What is the opportunity for AI in government? It’s vast - from identifying problems before they occur, such as using data to prevent expensive and complex interventions in health and social care, to digital twins of cities to model changes in infrastructure, and providing an always-on front door for government processes.
These are just a few examples that Craig Wentworth, Research Director at TechMarketView shared in his Future Gov Summit presentation. And there are many more waiting to be discovered – by piloting, assessing and measuring impacts, then scaling. The opportunity is all about finding the business value for AI applications – but it can be quite elusive.
Up till now, there has been a fairly narrow focus on efficiency, and time and cost savings using AI. This is to be expected, as they are AI’s fundamental promises.
But there should be more strategic thinking – for example, if AI can free up employees’ time, what should they do with it? And how can we measure and prove AI’s impact on service quality for citizens?
Some proven use cases
According to a recent National Audit Office (NAO) report on the use of artificial intelligence in government, 74 AI use cases have already been deployed in government (as at Autumn 2023) .
Some successful examples include a council using AI to predict and prevent homelessness, with its pilot project driving a 27% improvement compared to the standard non-technology approach.
A hospital used AI to identify patients who are at increased risk of hospitalisation within a year. These individuals were then invited in for holistic assessment, helping local teams work together on more personalised care support plans. This reduced A&E visits by 60% during a trial in care homes.
Another council has used an AI-powered multilingual voice assistant as a digital front door to its helplines. This helped achieve an 84% drop in calls transferred to its call centre at peak times, with over half of calls handled without human intervention. The tool has generated approximately £7.5m in savings.
A delegate at the event also shared an instance of using Copilot in a social care conversation setting – just by having it record and transcribe in the background meant not having to take notes, which improved the quality of the conversation.
Some projects have highlighted inconsistencies, though. A recent three-month trial of Copilot across departments delivered time savings of two working weeks per person per year. But another trial produced outputs that needed fixing in some instances, which affected those time savings. AI projects can also give rise to what Craig called ‘productivity theatre’ – creating the illusion of getting more done.
According to the state of digital government review, only 8% of AI projects show measurable benefits. And many pilots have not scaled – there are often vast gaps between a pilot and implementation, as they only prove a technical concept, not business value, and there is adoption inertia beyond a few willing enthusiasts.
Get the foundations right to give AI a chance
For AI to deliver on its potential, it must be combined with a strategic review or redesign of processes, workforce skills assessments (70% of respondents cite skills as the primary barrier to AI success, according to the NAO Use of artificial intelligence in government report), and data improvements.
Government data can be trapped in proprietary formats – 70% of government departments say their data landscape is not interoperable and does not provide a unified version of the truth, according to the recently published State of digital government review. As Gareth Davies MP observed in a speech to MPs and civil servants in January last year: “Poor data means arriving at wrong answers, only faster.”
But the underlying systems are blockers too. The ‘Cinderella systems’ propping up government services use technology stacks that are often incompatible with modern frameworks, because they rarely support the API layers that are needed to unlock access to their data. And they can also contain security vulnerabilities that prevent cloud deployment.
How to begin, and make progress
According to the NAO Use of artificial intelligence in government report, only 21% of public sector organisations had a strategy for how to approach AI at all (as at Autumn 2023). So, to understand – and harness – the AI opportunity, here are some tips Craig shared.
- Look around and learn from successful use cases in other sectors
- Start small but think big by understanding the full potential and selecting something specific to pilot
- Build on a flexible platform that will adapt to rapid innovation
- Address procurement shortcomings and consider using an open framework rather than locking in suppliers for years
- Engage stakeholders early and often, and listen to their feedback so the implementations are genuinely useful
- Build trust and understand AI’s limitations, e.g. when it makes things up if it doesn’t know the answer – AI tools are designed to be pleasers, so knowing these shortcomings means users can adapt their prompts
- Count what really counts rather than what’s easy to count – cost and time savings are obvious, but what about impacts on service quality?
- Pivot or abort if it’s not working because it’s better to fail and learn
- Create networks to share learnings so benefits can be transferred and scaled
With a combination of spadework and a structured approach, AI pilots can deliver benefits, scale, and pave the way for more pilots and wider adoption. This is the opportunity – now, more of government needs to follow the right steps to seize it.
Capita is hosting a series of free-to-attend AI bootcamps facilitating open, honest conversations about your hopes and challenges, and how we could help you.
To register send an email to bettergovernment@capita.com