Hippo joined two Global Government Forum events in London this October, beginning with the Data Summit on 14 October and continuing at Public Service Data.AI the following day.
At the Data Summit, Technology Director Ed Hiley and Commercial Director Phil Disbury joined the roundtable Improving data sharing across the public sector and making it happen. The session looked at why sharing data between public organisations remains difficult, even when there is broad support for doing it.
At Public Service Data.AI, which Hippo sponsored, Co-founder and Director Liz Whitefield moderated a fireside chat with Jake Richards MP, Minister for Sentencing, Human Rights and Technology. They discussed how data, AI and digital services could improve frontline services in the justice system. Phil also joined a panel on the shared vocabulary and infrastructure needed to help public servants find and use data held across government.
Key themes from Ed and Phil’s Data Summit roundtable
Data sharing agreements need to work in practice
There was clear interest in sharing more data between departments and public services. The questions were about how those arrangements would work from day to day.
Identity assurance was one concern. When services link records about a person, they need to know that they have found the right person. A wrong match can expose information or lead to the wrong decision. A missed match can prevent somebody from receiving support.
The UK digital verification services trust framework provides common rules for digital identity services, but the level of assurance still needs to reflect the service and the risk involved.
Accountability was another. Departments need to know which organisation controls the data, what purpose it can be used for and how each party will check that the agreement is being followed.
That includes the end of the arrangement. Review dates, retention periods, deletion and audit rights need to be considered before access is granted. The ICO’s guidance on data sharing agreements supports this approach, while leaving organisations to decide which controls are appropriate for the service.
Once the purpose and responsibilities are clear, teams can turn to the data itself and whether other people will be able to understand it.
Data needs context before it can support AI
Metadata was a significant part of the roundtable discussion. Sharing a dataset is of limited use if the receiving team cannot tell where it came from, who owns it, what its fields mean or whether it is suitable for the task.
Those questions become more important when data is used in an AI system. Teams need to understand the accuracy of the source, how representative it is and where gaps or bias may exist. They also need enough provenance to follow which information contributed to an output.
The same context helps public servants decide whether data is reliable enough for operational use now. Good metadata makes data easier to find, assess and govern, whatever technology is used later.
It also reflects Hippo’s existing work on AI-ready data, which puts origin, ownership and suitability ahead of choosing a tool.
As more decisions move away from the centre, local teams will need the data and authority to act on them.
Devolution needs leadership and local capability
The government’s plans to rewire the state brought localism, devolution and decentralisation into the discussion. Moving decisions closer to communities could help services respond to local needs, but responsibility needs to be matched by access to useful data and the capability to work with it.
Leadership matters here. Data work often spans professional and organisational boundaries, as well as political cycles. Policy, service, commercial, legal, data and technology teams need enough shared understanding to make decisions together.
The role of external consultants was part of this conversation too. Specialist support can bring experience and capacity, but it should also leave public sector teams better able to own the work. Hippo’s approach has been to work as one team, share decisions openly and build capability during delivery.
The idea of the citizen developer was also raised. Giving public servants more ability to shape tools around their work could be valuable, provided that skills, support and governance develop alongside it.
Devolution also changes how organisations think about data sovereignty, because not every type of data needs the same level of control.
Different data needs different levels of sovereignty
The roundtable considered how different layers of data may need different levels of sovereignty and protection. Highly sensitive operational data will not be treated in the same way as information already available to the public.
A useful model needs to distinguish between those levels, rather than apply one answer to every dataset. It should help teams decide where data can be held, who can access it and what technical or contractual safeguards are needed.
This also affects the cost and design of AI systems. Decisions about hosting, access and assurance need to reflect the data involved and the existing systems the new service must work with.
Public Service Data.AI: justice, vocabulary and infrastructure
In Liz’s fireside chat with Jake Richards MP, the focus moved to justice services. Better use of data could help frontline teams see more of the information they need, reduce administration and provide more joined-up support. Doing that safely depends on clear identity, access and responsibility across the organisations involved.
Hippo saw this in its work on Resettlement Passports for the Ministry of Justice. Research with more than 200 people showed how much time prison and probation staff spent finding information across separate systems. Bringing relevant information together, with access based on a person’s role, helped staff spend more time supporting people leaving prison.
Phil’s panel focused on whether public servants can find and understand data held elsewhere in government. That becomes difficult when organisations use different terms for the same thing, or the same term to mean different things.
A shared vocabulary can make data easier to discover and compare. Common infrastructure can make it easier to access. Neither requires every dataset to be moved into one central system. The current cross-government data sharing principles make a similar case for federating first, using common standards and reusing existing sharing solutions.
In some services, a focused integration will be enough. In others, people may need secure access to information without receiving another copy of it.
The choice should follow the service need, the sensitivity of the data and the responsibilities of the organisations involved.
What we would take forward
For public sector teams working on data sharing, the practical priorities are:
- Start with the public or frontline need and identify the minimum information required.
- Set identity, entitlement and access controls according to the risk.
- Make the purpose, ownership and responsibilities clear before data is shared.
- Agree definitions, metadata and measures of data quality.
- Decide whether users need a copy of the data or secure access to it.
- Set out how use will be monitored, reviewed and brought to an end.
- Build public sector capability throughout the work, rather than treating knowledge transfer as a final handover task.
Public sector data sharing is doing its job when staff spend less time searching for information, people do not have to repeat themselves and decisions are based on a clearer picture of their needs. It should also remain possible to explain why the data was used, who accessed it and when that use will end.
