Re:Shuffle – an open letter to Rt Hon Liz Kendall MP, Secretary of State, and Kanishka Narayan MP, Parliamentary Under-Secretary of State, DSIT
Directory of Strategy
Dear Liz and Kanishka,
Congratulations on your appointments at DSIT! As the department that makes the future happen, it couldn’t be a more exciting time to be at the helm.
DSIT is still a young department in government, and one of the most important things you can do is continue to drive a culture which is focused on the titanic changes the world is undergoing, rooted in new scientific discoveries and technological revolutions.
There are both opportunities and risks for Britain in this more dynamic world order, and DSIT carries a tremendous responsibility on their shoulders to manage both.
Since the election, DSIT has held the responsibility for both science and technology policy in the whole economy, and also responsibility for the ‘digital centre of government’ - including the Government Digital Service, Incubator for AI, and other functions brought over from the Cabinet Office. This combination has the potential for great synergies, but also means it’s easy to overlook some of the day-to-day management issues which government technology involves when presented with great strategic issues like frontier AI development.
Here are four priorities from us at Re:State for DSIT and GDS where you can drive significant advances in better digital government:
- Invest strategically
Investment in AI adoption across government is incredibly chaotic. Our research found duplication, and poor sharing of evaluations and lessons learned, in areas where the government is trying to use AI for public services transformation.
DSIT should drive adoption in areas where there’s an evidence base and already widespread private-sector adoption. The arguments that the public sector is special, and needs entirely different kinds of software to the private sector, are usually wrong. In areas like casework automation, translation and transcription, and coding co-pilots, government is unnecessarily lagging private sector adoption.
Rather than using complex and lengthy business cases for small AI investments, we have argued that DSIT should empower Programme Directors with the ARIA model – giving them widespread freedom to invest in AI adoption as they need, based on an overall single business case for an AI Transformation Fund.
- Balance risk and reward
Public services today, without much automation, frequently make mistakes. They ‘hallucinate’ processes and rules which aren’t founded in the law, aren’t accountable for the decisions they make, and often can’t even explain them. Despite relying on human intelligence, not artificial intelligence, modern public services often have the same failure modes.
But AI adoption plans in Whitehall often rely on benchmarks for model performance which involve high performance in 100 per cent of cases. These are unrealistic by the standard of any normal public service.
DSIT should issue guidance on measuring the performance of automated software directly against the level of performance in existing services, rather than artificial standards of ‘best practice’ which are not currently met. If AI can perform to the same level, or better, than a caseworker (and is cheaper), then that is still better value for money for the taxpayer.
- Buy better
Most AI procurement is still being done through services-based, ‘time and materials’ contracts for developing bespoke technical support. This has contributed to an information asymmetry, where the public sector is hugely reliant on ongoing support from private companies to deliver AI projects. Government buyers shouldn’t be afraid to use more ‘commercial off-the-shelf’ AI tools – the private sector moved to buying most software through this license-fee model years ago, but it’s still incredibly difficult to buy in Government outside of cloud platforms.
GDS should launch a new procurement framework for buying AI tools off the shelf, with light-touch contracting requirements so public services can determine which are best for them through prototyping and experimentation, using approaches like the new competitive flexible procedure.
- Fund teams and services, not projects
The funding and budgeting of Government IT is designed for an era of software development which has long passed. Rather than large, bespoke builds for discrete tools, most private companies fund software in an Agile way – paying for teams and services on an ongoing basis, and iteratively transforming them. The ‘big bang’ transformation programmes that the current Programme Business Case encourages are prone to failure, and accepting that digital transformation is a function, not a one-off cost, is key to getting better value for money.
The Treasury is piloting a new approach following the Performance Review of Digital Spend, and this should be expanded across Government to enable much more agile and portfolio-based funding decisions for digital transformation.
We wish you the best of luck,
Joe Hill
Policy Director, Re:State