AI for Merger Filing Analysis
We built a tool that automates merger-filing research for M&A transactions
Cross-border merger control is high-stakes business. Gun-jumping can lead to significant penalties, “stop-the-clock” mechanisms can disrupt dealflow. Even transactions that do not meet threshold requirements can catch the attention of regulators.
While there are unforeseen risks to every transaction, getting the fundamentals right goes a long way. These include jurisdiction-specific thresholds, evolving notification regimes, and enforcement priorities — shaped by sectoral focus and geopolitical dynamics.
We spoke with Janet Hui, Antitrust/M&A Partner at JunHe LLP, and the former Co-Chair of IBA’s Antitrust Committee. She explained that the “fundamentals” are far more complex than they seem. In recent weeks, the FTC released new filing thresholds and filing fees for 2026 — How would this be dealt with if the transaction takes two years to complete? Would the parties need to apply again if they’ve already secured a green light? When a conglomerate merger spans the globe, how do the parties manage complex competition rules across more than 100 jurisdictions?
Inspired by Janet, we attempted to build an Agent Skill to help lawyers or in-house counsel with their initial desktop research.
Agent Skills are folders of instructions, scripts, and resources that (AI) agents can discover and use to do things more accurately and efficiently.
In the assets/ folder, we packaged primary sources include legislation and competition authority guidelines from various jurisdictions, as well as secondary sources such as Merger Control 2025 by Chambers and Partners, World Competition Database by George Washington University, and Mergerfilers Global Legal Guides. This helps guide the AI on initial research, much like how a lawyer might open up a textbook when faced with a new area of law.
The SKILL.md file provides a detailed framework to the AI to instruct it on how to perform its research across different aspects, including but not limited to:
whether notification is mandatory or voluntary;
notification timelines;
sanctions for failure to file;
potential director liabilities;
recent enforcement trends, etc
The final ingredient involves the company-specific information, which includes publicly available data (e.g. annual reports) and non-public due-diligence information. Note that this information is independent of the Agent Skill to keep things modular — hence this Agent Skill can be reused for any M&A transactions. With all the elements in place, the AI is able to produce a tailored piece of desktop research in minutes.
Let’s take a case of Rio Tinto’s interest in a joint venture with Codelco to develop a new lithium project in Chile as an example. We provided the AI access to Rio Tinto’s 2024 Annual Report, as well as Codelco’s 2024 Annual Operational and Financial Report. Using the Agent Skill, it was able to generate a detailed desktop analysis for 40+ jurisdictions in 45 minutes.
The following is an excerpt from the Chilean analysis:
We were encouraged by how the AI performed, having:
Staying grounded in primary legislation and competition authority guidance we provided
Supplementing its research by performing web searches for recent enforcement trends across the different jurisdictions
Capturing the nuances such as how “change of control” is defined across jurisdictions (which are sometimes overlooked by antitrust lawyers!)
Lastly, it does have some “common sense” of a competent lawyer, with attempts to estimate regional revenue from annual reports and flagging uncertainties for further review
We also prompted the AI to generate a client questionnaire asking for deal-specific information, while highlighting any assumptions it made.
As a side experiment, we built a browser based tool to visualise merger control requirements across jurisdictions by way of decision trees.
Here, we turned the guidance for European Commission under Council Regulation (EC) No 139/2004 into an interactive experience, making it particularly helpful for quick overviews for unfamiliar jurisdictions, especially those where primary sources are unavailable in English.
Try it out here: http://merger-filing.terracotta.dev
Despite all its bells and whistles, this is no substitute for experienced antitrust counsel. Merger control analysis requires judgment calls that go beyond threshold calculations. What it does do well is compress the initial desktop research from days to minutes — giving lawyers a good starting point and reducing the risk of missed rules as a result of information overload.
We believe that AI enables practitioners to spend more time on the work that truly requires their strategic thinking and expertise.