The Flemish government wants a lean but high-performing administration and hopes to achieve this by leveraging AI. The intended savings are not impossible, but they are certainly not a given.
The Flemish government is cutting costs across numerous sectors, including the government apparatus itself. By moving toward a “lean government,” the Flemish administration aims to save 155 million euros. This will be achieved by reducing the workforce. Only one in three departing civil servants will be replaced.
Doing more (or the same) with less
However, it is also the ambition of the Diependaele government to maintain a high-performing Flemish administration. Doing more with less will, according to the minister-president and his ministers, be made possible by increasing the use of AI.
This is an ambition that, even at this early and vague stage, deserves some analysis. The letters A and I seem to possess an almost magical power today. Anyone looking to achieve something with AI can already count themselves rich simply by voicing those plans out loud.
The reality is quite different. AI is a broad umbrella term for a multitude of technologies. We assume the Flemish government is referring to generative AI and its applications. Even then, there are various ways to deploy AI.
The limitations of a copilot
The first is the simplest: giving every civil servant an AI assistant. This is a strategy that can certainly provide added value, although there is a consensus in the tech sector today that the value of an AI copilot is far from unlimited. The efficiency of individual employees may increase slightly. Emails are drafted more smoothly and documents interpreted faster.
That is useful, but it quickly hits its limits. Think of ten minutes of time saved here, and another fifteen minutes there: not enough to suddenly take on the workload of a retired colleague.
Deep implementation in government systems
True value from AI comes from deep implementation. AI systems must then be linked to a strong data foundation, with integrated IT systems. AI is then the icing on the data cake. If no unified data platform exists with compatible databases and clear semantics, AI cannot work miracles.
In other words: when data is partly in a SaaS solution, partly in a local legacy database, and partly in an Excel sheet that Bart from the Department of Education and Training has been meticulously filling out for five years, it is not yet time to talk about AI.
The good news is that Flanders is certainly not the worst student in the European class. Years of digitalization have already clearly borne fruit. Take the Flemish Citizen Profile, for instance, where citizens can already view and manage many matters from a central platform. There is indeed evidence of digital collaboration among Flemish government services.
No AI button
That does not mean there is now a big ‘AI’ button that Diependaele can press to suddenly realize savings. The rollout of AI capabilities is a complex IT project where, alongside data quality and a solid back-end, a clear project vision coupled with measurable parameters is essential.

And that’s not even mentioning the people: you have to learn how to work with AI tools. They can increase an employee’s capabilities, provided that employee knows exactly what AI’s strengths are and what its inherent weaknesses are. Without the necessary training, every digital project is doomed to fail, AI included.
Concrete figures show hurdles
This is not mere doom-mongering. Capgemini published a report on data and AI in government just this week. That study, based on a survey of 600 executives in sixteen European countries (excluding Belgium), suggests that implementing AI is not so simple after all. Barely 28 percent of government organizations indicate they are effectively succeeding in scaling AI initiatives.
On the other hand, the budget going toward AI is increasing. Capgemini speaks of seven percent of the annual IT budget by 2027.
Illusion of quick returns
Furthermore, the fact that AI does not (immediately) pay off is not due to the government context. This is made clear by research from PwC among 4,000 CEOs in 95 countries earlier this year. Those figures show that AI has effectively delivered savings and revenue improvements for barely twelve percent of the business leaders surveyed.
Mind you: the belief in the role of AI remains, but the realization is growing that AI, like all other new technologies, needs time to take root within an organization. Investment and patience therefore precede savings and results.
Nuance from experts
The urge to ‘do something with AI’ is strong, not only in government but in companies worldwide. Strategic insight and measurable KPIs are not always part of the package. “We often have to tell customers to tap the brakes for a moment,” said Steven Nuyts, Head of BTP Solution Advisory BeNeLux at SAP, in December regarding this situation during a roundtable discussion organized by ITdaily. “First, look at what levels of automation you already have. GenAI is not the solution for everything. We talk a lot about productivity, but there is still too little focus on revenue growth and truly tangible results.”
The number of tokens you need to match a human will ensure that you don’t necessarily come out cheaper by embracing AI.
Joris Dresselaers, Business Area Director Visma
“I notice that we often even overestimate the level of maturity and literacy in terms of data among business users,” noted Fen Lasseel, Managing Director of Datashift, in that same conversation. “We still need to bridge the gap to business users more and help them extract sustainable value from data.”
Magical savings fail to materialize
These insights are not groundbreaking, but they stand in stark contrast to communication where ‘more AI’ is suddenly supposed to help close a budget gap in the short term. Not a single expert we have spoken to about AI over the past year disputes the technology’s capacity to enable efficiency gains. However, they all note that AI is not a magical cost-saving technology.

More recently, Gianni Cooreman, Senior Director of Agentic Transformation at Salesforce, noted that companies in the context of AI have “an excessive focus on pure efficiency and individual productivity.” That statement also carries additional weight in the context of the Flemish government’s September declaration.
AI also costs money
Finally, we note that AI also costs money, and quite a lot at that. Those who consume AI via the cloud are faced with a combination of subscription costs per user for certain applications (such as a copilot), or costs per token for others (such as automated smart workflows). These can add up quickly. Without a clear strategy, measurable goals, and clear cost management, there is even a risk that the AI bill will turn out higher than the saved personnel costs.
“The number of tokens you need to match a human will ensure that you don’t necessarily come out cheaper by embracing AI,” said Joris Dresselaers, Business Area Director at Visma, in the previously mentioned expert panel.
“The question is whether Large Language Models (LLMs) are always the right solution. Using those models is very costly,” noted Véronique Van Vlasselaer, Head of AI & Data Science at SAS EMEA, at the time. One wonders if Diependaele has asked himself that same question.
Major challenge
The Flemish government is therefore facing a major challenge: it must achieve efficiency gains in the short term with the help of AI and do more with less. Reports, figures, and experts all suggest that such a thing is possible, but not as simple as it seems at first glance for many organizations.
In summary, ‘AI’ is not something that is just lying there waiting to be picked up to achieve efficiency and savings. AI is a new and powerful digital technology with a lot of potential, provided it is rolled out within the framework of well-thought-out projects, inside an organization with a strong digital foundation and high data quality.
If Flanders wants an AI-driven government that truly offers more than today but with less, savings will only follow if investments are made first. Furthermore, Flanders will have to plan and steer its AI projects better than other European governments and large organizations are doing today if it wants them to be successful quickly.
