The way this usually works is that a cloistered cabal of civil servants and hungover political advisers get together in a room and decide everything.

I figured that we could start to use technology to do things differently — by opening up the process to hundreds of thousands of public-sector workers for the first time, and enabling them to feed in practical ideas about ways to cut costs in their own departments.

So we created a simple website where government employees could submit their thoughts, and emailed half a million public-sector workers to let them know about it. The tech experiment was a big success — and within just a few weeks we were announcing specific money-saving changes that had come directly from this digital initiative.

If I’d have stopped there, it would have been fantastic. But I figured that if this kind of “open source” policy making had worked so well with government staff, we should take the next step, and open it up to the public as a whole.

So we set up a Facebook page called the Treasury Spending Challenge, and waited for the collective wisdom of the British people to reveal itself. Almost instantaneously, the page was inundated with porn and abuse, and some wag quickly built a Facebook page with a big photo of a goat on it, along with the caption: “Can this goat get more likes than the Treasury Spending Challenge?” Needless to say, the goat won easily.

So when it comes to using technology to improve government, the upside can be massive, but you also need to know that sometimes things will go wrong along the way.

Nowhere is that more true right now than the opportunity for artificial intelligence (AI) software to transform the way citizens interact with the state, and how public services are run.

When I was in government, we replaced more than 700 separate public-sector websites with a single one: gov.uk, which saved a bunch of money, and also made it easier for the public to find the information they needed.

Today, you can take the next step: start to use AI chatbot software so that people can ask questions like “Do I need planning permission for my garden shed in Ealing?” or “What government benefits am I eligible for as a new parent?”, and start to get detailed and helpful information back.

Think of it like ChatGPT for government services — with big potential cost savings, because citizens won’t have to speak to contact centre staff to get the information that they need.

Another obvious area where AI could be used in government is in healthcare.

There are lots of ways this could have a positive impact, but one example is to use software to transcribe and summarise the conversation between the patient and doctor, add it to patient records, and then automatically make bookings with secondary care. What a help that would be — for doctors and patients alike.

One more area where AI could be transformative is in policing, where officers investigating a case often end up drowning in millions of emails and text messages, trying to find a needle of evidence in a giant digital haystack.

Getting AI to read that content and then being able to ask it questions (like “Did person A ever invite person B back to their house?”) could be a game-changer for the police, saving countless hours and enabling them to focus on the parts of their job that only humans can do.

Some of these ideas might sound far-fetched, but the speed that AI software is progressing probably means that in a few years’ time, these suggestions will seem mundane compared to what’s actually possible.

Whoever wins the next election is actually in a surprisingly fortuitous situation, given that their time in government is coinciding with the most significant technological shift since the dawn of the internet: the AI revolution.

If Rishi Sunak or Sir Keir Starmer make the most of this opportunity, they could end up presiding over an era of reform that’s rare in British politics.

As my experience of getting my arse kicked by a goat shows you, not everything they do will work out. Hopefully that won’t stop them trying.