Project-based AI - Vanden Broele

The challenge

Each legislative term, local governments must draw up a multi-year plan outlining their policy objectives and associated actions. At the same time, they face an increasingly complex subsidy landscape, with hundreds of potential funding opportunities from European, federal, regional, and provincial levels, as well as private sources.

Until now, this often meant a manual, time-consuming search for subsidies that aligned with policy goals. Vanden Broele set out to transform this process, making it faster, smarter, and seamlessly integrated into the existing workflow of local governments. The use of artificial intelligence was a deliberate choice.

The solution: AI with semantic insight

Vanden Broele developed the Subsidiematcher, a feature that uses retrieval augmented generation (RAG) based on Azure AI services, and Axxes consultant Willem Dendauw contributed to the implementation.

The technical architecture includes the following steps:

1) Embedding of subsidy data

Each subsidy in the extensive database is converted into a vector representation (embedding) using the Azure Batch Embedding API. These are then stored in Azure Cognitive Search with semantic search functionality.

2) Semantic matching with policy actions

For each action item in a local government’s multi-year plan, we also generate an embedding. Using semantic queries in Azure, we retrieve the most relevant subsidies for each action, even when there is no exact keyword match.

3) Generative AI for evaluation and explanation

The identified subsidies are then passed to the Azure OpenAI Chat Completions API, which not only assesses whether a subsidy is relevant to an action, but also generates a structured output with a clear rationale for why the match makes sense.

The outcome is a report that not only provides concrete subsidies to local administrators, but also explains why they are relevant, significantly increasing the likelihood of a successful application.

The impact

Subsidiematcher cuts manual work from several days to just minutes. Users instantly receive an actionable, policy-aligned subsidy overview that includes relevant deadlines, requirements, and clear, well-supported explanations. This enables cities and municipalities to manage subsidies more proactively and strategically.

Subsidiematcher has now gone live for all customers and has already been warmly received by several pilot municipalities, including the municipality of Zoutleeuw.

In June 2025, Vanden Broele spoke with Peter Vanmechelen, Finance Director of Zoutleeuw, about Subsidiematcher. Reflecting on his experience with the tool, he said: “One of the strongest points for me was the matching rate. We have more than 300 actions in our multi-year plan, and for about 70 to 75% of them, we found a correct subsidy match. That was impressive. You can immediately see how well a subsidy fits a particular action. Even if the match is only 75% aligned, it is still shown—and I think that’s a smart approach. It helps in assessing relevance. In fact, through a so-called ‘mismatch,’ we even discovered a subsidy we could use for another project. So, unintended matches can also be valuable. I went from disbelief to genuine amazement. Anyone who makes this effort at the start of a policy period saves six years of time and efficiency.”

Read Vanden Broele’s full interview here.

This solution shows how artificial intelligence and semantic technology can deliver real, measurable benefits to local governments: from greater efficiency to increased financial capacity for policy execution.