For Yvonne Gallagher, Director, Digital Insights at the National Audit Office (NAO), the biggest challenge in successfully adopting AI is not the technology itself, but having the right organisational foundations in place.
While AI has generated significant interest across government, Gallagher argues that successful adoption depends on the same factors that underpin digital transformation more generally: an honest assessment of the organisation’s starting point, realistic aims and ambition, strong leadership, effective governance and high-quality data.
These themes sit at the heart of the NAO's recently published “Good practice guide for organisations using AI”, which aims to help audit and risk assurance committees develop an awareness of the “significant risks and challenges” associated with overseeing the planning, deploying and scaling of AI.
Across government and beyond, many organisations hope that AI will be able to make a significant difference to their work. And indeed, Gallagher pointed to a number of examples of impressive AI deployments in the public sector, ranging from medical diagnostic tools to fraud detection and risk assessment, adding that the NAO is planning to carry out research in a few months’ time to uncover further examples of successful adoption.
For Gallagher, one of the most important considerations for leaders seeking to deploy AI is understanding of how the technology will impact organisations or departments as a whole.
She said: “Understanding end-to-end workflows is key, and our work has shown that there's often limited understanding about them, and about where costs lie or who has overarching visibility.”
According to Gallagher, this lack of visibility is one reason why organisations often underestimate the difference between running an AI pilot and deploying AI successfully at scale.
"People have got very excited about the pilots. You can do them very quickly," Gallagher said.
"But really, there needs to be an acknowledgement that moving from pilots to scaled deployments is not an easy undertaking."
Scaling AI requires more than proving that a technology works. Developing a more comprehensive understanding of organisational workflows would enable leaders to set a clear vision for what AI deployment is intended to achieve and redesign processes accordingly.
This means rethinking processes, redesigning controls, training staff and ensuring that new ways of working are embedded across an organisation. Without that broader organisational change, initiatives can struggle to deliver lasting value, Gallagher stressed.
Good governance is also crucial. As organisations move beyond experimentation, effective oversight becomes increasingly important.
This challenge extends beyond internal oversight. Gallagher urged leaders not to overlook the wider context in which AI is being marketed. Organisations need to understand the commercial environment in which AI is developing, including consumption-based supplier pricing models and the long-term costs that may accompany large-scale deployment.
Another significant aspect of a successful AI deployment is data quality.
“Successful examples [of effective AI deployments] seem to be where there are rich datasets”, Gallagher said. Conversely, “if data is missing, unverified, or biased in some way, it will produce unreliable AI outputs.”
A contributing factor to the prominence of poor data is the prevalence of legacy systems across government, which often capture and store information in silos. The costs of maintaining these ageing systems also detract from the resources available to invest in AI and wider digital transformation, Gallagher argued.
Despite this, however, she believes that things are looking up. The government is looking into how to improve its legacy base, setting the foundations for future AI adoption. This will allow for functions such as improved financial reporting, an initiative recommended by the NAO which has the potential to help government become more productive, something which it has already started to work upon.
The “insights guide” published by the NAO is one of a series of guides to support the work of senior leaders across government. It was developed in response to demand from Audit and Risk Assurance Committee chairs, many of whom are now expected to oversee AI despite the area being new to them.
Gallagher said: “We targeted this at risk and audit committees and senior leaders, because we know that they're not specialists, but interestingly everybody now is having to think about AI…they're all grappling with the same issues.”
For Gallagher, that reflects a broader shift in how AI should be viewed across government, and highlights a central message of the NAO’s work.
“What we really wanted to do here was show, as we have done in our other work, that digital and now AI is a business issue,” she said.
She added that the organisations most likely to benefit from AI will not necessarily be those with access to the most advanced technology, but those with the strongest foundations in leadership, governance, data and organisational capability.