At a recent House of Commons event hosted by HCLTech, public sector leaders discussed what it takes to move AI and data from experimentation into delivery at scale.
The session brought together Arjun Sethi, Chief Growth Officer and Global Head, Public Sector, Aerospace & Defence and PE Practice at HCLTech, Sue Bateman, Chief Data Officer at the Department for Environment, Food and Rural Affairs (Defra) and Johnny Wolf, Head of Enterprise Architecture & FinOps Lead at the Metropolitan Police. Miranda Sharp, Trustee at the Centre for Cities, chaired the discussion.
Sharp opened with two questions that framed the challenge facing public sector data and digital teams: “What is the role of a digital function when everybody is now a live user and experimenter?” and “Can a data strategy be separate from an organizational strategy?”
The discussion that followed showed how closely those questions are now connected. AI is moving quickly into everyday workflows, but organizations still need to determine how experimentation, governance, data strategy and public value fit together.
AI is becoming an organizational question
Defra already has an overarching digital strategy, alongside data and AI strategies developed over the past few years. Bateman said the speed of change is now forcing the department to reconsider how those strategies are governed and applied.
“The world is changing so quickly,” she said. The task now is to develop “a new approach to governance” that allows the organization to operate safely while responding to demand from users for new tools.
Some of those tools may remain experimental. Others could significantly change services and operations. The challenge is creating an environment that supports both without losing control over risk, data or accountability.
Defra is also working to become more outcomes focused. Bateman said AI is likely to affect areas ranging from performance and risk management to digital and data operations, adding that “there’s no area that is going to be untouched.”
This expands AI beyond a technology program. Decisions about its use increasingly affect organizational priorities, workforce capabilities and the way public services are designed and delivered.
Connecting data to business outcomes
Sethi focused on the role of domain expertise in turning enterprise data into useful outcomes.
“A new job that is emerging is called an ontologist,” he said. The role combines deep domain knowledge with the ability to structure and connect data assets around the outcomes an organization is trying to achieve.
In defence, that may mean someone with direct knowledge of military operations. In retail, it may be someone who can “thread all the way from an SKU sitting on a storefront all the way back into the supply chain.”
This domain understanding helps organizations create knowledge graphs that connect information across systems and processes. The resulting context allows developers and data teams to build products around real operating needs rather than abstract technical requirements.
Sethi said HCLTech is placing “a lot of emphasis” on bringing ontologists together with developers and data analysts to “produce products and assets that work toward business outcomes.”
His point was that data creates value when it is connected to how an organization works. Technical skills remain essential, but domain specialists help ensure that AI and data initiatives solve the right problems.
Creating a safe space for experimentation
The Metropolitan Police has taken what Wolf described as “a very cautious approach to AI.”
As a highly regulated and risk-averse organization, the Met began by establishing a Responsible AI governance group. It then created an AI innovation hub where technologies and use cases could be explored without exposing sensitive operational systems or data.
Most deployments so far have focused on business operations rather than the most control-sensitive areas. The innovation hub also allows internal teams and external providers to test ideas using anonymized or controlled datasets.
Wolf explained that the distinction between a proof of concept and a pilot is important. A proof of concept is highly experimental, has a defined end date and can be dismantled once learning has been captured. A pilot may be designed with the potential to move into production and can then interact with controlled datasets under stricter conditions.
The data function supports this process by helping determine what can be tested, which information can be used and whether an experiment has a credible route to production.
Bateman added that public sector organizations need to provide both “the parameters for people” and “the environments for people to experiment safely.” That includes sandboxes and faster routes for testing emerging technologies within the protection of the operational structure.
Governance must move at the speed of technology
Wolf said one of the most difficult issues is balancing “ethics, governance and speed of delivery.”
Governance is necessary, particularly in policing, but the delivery process also needs to keep pace with the technology itself. He said one initiative aimed to deliver technology within three to six weeks but ultimately took six months. Even so, it remained considerably faster than previous delivery models.
“If you don’t keep up and we don’t adapt our governance and the way we work, we won’t keep up with this technology,” he said.
Bateman agreed that existing ethical principles remain relevant. Responsible innovation and established ethics frameworks still provide a foundation, even as the technology creates new questions.
The issue now is how those principles are applied within systems that can operate more autonomously. Bateman said organizations increasingly need to determine at “which point the human is actually in the loop.”
In this environment, governance needs to become part of the design and operation of AI rather than a review that happens after development. The aim is to give employees enough freedom to innovate while maintaining clear limits, oversight and accountability.
Sovereignty is becoming part of the AI conversation
Sethi said AI ethics is also becoming more closely connected to national sovereignty.
“Increasingly, the world is glocal,” he said. Countries want to ensure that technologies operating within their borders remain subject to their laws, regulations and public expectations.
For service providers, this means ensuring that AI systems respond to the requirements of the country in which they operate. Sethi described this as applying national regulation “within the four walls of the country.”
Data location, model control and the treatment of sensitive information are becoming part of the wider governance discussion. Sovereignty increasingly affects the infrastructure choices organizations make and the types of AI they are willing to deploy.
Bateman said this discussion is beginning to emerge across government as well. Defra holds distinctive environmental monitoring data, raising questions about how those assets should be retained, protected and potentially classified in the future.
“It’s early days,” she said, but the strategic importance of such data means the sovereignty question is likely to grow.
Delivery starts with the user experience
Toward the end of the session, Sharp asked the speakers what question they would put to the wider room.
Bateman wanted to understand where others had found opportunities, but also “what is the biggest challenge that you see around this project?”
Wolf said he wanted to see more of a private sector approach to experimentation in public services: “Fail fast, experiment, do things much quicker.” Since joining the public sector, he said his personal challenge had been “changing the norm or challenging the norm.”
Sethi brought the discussion back to the people public services are designed to support. Public sector practitioners, he said, need to ask whether they truly understand the constituent experience.
“If we can identify those friction points, then we know what we’re trying to solve,” he said.
The structural barriers to AI and data at scale sit across governance, delivery speed, skills, architecture and public trust. Addressing them starts with a clear view of the outcome, the user and the point of friction. From there, organizations can create safer environments for experimentation, connect data to domain knowledge and build governance that enables delivery at the pace now required.
