Read in Dutch
Explainable AI as the foundation of an open government
Introduction
The Open Government Act (Wet open overheid - Woo) has now been in force for four years and establishes the public's right to access government information. Its purpose is to ensure that government operates transparently, both in its processes and in its decision-making. This requires information to be properly captured from the outset, with decisions fully traceable and the reasoning behind them clearly documented. Not because someone may request that information at a later stage, but because openness is embedded by design in the way government works. In a truly effective open government, many Woo requests would simply become unnecessary not because less information is made available, but because trust is proactively built into the system itself.
That is the spirit of the Woo. The Act is not fundamentally about files or statutory deadlines; it is about public values: transparency, democratic accountability and trust between government and society. Citizens, journalists and oversight bodies are not asking for documents as an end in themselves they are seeking clear, understandable answers. They want to know how decisions were reached, why particular choices were made, and on what basis.
At the same time, this promise is under increasing pressure. In its 16 February 2026 policy letter on AI and the Woo, the Dutch government outlines how technology can support better implementation of the Act. The letter also makes clear that many current initiatives are still in their infancy and that meaningful scaling will require stronger choices around collaboration, governance and the design of information management. The political and societal urgency is evident, but the challenge extends well beyond policy or technology alone.
The fundamental question is therefore not whether government wants to be transparent, but whether it is capable of following, explaining and reproducing its own decision-making processes. As long as transparency remains primarily a retrospective exercise requiring fragmented information to be reconstructed after the fact public trust will remain fragile.
This approach must change. And it must change quickly.
Mapping the challenges
Increasing pressure on transparency and public trust
Implementing the Woo creates persistently high costs, lengthy processing times and growing societal friction. Dutch ministries take an average of 143–188 days to respond to a Woo request, despite the statutory deadline of 42 days. With approximately 2,000 requests each year, requiring 30 to 60 hours of work per request, direct implementation costs are estimated at €5–10 million annually, excluding penalty payments and other associated costs.
The root cause lies not in the legislation itself, but in fragmented information management and heavily manual processes. For journalists and civil society organisations, this hampers their ability to perform their democratic scrutiny role. For public sector organisations, it creates a structurally high administrative burden and an overreliance on individual knowledge. Senior officials and political leaders increasingly find themselves having to explain decisions retrospectively, based on incomplete or dispersed information.
This pressure has a direct impact on public trust. Responses often arrive too late, provide only limited information or fail to address the core of the question. What may be legally compliant often feels inadequate in practice. The result is a vicious cycle of follow-up questions, increasing numbers of Woo requests and ever-greater pressure on public organisations.
The Woo does not create these problems it exposes them. It reveals that transparency and traceability are still insufficiently embedded in the way many organisations manage information and organise decision-making.
Current initiatives: logical, but insufficient
It is entirely understandable that significant investment has been made in technological solutions over recent years. AI applications for search, classification and anonymisation promise faster processing and reduced workloads. Many organisations have launched pilot projects aimed at addressing specific bottlenecks, and these initiatives are both logical and often technically successful.
However, most prove difficult to scale. Pilot projects improve individual steps within the process but do not fundamentally change the process itself. They address symptoms where pressure is greatest, while leaving the underlying foundations untouched. As a result, organisations accelerate a way of working that still lacks proper control over its information.
The experience of the Platform Open Government Information (PLOOI) illustrates this pattern. The ambition of creating a single central access point for government information was both logical and socially desirable. In practice, however, centralised publication proved unworkable without consistent information capture, standardised metadata and clear ownership arrangements at the source. The lesson is clear: discoverability does not begin with a platform it begins with control over information at its source.
The core problem: a lack of control
Beneath the operational pressure, the limited scalability of current initiatives and the growing number of Woo requests lies one structural issue: many organisations are not sufficiently in control of their information and decision-making.
Decisions are made across multiple systems, files and more frequently in informal communication channels. Information is dispersed, context remains implicit and documentation is inconsistent. As a result, the same question can produce different answers depending on who searches, where they search and how they interpret the available information. Traceability is lacking: it is difficult to reconstruct how a decision was reached, which considerations were weighed and which information informed the outcome.
Transparency therefore becomes a retrospective recovery exercise. Every new request requires information to be searched for, interpreted and assessed all over again. AI can accelerate this process, but as long as the underlying problem remains unresolved, faster processing merely amounts to treating the symptoms.
Organisations that focus primarily on speeding up request handling, without first establishing explainability and traceability, simply defer the problem while perpetuating this fundamental vulnerability.
Our vision
Sopra Steria views the Woo not as a document management process, but as a public value challenge. Trust is not earned through the publication of individual documents, but through information that is discoverable, traceable and explainable. Achieving this requires a fundamentally different approach. Rather than placing documents at the centre, the focus should shift to decision-making itself: how choices are made, which considerations are taken into account and what information underpins them.
As long as openness is organised around documents and requests, it will inevitably remain reactive. Organisations that seek to improve transparency in a sustainable way must embed it within their processes, data and governance. Openness then ceases to be the end point of a request and instead becomes an inherent characteristic of the information ecosystem itself.
From reactive accountability to proactive transparency
This shift requires a fundamental change in approach. Rather than reconstructing what happened after the event, organisations should create the ability to provide meaningful insight from the outset. By understanding which decisions, themes and information consistently generate public interest, transparency can be delivered in a far more targeted and proactive way.
Proactive transparency does not mean publishing everything. It means making the information that matters accessible in a clear and understandable way. This not only reduces the pressure on Woo processes, but also strengthens public dialogue. Citizens and journalists gain context rather than isolated fragments of information. Trust is no longer secured through procedural compliance alone, but built through meaningful insight.
Raising the bar for AI: explainability and traceability first
AI has an important role to play in this transformation but it is not an end in itself. Its real value lies in accelerating scalable traceability. For that reason, government must set higher standards for how AI is designed and deployed.
AI systems that support transparency must do more than deliver speed or efficiency. They must be explainable, auditable and governable. They should make it clear what the system does, why it does it, and on which information its outputs are based. Only then can AI strengthen confidence in public decision-making rather than undermine it.
Good intentions or technical performance alone are not enough. Organisations deploying AI in this domain must design for governance, quality and accountability from the very beginning.
Openness and privacy as a design challenge
Transparency and privacy are often portrayed as competing interests. In reality, they are two sides of the same design challenge. Responsible disclosure is only possible when decisions, considerations and justifications are explicitly recorded and remain traceable.
When decision-making, contextual information and the grounds for exemptions are properly documented and traceable, privacy protection becomes stronger rather than weaker. Not because less information is disclosed, but because government is better able to explain why information is or is not made public.
Reliability is a choice
This vision is ambitious, yet entirely achievable. The challenges surrounding the Woo are not inevitable; they are the result of earlier design choices. Different choices can therefore produce different outcomes.
The Woo currently costs the Dutch government millions of euros every year, largely due to avoidable inefficiencies. AI only delivers lasting value when it forms part of a structural approach centred on information quality, reusability and explainability.
Organisations that treat transparency as an inherent characteristic of decision-making, adopt traceability and explainability as core design principles, and deploy AI deliberately as an accelerator can strengthen trust and reliability in a sustainable way.
In other words: this can be achieved provided it is designed properly from the outset.
Sustainable Management of the Woo
Intended Outcomes: From Reactive Processing to Proactive Transparency
The greatest potential of AI does not lie in processing Woo requests more quickly, but in reducing the need for them altogether. By proactively making frequently requested information available in a clear, understandable and consistent manner based on patterns identified in previous requests three tangible benefits can be realised:
- Shorter processing times and fewer appeals, through improved discoverability and more consistent decision-making.
- Lower administrative burden through reuse, as frequently requested information is proactively published in a consistent format.
- Greater public trust through traceability, enabling citizens to understand which information underpins decisions and how key considerations were reached.
End-to-end: solving the whole problem instead of fighting fires
The objective is not to accelerate a single step in the Woo process, but to redesign the underlying value chain from information capture at source, through discoverability and decision-making, to public accountability.
This means looking beyond simply "searching faster" or "redacting faster". Instead, the focus should be on creating the conditions that make decisions inherently traceable and explainable. Only then does openness become an intrinsic characteristic of information management rather than the end result of responding to a request.
Achieving this requires a mature information management capability in which key information is systematically captured through case management systems, metadata, audit logs and event records. Documents become the human-readable representation used to explain both the decision itself and the process behind it to citizens and businesses.
Without this foundation, every intervention including the use of AI remains little more than symptom management.
AI as an accelerator when used responsibly
AI can make the implementation of the Woo more manageable by supporting knowledge discovery, document classification and anonymisation. However, acceleration only delivers value when AI forms part of a responsible design and operating model rather than being deployed as a standalone tool.
Careful implementation is therefore essential. In practice, this requires robust governance, frameworks for responsible AI design, investment in AI literacy and skills, and above all a mature information management foundation built on trusted data quality and metadata.
Key elements include:
- Clear governance: explicit roles and responsibilities, design and usage frameworks, and integration into policies and operational processes.
- Continuous monitoring: oversight of system use and performance, including the identification of anomalies, supported by regular maintenance to refine models, prompts and working methods.
- Traceable audit trails: not designed for openness in themselves, but structured so that actions and decisions can always be reconstructed afterwards, providing the foundation for transparency and accountability.
- Human-in-the-loop decision-making: processes and interfaces that demonstrably support professionals in making decisions, while providing visibility into both human judgement and AI-assisted actions.
- The EU AI Act reinforces this approach. AI applications used within the public sector and particularly those supporting Woo requests are likely to fall within a high-risk category, requiring demonstrable governance and risk controls. Even where systems fall outside this category, transparency and explainability remain essential whenever AI affects citizens' access to information or their information rights.
Proven open-source components: modular, scalable and reusable
To scale reliably across different public sector organisations, solutions should be designed as modular architectures built from proven, reusable and, wherever possible, open-source components. Centrally managed building blocks enable organisations to address common challenges more quickly and consistently, avoiding the need to reinvent solutions for each implementation.
Our Advanced Search & Retrieval solution explained
A practical example of this modular building-block approach is our Advanced Search & Retrieval solution. It enables semantic search searching by meaning rather than relying solely on exact keywords. This provides a faster and more comprehensive view of the information available on a particular topic, while significantly reducing the risk of overlooking relevant information due to differences in terminology.
The solution can be used to support the handling of Woo requests, but equally serves as a means of strengthening information management in a more structural and sustainable way.
Its value lies not only in the technology itself, but also in a number of deliberate design choices that ensure both effectiveness and governance:
- Continuous evaluation and optimisation through automated validation. Different configurations and large language models can be compared, evaluated and optimised automatically, demonstrating which combinations consistently produce the best answers to frequently asked questions. This also enables organisations to revisit those choices regularly for example, when new frequently asked questions emerge or when a new language model becomes available. The result is a sustainable solution in which technology choices and dependencies can be reviewed and justified over time. In research conducted by Sopra Steria in 2025, we demonstrated that configurations such as hybrid search combining traditional keyword search with semantic search can be systematically optimised for maximum performance.
- Centralised search across data silos. Organisations can choose how individual data silos are connected. Where information is not yet searchable, it can be ingested into the solution. Where search capabilities already exist within the source system, a federated search approach can be adopted, allowing data to remain in its original location while AI is layered on top of existing search capabilities. This simplifies governance because users retain access only to the information they are authorised to view, while still benefiting from the capabilities of generative AI. Regardless of how data sources are connected, users can perform consistent, organisation-wide searches across all information they are permitted to access.
Our solution is built as a modular architecture using open-source technologies and can be deployed in virtually any environment: public cloud (such as Microsoft Azure), private cloud (for example, the infrastructure used by the Dutch National Police), or on-premises, whether on a laptop, in a vehicle, on a local server or within a data centre.
You only deploy the components you actually need. Existing capabilities such as platforms, identity and access management, language models, storage solutions or data repositories can remain in place, with our components complementing rather than replacing them.
Together with our clients, we define how the performance of the overall solution should be measured and optimise the configuration accordingly, ensuring its effectiveness can be objectively demonstrated. User testing then validates these results in operational practice. This is how Sopra Steria works alongside government organisations to demonstrate that its AI solutions deliver measurable value and perform reliably in real-world environments.
From ambition to delivery
The coming years will be decisive in shaping the role of AI in the implementation of the Woo. This is the moment for public sector organisations to make choices that are not only effective and scalable today, but remain explainable, maintainable and trustworthy tomorrow. Sopra Steria Netherlands supports government organisations throughout this journey from strategic vision to successful implementation and large-scale adoption.
Sopra Steria believes that governments should apply more demanding assessment criteria when evaluating AI solutions, particularly in relation to modularity, reliability and scalability. These expectations are both realistic and achievable. Yet we continue to see pilot projects launched that demonstrably fall short of these standards. Organisations should also require clear evidence that a pilot delivers genuine added value. Launching AI pilots without establishing robust evaluation criteria represents an inefficient use of public resources.
Drawing on extensive expertise across government and other critical sectors, Sopra Steria recognised some time ago the need for an alternative to both large technology vendors where greater strategic independence is increasingly important and early-stage start-ups, where long-term scalability may remain uncertain. For this reason, Sopra Steria has invested for many years in developing its own Advanced Search & Retrieval platform: a solution built on open-source components, deployable across a wide range of infrastructures, and specifically designed for responsible, reliable and scalable AI.
The platform supports not only the operational delivery of the Woo for example, by enabling information to be found, interpreted and assessed more quickly but also strengthens information management at its foundations, allowing openness to become an inherent capability rather than something achieved only in response to individual requests.