Government panel discussion, “Digitise, automate and innovate — Paving the way for AI,”

Digitisation and legacy modernisation: Setting the foundations for government AI

Jamil Hussein

AI might grab all the headlines, but it’s the essential foundational work — like dealing with legacy systems and processes to prepare good-quality data — that will allow it to deliver real value in government.

AI is only as good as the data — and the data is only as good as the infrastructure supporting it.

That was the key message from the recent government panel discussion, “Digitise, automate and innovate — Paving the way for AI,” featuring digital leaders from the Department for Business and Trade, the Home Office, the Cabinet Office and Zaizi.

Watch the webinar: Digitise, automate and innovate — Paving the way for AI

Legacy modernisation: The great escape

Digitisation, automation, and legacy modernisation in government are not just about addressing inefficiency and security concerns but also about seizing opportunities, particularly AI.

Chad Bond, director of strategy and innovation at Zaizi, framed the challenge as “The great legacy escape.”

“Everyone talks about AI, but actually it can’t thrive on some of these creaking legacy systems,” he said. “Neither can they operate in spreadsheets. There are numerous instances in many departments everywhere that still rely on shadow IT or spreadsheets.”

Chad gave examples of how Zaizi helped government clients move away from spreadsheets to a system that allows data sharing and prepares the organisation to innovate with AI. 

Crucially, addressing legacy and rolling out AI can happen in parallel, added Chad.

Richard Appiah, head of data strategy at the Home Office’s Migrations and Borders Group, agreed with Chad’s view on getting the foundation right. 

“AI can do all these magical, wonderful things — but there’s something important about quality, about getting the foundations and the basics right,” said Richard, adding that to unlock AI’s full potential, it’s crucial to focus on “data, people, and culture.”

Richard said his department must deliver clear outcomes, which are all “underpinned by really good quality data.” He gave a scenario of how inconsistent data from siloed teams and reliance on spreadsheets hinder the ability to provide reliable, critical information.

To tackle these fundamental challenges, Richard explained they’re assessing their maturity model. It will allow them to understand where their data is and pinpoint specific areas for improvement.

Getting the data right

Working in the relatively new Department for Business and Trade, chief data officer Sian Thomas MBE considers herself lucky not to deal with some of the legacy infrastructure that older departments face. 

She said it also isn’t by accident: “That’s because my teams work really hard to make sure that, as we’re doing the sexy AI stuff, we’re actually thinking about what improvements we need to make to the infrastructure and how we justify the funding for the nice, new shiny.”

Sian also acknowledged the “no-win situation” some departments face with legacy systems that become so outdated that the cost of replacing them outweighs the value they deliver.

Ensuring data is in a fit state extends beyond just numbers to include ”every piece of information” collected and held, says Sian. “We’re taking better care of our data, but we need to adopt that mindset for the other types of information,” she added. 

Dr Ravinder Singh Zandu, head of the digital and systems team in the Cabinet Office, also spoke about the importance of data.  

He touched on common discussion points around data volume and variety, and emphasised the importance of roles like data stewards and data ownership in streamlining data management. “Those new roles have to be identified and people made accountable,” he stressed.  

Dealing with expectations & educating the top

The AI headlines fuel inflated leadership expectations. It’s a big challenge for those delivering solutions — having to explain to senior leaders why the fragile data ‘under the hood’ can stall AI progress. 

The solution? Alignment, education and exposure. 

Dr Ravinder suggested showing the specific steps required to achieve an outcome — this will align expectations and help seniors understand what’s realistically possible.

Departments like the Home Office are improving data literacy amongst leaders by running hands-on “data-heist” workshops where leaders experience firsthand the consequences of poor data quality in simulated environments. “That really raised their awareness,” said Richard. 

Chad said it’s very hard for those at the top to understand complex technology. “They just want to see results,” he explained. “So start small, build the thing, show the thing. Get them involved and then you build a culture and a mindset of that conversation with the seniors.”

He used examples of how Zaizi worked with senior stakeholders in collaborative workshops to achieve consensus and demonstrate what’s possible.

But AI shouldn’t be used as a hammer in search of a nail — a point that needs to be communicated to seniors.

“Just because you can do something, doesn’t necessarily mean that you should,” said Sian.

She added that leaders must consider the ethical implications. While many departments have ethical frameworks, she doubts that senior leaders consistently consider ethics in their decision-making for AI projects.

How to start your AI journey

The panellists highlighted several approaches, from defining the desired outcome and identifying the most optimal use case to starting small, iterating, and using that success for wider roll-out. 

“To Chad’s point, it’s starting small,” said Richard. “It’s about understanding actually where do we get the most value add that supports our outcomes… and build on.”

Sian stressed again the importance of getting the foundation right — what she called “the really boring stuff” — like knowing what the data is, where it lives, and how it’s used.

Equally important is getting the people and culture right.

The panellists all agreed on the importance of a workforce that understands, trusts, and can effectively use AI. 

Chad cited the example of how Zaizi’s work with Border Force on ScanApp earned the trust of frontline staff. Zaizi’s user-centric approach meant we worked “hand-in-hand” with end users to understand their needs and pain points — gaining their trust and buy-in.

There was discussion on the importance of having ‘humans in the loop’ — both to train the AI models and for oversight. They also highlighted why it’s important to explain to staff that AI will automate repetitive and laborious tasks, allowing them to focus on more insightful and creative work.

Sian did raise concerns about agentic AI and how that might mean “less about humans in the loop and more about humans over the loop.” But she again reaffirmed the importance of having sensible and strong AI governance and ethics to manage those risks.

Collaboration and continuous improvement

Finally, the discussion underscored the value of continuous learning and collaboration.

A key part of that is improving data literacy across organisations. Richard wants everyone in the Home Office to understand that “data is part of their job. It’s not an add-on.”

He also spoke about the importance of sharing ideas both internally across departments and learning from external providers.

Chad highlighted the expertise within the private sector that government can learn from. He used the ScanApp example again to illustrate how Zaizi empowered civil servants to run the solution themselves. He stressed that every contractual arrangement should include knowledge transfer to build long-term capability within government.

To dive deeper into these discussions and gain more insights from these digital leaders, watch the full panel discussion, “Digitise, automate and innovate — Paving the way for AI“

Thanks for joining us! We’ll keep you informed with regular updates.

Subscribe to our monthly newsletter, including updates about our content and events

Consent(Required)
This field is for validation purposes and should be left unchanged.