UKGI is using AI. Its project management function is AI, but also digitally-enabled to map out hundreds of activities across hundreds of individuals, manage interdependencies, and automate risk reporting.

This ensures that projects can be critiqued and reviewed before being put to the senior management team.

It has also developed its own secure GPT models, which are utilised for specific use cases.

“If you think about what government often does in any kind of advisory or policy context, it is capture information, assess the gaps, consider comparators, do an options appraisal, and put out a submission.

“Each one of those five steps can have a generative AI element to it. And what we’ve done is walk our teams through those processes and show them how they can enable themselves through each step using the secure model that we’ve developed – we’ve seen massive benefits to that,” he said.

It also uses analytics to look at the government’s financial investments and liabilities, using machine learning and various algorithms to assess things like downside stress scenarios, for example, which can then be published and shared with the Office of Budget Responsibility.

“All those things took time,” Junejo said. “It does require leadership at all levels… and the ability to really be able to speak to the people that you’re trying to bring on the journey – which a lot of people talk about, but don’t necessarily get right.”

The ‘Making AI Transformation Happen in Government’ webinar was held by Global Government Forum in partnership with knowledge partner Salesforce on 9 September 2025. Watch the webinar in full here and hear the panellists’ answers to a wide range of questions on topics including:

The importance of avoiding shortcuts when laying the foundations for AI
Structuring internal data to deliver value
Applying AI to discrete tasks and processes
Use of AI to help expediate more joined up and innovative government policy
The energy consumption and environmental impact of AI
The evolution of AI and new technologies
And more