Digital government teaching case study

Introduction

It was early 2023 when Piet Verstraeten, Innovation Manager at Belgium’s National Employment Office (Rijksdienst voor Arbeidsvoorziening, RVA), stood at a crossroads. After leading a successful proof-of-concept that reimagined how citizens apply for thematic career breaks, he found himself contemplating a much more difficult challenge. The pilot had shown that a user-friendly, AI-enhanced process could make a traditionally bureaucratic service feel almost seamless. But the next step — scaling that innovation across the country — would test not only the technical and operational limits of the RVA, but also its institutional culture, legal adaptability, and public legitimacy.

The project had started with a deceptively simple idea: how could the government make it easier for citizens to take time off to care for their children, support a terminally ill family member, or cope with personal medical needs? But as the pilot evolved, it exposed much deeper systemic questions about how digital tools could shift the citizen-state relationship, how data and algorithms could personalize entitlements without undermining trust, and how a large public agency could adapt to a service philosophy centred on the user rather than the procedure.

Now, Verstraeten had to decide how to move from experimentation to transformation. Could the new system integrate into legacy IT infrastructure? Would staff embrace a digital-first approach? How would the RVA ensure that digitally vulnerable users were not excluded? And perhaps most importantly: how could the agency maintain the ethical high ground while using artificial intelligence to guide decisions?

These were not just implementation details. They represented a broader managerial dilemma — how to preserve the integrity and public mission of a long-standing federal agency while heading it into a new era of data-informed, user-driven service delivery.

Background: The National Employment Office (RVA)

The RVA is a federal public institution responsible for managing unemployment insurance, bridging pensions, reintegration measures, and career breaks. Since its establishment in 1936, the RVA has operated under the Federal Public Service for Employment, Labour and Social Dialogue and employs over 4,000 people across regional offices in Belgium. It plays a critical role in balancing social protection and labor flexibility, with digitalization becoming an increasingly central part of its service strategy.

The major responsibility of the RVA is to administer applications for career interruption, including thematic leaves, such as parental leave, medical assistance leave, and palliative care leave. These forms of career breaks are vital tools for employees to balance work with caregiving or personal needs. Yet, in practice, the process of applying for such breaks had become notoriously complex. In 2022 alone, 553,695 total leave requests were received, of which 341,109 were for thematic leave. Many were incomplete or submitted in error.

Applicants had to interpret dense legal texts, navigate multiple layers of rules, and frequently misunderstood the eligibility requirements. The result was a process prone to error and inefficiency. Internally, the RVA’s Front Office was overwhelmed: each month, it handled 40,000 phone calls and 28,000 emails. Inadequate applications led to a 15% return rate due to missing data, and 10% were rejected due to non-compliance with legal criteria.

This inefficiency imposed a burden on both employees and employers. Workers often faced long delays, frustration, and anxiety over incorrectly filed applications. Employers, who had to verify requests, encountered delays and confusion on their end as well. Internally, the administrative overhead strained the capacity of RVA personnel. While the RVA had tried to improve clarity by adjusting website content and deploying a static chatbot, these incremental changes fell short. The root problem — a lack of personalized, accessible guidance and a complex bureaucratic logic — remained unaddressed.

Launching the Innovation Initiative

Recognizing the limitations of patchwork solutions, Piet Verstraeten took a bolder step. For years, the RVA had made incremental adjustments — rewriting sections of the website, improving legal explanations, and deploying a simple FAQ-based chatbot — but these efforts had not resolved the core frustrations of users. Applicants remained overwhelmed by the volume and opacity of the information, while RVA employees were overburdened with processing errors and responding to redundant inquiries. It became increasingly clear that a more fundamental intervention was needed — one that would reframe the citizen’s experience from the ground up.

In late 2022, Verstraeten joined forces with NIDO, the Belgian federal government’s innovation lab operating under the Federal Public Service for Policy and Support (FOD BOSA). NIDO supports public organizations in tackling complex policy and service design problems by applying human-centred, data-informed, and experimental methodologies. Together, they prepared a proposal for the Federal Innovation Award — a competitive challenge aimed at supporting pioneering ideas in government services. Rather than proposing another layer of static digital information, Verstraeten and NIDO framed the project as an opportunity to radically redesign the career break application journey, leveraging generative AI and UX design to rethink how people interacted with complex administrative rules.

The proposal was selected by NIDO because of its relevance, replicability across government services, and potential to address structural inefficiencies in one of Belgium’s most-used social service channels. The project was named "Loopbaanonderbreking in een vingerknip" (Career Break in a Single Click) to reflect its ambition: to make taking a career break as simple and accessible as ordering a service online. NIDO awarded €30,000 to fund a short-cycle, high-impact proof-of-concept (PoC) that would test a bundled set of innovations.

From the outset, the initiative stood out in its framing. Rather than asking what technology could do, the team asked what citizens actually needed—and then explored how new technology might help meet that need. The challenge was to bridge the gap between regulation-heavy government services and everyday decision-making, turning legal complexity into understandable, actionable guidance. The envisioned solution would need to be not only functional but intuitive, empathetic, and legally accurate.

The main objectives of the PoC were threefold: first, to reduce the keuzestress (choice stress) that citizens experienced when navigating eligibility and application criteria; second, to reduce the operational burden on RVA by lowering error rates and dependency on help desks; and third, to build an adaptable digital framework that could be scaled or reused across similar services.

By early 2023, the project had attracted interest across the Belgian civil service as a potential model for other government agencies struggling to make complex services more citizen-friendly. It also tested NIDO’s model of small-scale experimentation — delivering early insights, avoiding sunk costs, and prioritizing learning and iteration. For Verstraeten, this project became more than a service improvement; it was a test of whether a public agency could truly design around the needs of its users, even when those needs pushed against entrenched procedures and organizational comfort zones.

Designing the Solution: A User-Centered Approach

The project team partnered with Humix, a Belgian UX design agency renowned for its inclusive and data-informed approach to public sector transformation. The agency was selected not only for its technical expertise but also for its track record of navigating the complexity of government processes with a focus on the needs of real users. Together, they designed a three-part intervention that would reimagine the application experience for thematic leave.

First, the team undertook a full UX Application Flow Redesign. This was not just a surface-level revamp, but a foundational shift in how users would interact with the RVA’s services. Using iterative wireframes and low-fidelity prototypes, Humix developed a step-by-step user flow that was intuitive, streamlined, and legally accurate. Each stage was broken down into digestible chunks, with visual cues and progressive disclosure of information. Personalized summaries offered employees a clear starting point, while contextual help at each stage guided them with plain-language explanations. Validation tools checked inputs in real-time, helping users avoid common mistakes before submitting their application. This reduced the cognitive load for users and replaced the traditional form-heavy approach with a dynamic, responsive experience.

Second, the solution introduced AI-Driven Personalized Leave Proposals. Leveraging historical application data, the system would generate three personalized scenarios that matched the applicant’s likely eligibility and preferences. This approach aimed to reduce keuzestress — the stress of making decisions when confronted with too many poorly explained options. Instead of forcing users to navigate complex regulatory logic, the assistant provided concrete, relevant pathways tailored to their situation. The user could compare these proposals side by side, helping them feel more informed and in control. Although in its early stages, this module demonstrated how AI could shift the service from a reactive bureaucracy to a proactive support system.

Third, a Conversational Assistant was embedded into the user journey. This assistant, built on natural language processing and trained with public-facing RVA documentation, served as a round-the-clock digital guide. Unlike traditional chatbots that often deliver pre-programmed, generic answers, this assistant could interpret context and deliver personalized, jargon-free responses. It not only answered questions but also helped users navigate choices, corrected misconceptions, and offered reminders of next steps. During testing, users noted that the assistant felt less transactional and more like a knowledgeable guide. While some remained hesitant to fully trust an AI-powered interface, many appreciated the ability to ask questions on their own terms, without the pressure of time-limited calls or in-person appointments.

Together, these three innovations formed the backbone of a new service philosophy for the RVA—one that centered the user, simplified complexity, and leveraged technology not to automate public service delivery, but to humanize it. The assistant helped users understand requirements, clarified steps, and even corrected misunderstandings during the process. Unlike traditional chatbots, it used context to provide accurate, relevant answers without legalistic jargon.

The solution used a user-centered approach with the following characteristics:

• The early problem framing around user pain points (keuzestress, legal complexity, emotional overload in decision-making)

• The collaborative prototyping process with Humix, which involved rapid iteration based on real user feedback

• Simplification through modular flows, contextual help, and progressive disclosure

• The assistant and proposal engine are examples of anticipatory service design, not just reactive design

• The commitment to accessible language, and the plan (albeit not yet implemented) for multilingual support

Pilot Outcomes and User Reactions

The PoC ran over a three-month period during the summer of 2024. Usability testing revealed strong satisfaction among users, who responded positively to the more intuitive, accessible, and guided experience provided by the redesigned application interface. Many applicants reported that, for the first time, they felt confident in navigating a public service process without needing to consult dense legal documents or seek external help. The streamlined user flow allowed them to progress step-by-step, supported by contextual guidance and pre-filled fields. The AI-generated leave suggestions also gave them clarity by narrowing the decision set to three plausible options tailored to their circumstances. For many, this eliminated guesswork and reduced the stress associated with making the "right choice."

Employers, who play a key role in validating leave requests, observed that the new system significantly reduced confusion among their staff. Fewer clarification emails were sent, and HR departments spent less time walking employees through basic procedural steps. On the institutional side, the RVA noted a measurable reduction in the number of incoming inquiries to its Front Office. Early analytics indicated a drop in email volume and call center load, confirming the system’s potential to alleviate pressure on administrative services.

Quantitatively, completion rates for thematic leave applications improved, and the average time required to finalize an application decreased. Error rates also declined, with fewer submissions needing to be returned or followed up due to missing or incorrect information. In parallel, the conversational assistant proved to be highly effective: during testing, it was able to resolve over 80% of user queries without human intervention, offering real-time answers in clear, non-technical language.

Nonetheless, several important areas for improvement emerged. Some users expressed hesitation about interacting with the AI assistant, equating it with earlier generations of underperforming chatbots. Although the new tool was more sophisticated, its identity as a machine still prompted caution among more skeptical users. Another frequently mentioned gap was the lack of financial transparency. While the assistant and the application interface offered clarity about procedural requirements, users wanted detailed estimates of how different leave options would affect their net income, taxes, and pension rights. The system did not yet integrate this kind of simulation, and for some users, this lack of foresight undermined the utility of the recommendations.

Finally, the pilot highlighted digital inequality as a significant concern. For users with limited digital literacy or less frequent access to digital services, the tool required complementary human support. Although the design was meant to be intuitive, the assumption of a baseline digital fluency created a barrier for certain groups. As such, any nationwide rollout would need to include hybrid models that maintain accessible channels—such as local offices, telephone support, or guided walkthroughs —t o ensure that no one was left behind.

Ethical and Operational Considerations

From an organizational perspective, Verstraeten had to ensure compliance with the General Data Protection Regulation (GDPR)  and public procurement rules, maintain accessibility across Belgium’s three national languages, and manage expectations internally. The design and execution of the project also had to align with the broader objectives of digital transformation in government — efficiency, inclusivity, and user trust — without disrupting core service obligations. Integration into legacy IT systems, resource allocation for technical maintenance, and defining ownership over AI-driven components were key practical issues that required careful orchestration between IT, legal, and operational units within RVA.

The project introduced a new way of working for civil servants and required training to help staff understand how AI could support, rather than replace, their responsibilities. Operational workflows had to be redesigned to ensure that manual interventions complemented automated guidance, especially in cases where edge scenarios fell outside the standard AI logic. Additionally, the multilingual context of Belgium demanded that all features — including the assistant — eventually be made available in Dutch, French, and German, without loss of nuance or clarity.

Ethically, the use of AI raised questions about transparency, decision accountability, and data privacy. Verstraeten addressed these by ensuring that no automated decisions were made — AI was used purely to support human choices — and by maintaining strict anonymization in data processing during testing. The system provided interpretive assistance rather than taking autonomous action, respecting legal standards around public service decision-making.

Importantly, the design emphasized user autonomy and control. Users could review and modify all proposed options, with a clear overview of what data was being used. At no point were users forced into algorithmic recommendations; the suggestions were just that — suggestions — and could be fully overridden. Feedback mechanisms were integrated to collect user sentiment and detect misuse or confusion in real time. In line with privacy-by-design principles, the conversational assistant was trained exclusively on publicly available information, and any future integration with personal data would be subject to explicit user consent and secure authentication via national ID systems.

These considerations made clear that innovation in the public sector is not just a technical issue but a deeply institutional one. Ensuring that digital tools align with public values, regulatory frameworks, and user expectations is a precondition for long-term legitimacy and adoption.

The Scaling Challenge: The Public Servant’s Dilemma 

By the end of 2024, Verstraeten faced a defining moment. The pilot project had demonstrated clear value: users felt more empowered, employers were less burdened, and the RVA was seeing early gains in efficiency. Yet the stakes were rising. What had started as a targeted proof-of-concept, underwritten with a modest innovation budget, was now positioned to challenge deep-rooted public service processes. Could this innovation scale? And if so, how could it be done without compromising legal safeguards, organizational stability, or user trust?

Verstraeten understood that institutional alignment would be critical. Internally, the RVA consisted of siloed departments with their legacy systems, workflows, and risk appetites. While the pilot had succeeded with a small cross-functional team, scaling would require formal buy-in from leadership, new interdepartmental coordination structures, and a long-term funding model. Many RVA staff remained cautious. Would the conversational assistant add complexity or reduce their role? Would a new digital-first process increase pressure on frontline staff to support digitally excluded citizens?

The IT dimension raised equally urgent concerns. The assistant and application interface would need to be extended to support Belgium’s three national languages, without losing nuance or reliability. Integration with authentication tools like Itsme and e-ID had to be scaled securely. The platform also needed to anticipate future regulatory changes, including eligibility rules that could shift with new political coalitions. Verstraeten also had to consider how to keep the AI logic current and trustworthy. Who would be responsible for maintaining its accuracy, and how often would it need to be retrained as new policies or user behaviors emerged?

Beyond the operational realm, policy challenges loomed. Verstraeten began asking a more fundamental question: should the process of requesting thematic leave be treated as a standalone transaction, or as part of a broader ecosystem of digital rights and entitlements? Could users be offered simulations that show not just eligibility, but income impact, tax implications, and pension effects? Doing so would require collaboration with other federal agencies — FPS Finance, FPS Pensions — and potentially new legislative frameworks.

At the heart of his dilemma lay a cultural question: could citizens be persuaded to trust a system that leverages their data to simplify bureaucracy, rather than obscure it? And could civil servants shift from being gatekeepers of procedure to facilitators of user-driven journeys? The assistant was designed to support autonomy, but perception mattered. If users feared surveillance, data misuse, or automated decisions, adoption would stall. Verstraeten also worried about the ‘AI fatigue’ prevalent in public discourse — would the project be misunderstood or politicized?

Each of these questions pointed to different risks and levers: organizational readiness, technological maturity, legal resilience, political support, and societal trust. None of them could be resolved by Verstraeten alone, yet his leadership was pivotal in orchestrating the next steps. He needed to formulate a scalable governance model, build a coalition of internal and external champions, and articulate a roadmap that translated vision into operational realism.

As he reviewed the final feedback report from the PoC, he realized this was not just a decision point — it was a test of public innovation leadership. Would the RVA stay confined to legacy systems and incremental change, or would it seize the moment to define a new standard for digital public service?

As the proof-of-concept wrapped up in early 2023, momentum was building. User feedback was positive: citizens navigating the career break application for the first time reported less stress, fewer mistakes, and greater confidence in their eligibility choices. Employers appreciated a drop in routine queries from staff, and internal RVA teams began noticing a modest reduction in front-office workload. The redesigned digital journey — combining step-by-step UX flows, personalized leave suggestions, and an AI-powered assistant — had struck a chord.

Yet the initiative stood at a crossroads. The three-month pilot had been funded as an isolated innovation experiment, not as a blueprint for nationwide adoption. Key questions loomed. Could this solution be embedded in the RVA’s core systems and extended to Belgium’s multilingual population? Would legal teams endorse the assistant’s interpretive responses as compliant with complex regulations? More fundamentally, would senior leadership support a transition from process-driven service delivery to user-led design logic?

For Piet Verstraeten, the pilot's lead and Innovation Manager at the RVA, the path ahead was uncertain. His small, cross-functional team had proven that change was possible — but scaling up meant shifting from experimentation to execution. It would require not just more funding, but deeper institutional commitment and a willingness to rethink long-standing assumptions about how public services should operate. A promising prototype now faced the harder test: transformation.

Conclusion

As Piet Verstraeten reflected on the three-month pilot, the results were encouraging — improved user satisfaction, reduced error rates, and a glimpse into a more humane, intelligent public service experience. Yet for all its promise, the prototype had only scratched the surface of what a full-scale transformation would entail. The case left him — and the RVA — with a critical set of unresolved questions.

Would institutional culture, long shaped by legal precision and risk aversion, accommodate a user-led design philosophy rooted in empathy, iteration, and digital fluency? Could a conversational assistant, powered by AI but framed as a human-centered tool, earn the trust of both citizens and civil servants in an era increasingly skeptical of algorithmic authority?

Operationally, Verstraeten faced choices about governance, funding, and technology architecture. Should the RVA develop internal AI expertise or outsource maintenance to external partners? How would they ensure accountability as the assistant evolved and interacted with new user segments? Would success require centralization of innovation capacity, or a distributed, agile model across departments?

Moreover, the case forced a deeper policy reflection. Was this digital assistant merely a better interface for an outdated system, or a gateway to a more integrated model of citizen entitlements — one that spanned across agencies and offered proactive guidance on life events? Could such a transformation preserve individual agency while also reducing the administrative burden on users?

The pilot had reframed a technical project into a strategic inflection point. Verstraeten’s challenge was no longer about scaling a solution, but about shaping a broader narrative: “What kind of public service do we want in the age of digital intelligence — and who gets to design it?”