How can government scale up AI responsibly?

In the new TNO report ‘Scaling AI responsibly in government’, researchers explore which strategies, choices and considerations play a role in scaling AI applications within the Dutch government. The study combines insights from scientific literature with practical experience from concrete scaling initiatives.

Seven strategies for scaling AI

The research shows that scaling is neither a uniform nor a linear process. Depending on objectives, context and collaboration, organisations consciously or unconsciously choose different forms of scaling. The report identifies seven AI scaling strategies, including scaling in (scaling within one’s own organisation, such as ChatILT), scaling out (reuse by other organisations within the same domain, such as GovChat-NL), and scaling together (where multiple parties jointly develop an AI application with scaling as a starting point, such as OmgevingsChat).

Eight factors that make or break scaling

In addition to scaling strategies, the researchers have identified eight aspects that play a role in almost every scaling trajectory. These range from technology and documentation to culture and organisational support, and from AI literacy to funding and legal frameworks. These aspects can act as either success factors or barriers, depending on how they are addressed.

‘What we see in practice is that organisations are often willing to scale, but do not always have a clear understanding of how to achieve it,’ says Tessa Bruijne, one of the report’s authors. ‘Scaling is not an automatic next step after a successful pilot. It requires strategic choices, collaboration, and attention to public values.’

An important conclusion is that scaling AI is never solely a technical challenge. Organisational maturity, user involvement, clear responsibilities and structural funding prove to be at least as decisive for success.

A conversation starter for practice

To support government organisations in making well-informed choices, the report also includes a practical conversation starter. This helps teams, at the beginning of a scaling trajectory, to jointly map out their objectives, determine which strategy fits best, and understand the implications for collaboration, development and investment.

The report also emphasises that scaling should not become an end in itself. Not every AI application is suitable for scaling, and in some cases, choosing not to scale is the responsible option.

With this exploration, TNO provides concrete guidance for government organisations that want to scale AI in a responsible, effective and future-proof way.

Image header: Sinem Görücü