The strategy is transformation, again

For those of us who went through the digital transformation of government at the start of the early 2010s, these times feel familiar. We are on the brink of another wave of transformation through AI. Back then it was analogue to digital, channel shift and reimagining what public services would be like if they were designed digital first, using internet era ways of working. 

With the exemplar services going through transformation and gov.uk becoming the single domain of government, the sense of possibility for how to improve and reimagine services was obvious. I remember working on the digitisation of ‘Renew Your Passport online’ at the Home Office. Watching users realise they could simply take a passport photo on their phone instead of visiting a high street store or photobooth felt like magic. Our research sessions were often derailed as participants stopped to confirm – "Wait, I can just use my phone?"

We spent months designing, testing, and refining instructions to ensure selfies would pass official checks. More months went into understanding the challenges of transferring photos from a camera or phone to a computer and then uploading them. New, exciting solutions required new ways of helping users navigate this new world.

This time round feels different because it is. The world is a different place and we are feeling the effects of tech distrust and fatigue. Watching charlatans profit off the hype cycle of things like NFTs or the blockchain, or indeed AI. Also, AI transformation isn’t just digitisation or service improvement like it was in the early 2010s; AI is novel and we don’t know its impact yet. And people have legitimate concerns about its effect on environment, equity, accuracy and safety. How do we embrace this transformation, but also develop it in a way that makes sure it delivers fair and safe outcomes? Keeping pace with rapid technological change while ensuring services remain user-centred and inclusive.

The techniques and tools that were guiding our work back then are just as relevant now. Here are some key takeaways from the past that still hold true. 

1.Start with user needs

As Mike Bracken wrote back in 2013 – “it’s usually the way with all large, rules-based organisations, that more time and effort is spent on internal logic and process, than on listening to and understanding real user needs.” 

The risk for departments is focusing only on internal AI applications. Or simply thinking that applying AI to a transactional service – without considering the user needs in the context of that service – will add more complexity or even make the experience worse. Ask yourself: What does the user need? What problem are you solving? Use the answers to these questions as the guardrails for every decision you make going forward. They will make sure your technology is tightly scoped to your outcomes.

2. Work in the open

Back when gov.uk and transactional digital services were forming, there was a flood of blog posts documenting the process, making it easy to follow along and learn from the work being done. It was great. But skip to today, and I’ve read almost nothing about the experiments going on in AI across government. 

The reality is, people don’t know what others are up to, even when they’re working in similar areas. Let’s start sharing our experiences again. What have you learnt that others would benefit from? This reduces duplication of effort. It can help provide guidance so you can recognise and spot familiar mistakes early. The lack of this sharing is a missed opportunity for collaboration, learning and improvement.

3. Test and learn

The test and learn approach, trumpeted by Pat McFadden, is how we should approach the process when developing AI for government and public services. Working in government the last few years, anecdotally it feels we have taken a step back from testing and learning. Powerpoint decks and Word docs seem to be the prevailing currency of some digital teams, not working software. As Anna Gos wrote, it’s 2024 (now 2025) on the internet, so let’s act like it. Let’s push ourselves to deliver AI-powered solutions into the hands of real users – testing, learning and documenting as we go.

4. Deliver often, iteratively and repetitively 

One of the mantras during the last transformation was that the strategy was delivery. This could be misconstrued as valuing delivery over quality. My interpretation was always: ‘don’t let perfect be the enemy of good, and get stuff shipped’. I understand how difficult it is to not get stuck in endless planning or perfectionism when working in government, but when you focus on delivery, it really makes a difference. 

If we want to harness the power of this new technology, we need to show how people are using it, rather than discussing its potential. I worked on the Alpha of gov.uk Notify. In meetings, our government colleagues expected us to discuss the service. Instead, we arrived with the service – ready for people to sign up and use it on the spot. 

In Estonia, AI is integrated into public services to make them faster and more accessible. Instead of lengthy discussions, they’ve built AI applications that automate communication between citizens and government agencies – reducing response times and improving service delivery. 

Let’s not get bogged down in policy and governance; let’s prioritise the fast delivery of working use cases.

5. Safety first 

As our CEO Marc wrote last month, “AI is the most important technology of our time”. And it’s only by safely leaning into this transformation that we’re going to make the most of it. 

When looking at how AI can transform your department or service, think about safety first. What unintended consequences should we design for? What different groups of people, circumstances, and behaviours are important? Who might experience negative consequences from our decisions? Can users understand how the AI makes decisions? How is the data collected, stored and protected?

These are a few questions you can start thinking about. AI is a strong tool. It needs careful safety measures to prevent misuse. This includes stopping false information and biased content from spreading.

6. Outcomes over outputs 

Good service delivery is all about outcomes. Be specific about what you expect AI to help you achieve. It will ensure you deliver real public value. What is it we’re trying to achieve with the service? How will you measure whether or not what we’ve done is successful?

You should have the answers to these questions before you kick off your project. Government budgets are limited so every AI investment should drive measurable service improvements. For example, instead of just automating form processing (output), the goal should be reducing wait times and improving citizen satisfaction (outcome).

What should we do differently this time?

We’re at a turning point. Data literacy and AI-driven data science – large language models (LLMs) and other advanced systems – are no longer just concepts; they’re mature enough to drive real change. Instead of just talking about transformation, we should be using these tools to improve government services now. 

From improved efficiency to enhanced decision-making, there are so many opportunities to deliver better public services. Service communities were too early in the early 2010s, and the infrastructure was not in place. Now it is. This is our chance to test and learn outside of the department verticals. We need to share service transformation between departments. Let's create reusable platform components. These will help us meet user needs with greater speed and efficiency.

Bringing it all together

We don’t need to start from scratch. The same principles that have shaped great digital work still hold up today. As AI becomes a bigger part of our work in government, sticking to these foundations will help us build solutions that are useful. It’s easy to get swept up in the excitement. But what counts is putting working solutions in people’s hands. Then we must learn from real use, and adapt quickly.

Let’s keep it simple. Share our plans, test ideas safely in the real world, and strive for AI that truly makes a difference.

If you were around last time, what would you do differently this time round when it comes to enhancing public services in this AI era?