How should senior leaders approach AI? Practical steps for public sector leaders

Senior leaders don’t need to become AI engineers, but they do need to become confident, informed owners of change.

That was the message from a panel discussion this week at Think AI for Government (pictured), which explored how leaders across government should approach the opportunities and risks of AI.

Rather than asking how quickly government can deploy AI, here are eight steps that focus on how leaders can create the right conditions for responsible adoption.

1. Start with the problem, not the product

Many organisations still begin with the tool and then search for a use case. The panel argued that leaders should reverse that logic.

“Quite often people will come to us with a solutionised problem and say, put AI here,” said Pippa Fernandes, deputy director I.AI strategy at DSIT.

“You do something for a purpose. You do something for an outcome. You don’t do it just for the sake of it,” agreed Fawad Qureshi, global field CTO at Snowflake.

Instead, leaders should define the challenge and let expert teams explore the best response.

“One of the best things that leaders can do is come to us with the problem they actually want to solve, and leave the space and agency for tech teams to work out what the answer is,” said Fernades.

That approach reduces the risk of automating poor processes and increases the chance of meaningful transformation.

2. Stay sceptical and challenge the hype

Jason Kitcat, director of digital, data & technology at the Department for Business and Trade, warned leaders to be “Very mindful, thoughtful around the claims that we are being told by suppliers.”

He noted that vendors have strong incentives to oversell. Public sector leaders therefore need to ask hard questions about capability, cost, scalability and evidence before committing public money.

3. Measure value from day one

One of the strongest themes throughout the discussion was evaluation. Leaders need to understand whether tools are actually improving productivity, quality or user outcomes.

“How do you know it’s actually working and delivering value?” asked Kitcat, who argued for before-and-after measurement so organisations can compare outcomes properly.

Fernandes added that evaluators must be empowered to be honest: “Look for the right answer, not the answer they expect… they know you want to hear.”

4. Fix your foundations first

Excitement about AI should not distract from the basics: data quality, legacy systems and process design. Poor foundations will undermine even the most advanced tools.

Said Fernandes: “Fix the data.”

She described the current moment as an opportunity to improve infrastructure that‘s often been neglected. In many organisations, the real value of AI may be that it finally creates momentum to modernise core systems.

5. Put trust and ethics at the centre

Public sector AI must command confidence from citizens and staff alike.

“Trust becomes the foundation. We are a lot more tolerant of mistakes committed by humans. We are much less tolerant of mistakes committed by machines,” said Qureshi.

He warned that people are often less tolerant of mistakes made by machines than mistakes made by humans. That means governance, transparency and clear accountability are essential.

He also cautioned leaders about reusing citizen data for purposes people did not expect: “Whenever you start using data for secondary purpose, double and triple check.”

For public bodies, legitimacy matters as much as efficiency.

6. Create a culture of curiosity

Leaders don’t need to become technical experts, but they do need to engage personally with AI and create space for others to experiment.

Fernandes challenged organisations to think about how many senior leaders have actually used the tools available to them.

She also argued that innovation should be encouraged rather than feared. Teams need permission to test ideas, share learning and improve services iteratively.

That cultural signal often matters more than any strategy document.

7. Build diverse teams to reduce bias

AI systems do not operate in a vacuum. They are shaped by the data they are trained on, the assumptions built into them and the people designing them. For senior leaders, that means diversity is not a communications exercise – it is a delivery requirement.

Qureshi said that bias in AI often reflects bias that already exists in society: “Bias naturally exists in our world. It is a reflection of human choices.”

If organisations automate decisions or recommendations without recognising that reality, they risk scaling existing inequalities. His advice was to involve people with a broad range of perspectives and lived experiences from the start of any programme.

“You need people with those lived experiences who can provide you those insights on how the data should be used, he said.

He illustrated this with an example from homelessness services, where raw data alone could not explain why some women declined access to showers. Only direct engagement revealed that remaining visibly unclean was, for some, a safety strategy to reduce risk of assault.

“Data sometimes does not give you the entire context.”

7. Empower your experts and hold the risk

The panel concluded that AI success depends on trusted, capable teams being given the authority to solve real problems.

Fernandes said leaders who are willing to support their specialists and accept some risk can unlock significant change.

“If you’re prepared to hold some of the risk… you can make huge change.”

In practice, that means backing digital, data and operational experts to work together, rather than expecting technology alone to deliver transformation.