The road to AI maturity
In June 2025, representatives from 17 parliaments gathered in the Senate of the Netherlands in The Hague to discuss the journeys they were taking towards AI maturity. The meeting, convened by the Centre for Innovation in Parliament’s (CIP) Parliamentary Data Science (PDS) Hub, highlighted how parliaments worldwide were moving beyond AI experimentation and towards strategic implementation. The discussions revealed that AI adoption in democratic institutions was not about the technology itself, but rather about solving real parliamentary problems.
The most successful parliamentary AI initiatives share a common characteristic: they start with specific business challenges rather than technological capabilities. One participant noted that parliaments needed to be “business-focused, looking at where AI adds the most value”. This mindset represents a significant shift from the technology-driven approaches that dominated early AI adoption.
The UK Parliament’s capability-based planning exemplifies this approach, with the institution using enterprise architecture heat mapping to identify high-potential areas for AI implementation. Rather than deploying AI everywhere, the legislature treats it as “just another new technology”, applying AI strategically where it can deliver measurable improvements in parliamentary operations.
Perhaps the most practical insight to emerge from the discussions was the “golden rules” approach to AI governance. Instead of calling for comprehensive frameworks that take months to develop, this approach advocates for starting with between 5 and 10 basic guidelines written in accessible language, then further developing these guidelines on the basis of real-world experience.
This iterative approach acknowledges the critical reality that parliamentary staff are often already using AI tools before formal policies exist. In response to substantial “bring your own AI” adoption, the House of Commons of Canada has rolled out a “toddler strategy”, under which it provides warnings and teaches safe AI practices while allowing exploration without shutting down access to AI tools. This approach has proven more effective than introducing restrictive policies.
Traditional waterfall procurement models – where sequential project phases are completed one after another with no overlap – are incompatible with AI development cycles. Requirements often only crystallize when a pilot project is 90% complete, forcing parliamentary procurement teams to embrace agile approaches. The Senate of Spain uses framework contracts to manage situations where requirements cannot be fully defined at the outset of a project.
Parliaments must accept that bias is inevitable in AI systems. Institutions that have successfully adopted the technology focus on identifying and managing acceptable bias levels rather that pursuing bias-free solutions in vain. This pragmatic approach enables progress while maintaining democratic accountability.
Every successful parliamentary AI implementation begins with data quality. Poor data hygiene undermines the effectiveness of AI more than any other factor: indeed, as participants consistently noted, effective AI cannot be built on poor-quality data. To solve this, institutions need to prioritize data governance before major AI investment. The comprehensive data governance approach adopted by the Chamber of Deputies of Brazil serves as an example of how data stewardship becomes crucial as AI adoption scales.
Parliamentary AI adoption demands competencies across three distinct layers: technical infrastructure, business transformation and individual AI literacy. The most significant challenge is not technical but cultural in nature. Traditional technology adoption patterns are reversing, with end users demanding AI integration while IT departments resist owing to security concerns.
Successful institutions are addressing this challenge through cross-functional partnerships between human resources, which owns training responsibility but lacks technical capacity, and IT, which has expertise but possesses limited delivery capabilities. Such collaborative approaches combine pedagogical expertise with technical knowledge.
During the meeting, participants explored a draft AI maturity framework encompassing five levels: foundation, experimentation, strategic implementation, integration and innovation leadership. This framework serves as a road map and an evaluation tool, helping institutions understand their current position and logical progression steps. The CIP is now taking the development of this framework forward with the support of PDS Hub members.
The consensus that emerged from meeting in The Hague was clear: while AI is a genuinely disruptive technology, it can still be managed through established governance frameworks. AI’s true disruptive power lies not in its ability to overturn democratic processes, but rather in its potential to make parliaments more efficient and effective.
