Market consultation Audit Vlaanderen: AI to support auditors

Who are we looking for?

Audit Vlaanderen (AV) carries out both regular (based on themes, organisations & processes) and forensic audits for the Flemish government and the Flemish local authorities.

AV's audit process involves several steps, from collecting and processing structured and unstructured data (such as text documents, spreadsheets, photos, emails with attachments, etc.) to searching through this data, gaining insights and creating a final report that backs up all conclusions with the relevant documents & pieces of collected data or information.

In addition, the auditors themselves are responsible for a lot of time-consuming audit work:  the preparatory research, the retrieval of information and the follow-up of its delivery, data collection, cleaning and analysis, the analysis of the collected information and the concise reporting of the most relevant findings. 

AV wants to improve its way of working with innovative technologies such as artificial intelligence (AI), machine learning (ML) and/or robotic process automation (RPA). The goal is twofold: to make the work of auditors more efficient and to further increase the quality of audits. 
The most promising opportunities are situated within three areas, namely AI Assist, AI Detect and AI Generate.

For this project, intelligent datamining for audit processes, we thus primarily address companies and (research) organizations with expertise in one or more of these three areas in order to take hands-on steps forward together:

1. AI Assist: (Intelligent) retrieval & transformations with natural language processing (NLP): technologies to facilitate and follow up on the retrieval of information from all parties involved and/or to help process received information in various ways (e.g. storing files in the right location, adding metadata, extracting elements of the content, categorizing content, etc.) and/or to facilitate the administration of audits
2. AI Detect: Search with NLP and/or data mining & pattern/anomaly detection:  intelligent technologies to efficiently search, structure and interpret large volumes of unstructured data (such as documents, emails, statements, etc.) and/or for the automatic detection of risks, patterns or irregularities in large datasets
3. AI Generate: (Generative) artificial intelligence: use of generative (language) models to create textual and/or visual output or to improve conclusions and/or texts, to support (mainly) the reporting phase

The three sub-areas in more detail:

1: AI Assist

This area focuses on the use of AI as a supporting and facilitating technology within the audit process. By using AI-based assistance systems, such as virtual chatbots and agents, intelligent processing, and contextual recommendations/coaching, the goal is to automate repetitive tasks, handle all received information efficiently, and thus access relevant information faster. Within this area, we expect techniques such as information retrieval, natural language understanding (NLU) and other techniques for artificial intelligence to play a central role.

Purely illustrative, one of many possible examples: In order to evaluate whether subsidy regulations of a local government contain provisions to limit the risk of irresponsible subsidy payments, subsidy regulations must be retrieved from the website of the local government as well as from a number of discussion partners (who occasionally need reminders and sometimes ask for extra time to respond).  These are then stored with metadata to organize them according to the respective policy fields. (Support can be comprehensive but can also be set up for just one or a few aspects. That can also grow if necessary.)

2: AI Detect

In this area, the focus is on intelligent searching, understanding and structuring of the content of large volumes of unstructured data (such as documents, emails, interview reports, etc.) and on analytical intelligence: the application of AI to find information from a multitude of documents/data and automatically detect patterns, anomalies or relevant relationships in large and diverse datasets. This concerns both structured data sources (such as transaction lists or tables) and unstructured documents (such as reports, statements or emails). We expect technologies such as anomaly detection, unsupervised learning, statistical modeling and NLP to form the core of the added value here.

Purely illustrative, one of many possible examples with regard to intelligent searching, understanding and structuring: In the course of the audit, it may suddenly become apparent that there is still a relevant element that did not seem relevant before. AI can help the auditor by indicating which information about that element is reflected in which documents and by summarizing them.

3: AI Generate

This area focuses on the use of generative AI technologies, such as large language models (LLMs), for the creation of new content based on existing data. Within the audit context, this provides valuable support in structuring information, preparing reports, and generating useful visualizations. 

Purely illustrative, one of many possible examples: As an auditor, I can use generative AI to rewrite texts/reports to a consistent or more concise style/format, so that all reports (for dissemination to different channels) meet expectations in terms of quality (to-the-point, high readability and consistent style).
 
Good to know:

• Flexibility is more important than one-off setup: Almost every audit focuses on a different process within a different organization & with different data.
• The security of the information being processed is crucial and there is little room for the management of the solution(s) within our relatively small organization. As a reflection w.r.t. the feasibility of any proposed solutions, it is pointed out that Microsoft Copilot is currently being rolled out within the Flemish government. This does not mean that other technologies cannot be considered, but it does mean that the creation of a secure context with professional management and active technical follow-up as well as the licensing costs for Copilot are already covered but would have to be factored into the proposals for other technologies.

Purpose of the market consultation

Based on workshops with the auditors of AV, opportunities were listed and prioritized in the form of use cases. From this, use cases were then selected for which AI may be able to provide support.

Through this market consultation, we want to examine the extent to which the proposed use cases are technically feasible, which relevant solutions may already exist, which risks and preconditions are relevant, which cost price framework is realistic for the respective business cases and how we can best concretize our needs and translate them into specifications for a public tender.

In addition, we also want to gain insight into the willingness of  innovative players to set up or develop tailor-made solutions in collaboration with us within one or more of the above-mentioned areas  in order to gradually improve the efficiency, effectiveness and quality of our audit work.

Approach to market consultation and processing of information

The consultation will consist of a plenary session, in which we invite the participating market players to provide an assessment of the technological feasibility, potential complexity and possible risks for each use case.

In addition, we also organize optional 1-on-1 sessions to go deeper into the proposed solutions and any preconditions, including an estimate of the cost price and development time. Via the registration form you can indicate your interest for such a session 1-on-1 session.

We do not expect any preparation from you. Addestino Innovation Management will moderate and facilitate the discussions interactively. For a smooth discussion and good results, it is advisable to have your company or research group represented by an employee with technological expertise. They can usually estimate better and faster what the technological solutions can be for each need or desired functionality.

The market consultation will partly take place in English if there are attendees who do not speak Dutch.
      
Addestino Innovation Management will summarize the insights from the market consultation into a publicly accessible report, respecting the GDPR rules and the confidentiality of all information. The participants in the market consultation will receive this report. It will also be an appendix in the later tender.

Where and when? (customizable)

The market consultation will take place online on June 24th, 2025 from 2 pm to 5 pm (Brussels local time). You will receive the link for participation in time after registration.

The 1-on-1 sessions for those who are interested will take place on June 26th, 2025 in the afternoon or (in case of scheduling conflicts on the 26th) on June 27th, 2025.

Register

Participation is of course free, but registration is required. You can do this by filling in the registration form at the bottom of this page until June 19th, 2025, 1:30pm.