Partnerblog
AI assistants can save in-house lawyers a lot of time on questions about their companies, directors and shareholders. But what if the data on your own group is scattered, incomplete or contradictory? What should you know as an in-house lawyer before letting AI loose on your entity data?
Who sits on the board of which subsidiary? When do the mandates expire? Who holds which shares, directly or indirectly? For in-house lawyers, these are everyday questions. Today, answering them often means searching through spreadsheets, registers, minutes and inboxes. AI assistants promise to change that. They can, but not without conditions.
Much has already been written about the reliability of AI in legal research, in legislation, case law and legal doctrine. But there is a second question that gets less attention: how reliable is the data on your own companies?
The good: AI that works directly with your data
Until recently, companies connected their systems through APIs: fixed links between, for example, an entity management tool, accounting, a tax tool and a contract management system. Such a link does exactly what it was built for, but every new question requires new development.
The Model Context Protocol (MCP) changes that. MCP is an open standard that lets AI assistants such as ChatGPT, Claude or Copilot connect securely to business software. Think of it as a universal plug: the AI assistant you already use can retrieve data from different systems and combine it to answer your question. So there is no need to switch to a separate AI tool.
For in-house lawyers, that opens up concrete possibilities:
- Ad-hoc questions without technical knowledge: “Which director mandates expire this year in our Belgian subsidiaries?”
- Custom reporting, for example an overview of the group structure for an audit or due diligence, without first having to rework an export.
- Combining data from different sources, such as entity data, financial data and contracts.
The result: less time spent searching, more time for legal analysis.
The bad: contradictory sources and missing history
The biggest risk is not in the AI itself, but in the data. Entity data is often scattered:
- in public sources such as the Crossroads Bank for Enterprises (KBO/BCE) and the Belgian Official Gazette, and in internal sources such as the share register and minutes;
- split between legal, finance and tax, each keeping their own version;
- in spreadsheets and documents, which are not structured data;
- partly with external advisors.
What happens when an AI agent pulls the same data point from two sources, and they contradict each other? In legal work, 99% correct is not good enough. A wrong director or expiry date can have real consequences.
So the question is: where is your master data, the one place that counts as the truth?
Anyone who wants to centralise data usually runs into the same problems: unstructured data, missing details, incomplete history and inconsistencies between and within sources. AI can help, but it does not solve this on its own. Two examples from practice:
- During a migration of entity data, an LLM was used to automatically recognise the components of 30,000 addresses (street, number, postal code, city). It succeeded in 80% of cases. The remaining 6,000 addresses needed manual review and correction.
- For a group with 1,800 shareholders and numbered shares, twenty years of share history had to be reconstructed. Here too, human work remained essential.
The lesson: AI speeds up the work, but responsibility for correct data stays with people.
The ugly: governance and control
As soon as entity data becomes accessible to a wider group of users through AI, governance questions become urgent:
- Who owns the entity data? Legal, finance or accounting? Without a clear owner, there is no clear responsibility for its quality.
- Are there rules on the confidentiality and use of that data? And do employees know them?
- Who can see what? The more people gain access through AI, the more important a clear overview of user rights becomes. An AI assistant should never show more than the user is allowed to see.
- Where does the data go? Is it used to train AI models? Where is it processed? And how much risk is your organisation willing to take?
These are not technical questions, but legal and organisational choices. In-house lawyers therefore play a key role.
Five questions to ask today
Anyone who wants to use AI safely for entity data can start with these questions:
- Do we have one central, up-to-date source for our entity data?
- How complete and correct is that data, including its history?
- Who owns that data, and who maintains it?
- What rules apply to access, confidentiality and use?
- Which AI tools may employees use, and under what conditions?
AI will profoundly change the way in-house lawyers work with entity data. Those who get their data in order today will reap the benefits tomorrow.
Want to know more? On 14 December, Olivier Van Borsel and Andy Kellens of Corporify will discuss this topic in an IBJ-IJE partner webinar, with concrete examples from practice and time for your questions.
