Artificial intelligence (AI) is a transformative technology, and the mergers and acquisitions (M&A) space is no exception. With recent focus on the advancements in generative AI, it is worthwhile to differentiate generative AI from the AI used in predictive analytics and other current generation AI tools.

Predictive AI broadly speaking makes predictions or forecasts based on historical data patterns and existing data. Examples of predictive AI applications include weather forecasting, stock market prediction, and customer behavior analysis. AI technologies based on predictive AI have been used in legal tools for years. Examples include legal document classification in knowledge management systems, clause extraction in due diligence tools and relevance prediction (i.e., technology assisted review) in document discovery platforms.

Generative AI, on the other hand, is focused on generating new content or information rather than making predictions. Examples of generative AI applications include image synthesis, text generation, software code and music composition.

In the legal industry, law firms already use AI in M&A transactions. Current applications of AI in the M&A context include the following.

• Due diligence: The widespread adoption of cloud-based virtual data rooms (VDRs) has created an accessible collection of unstructured data. In the past, these documents were reviewed manually by large teams of lawyers. In recent years, AI-enabled tools using due diligence-specific models have been deployed to more efficiently review and analyze VDR documents, identifying, categorizing and extracting information relevant to the transaction. These tools are used to extract and organize key provisions such as “change of control,” “most favored nation,” and “termination,” making it easier to track third-party consents, understand the jurisdictional scope of the target’s contracts and spot problematic contract clauses. They can also greatly speed up more mundane tasks like detecting missing signatures and undated agreements. Results can be presented in user-friendly dashboards to facilitate fast red flag reviews. Whether generative AI’s due diligence-related capabilities outperform the “finding” function of existing due diligence-targeted AI models remains to be seen. Where generative AI is potentially additive is in helping to quickly produce polished text-heavy reports and presentations of results for different audiences.

• Contract drafting: Transactional M&A practice is heavily reliant on forms, precedents and templates. Existing tools, including those that are not AI-based, allow for automated production of both simple and complex contracts. We expect to see increasing integration of generative AI into the drafting process. A simple example would be to use generative AI to offer suggestions on modifying specific contract terms to make them more or less “friendly” or protective of a specific party in a deal. As generative AI tools mature, we expect to see increasing usage of the technology to used to alert users where a contract departs from specific agreed or expected terms, flag provisions that do not comply with a set of policies, report the risks of using non-compliant language and provide drafting suggestions from internal template banks and industry standards.

Some commentators have suggested that ChatGPT can be used to produce contracts such as share and asset purchase agreements. In our testing, the current version of ChatGPT could produce simple short agreements but could not produce longer more complex agreements. While the underlying GPT models are being used in legal technologies today, due to its limitations and data protection concerns, we do not expect ChatGPT to be used in the near-term by professionals to draft complex or context-specific legal documents.

• Legal research: M&A is a process-driven area that does not necessarily involve regular legal research. Where research is required, generative AI can be used as a supplement to traditional methods. Generative AI will quickly produce answers accompanied by explanations and supporting sources. Since answers may not be correct, and confidential and privileged information should not be entered into a publicly available AI platform, this new tool must be used with caution.

Emerging use cases in M&A practice include:

• Quantitative analysis of deal terms: The speed of AI can potentially be deployed during real time negotiations or compressed auction processes. Predictive AI can perform rapid analysis of positions in order to see what the overall effect proposed terms might have on the deal.¹ For example, in a hypothetical test case by our team, ChatGPT could model the impact of different earn-out scenarios on shareholders and employees of a target business during a real-time mock negotiation.²  ChatGPT was able to (i) identify patterns, correlations, and trends that could inform the evaluation of earn-out scenarios, (ii) forecast future financial performance, market conditions or other relevant variables that impact earn-out scenarios, (iii) analyze various risk factors associated with earn-out scenarios, such as market volatility, customer retention, competitive landscape and regulatory change, and (iv) perform sensitivity analysis to assess the impact of different variables on earn-out scenarios by adjusting parameters such as revenue growth rates, cost structures or market share, to simulate how these changes affect the earn-out calculations and help evaluate the risks and opportunities associated with each scenario.

• Enhanced risk analysis: Analysis of risk related to outstanding litigation can often slow the cadence of a M&A transaction. Predictive AI can be harnessed to analyze the prospects of successful litigation by identifying decisions based on similar factors and outcomes, identifying leading authorities, and detecting how key factors can impact a client’s position. While these tools are to our knowledge only applicable today in niche areas, given that the large published volume of judicial decisions it seems like a natural area for future development. There are also AI-driven software tools that can perform audit-like functions to review a target’s financial reporting. It is possible to imagine a day where these AI-assisted risk analyses are performed on a routine basis. A precedent for this exists in the form of third-party source code scans that are now routinely deployed in acquisitions of software companies.

• Success predictions: In theory, by analyzing publicly available historical business and financial data, predictive AI can provide analytics to assist parties in assessing the potential success or failure of certain strategies including acquisitions. In the M&A context, AI could analyze information such as a company’s brand, management team, trajectory, resources such as product diversity, mineral rights and productive capacity, and financial information, and determine which entities, when combined, would form a more profitable entity.³ In an asset purchase transaction, AI could be utilized to determine which assets or elements of a business would be worth more if sold individually.⁴ All of these tools, if harnessed correctly, can help purchasers and lawyers make more informed decisions, evaluate risks and develop effective business or legal strategies.

• Post-acquisition integration: AI is also potentially useful to analyze data from the post-acquisition integration process to identify areas for improvement and optimize operations. For example, based a precedent bank of past transactions can be analyzed and used to forecast which actions would result in more desirable outcomes.⁵ For example, AI-powered software can analyze data and identify cross-selling opportunities or flag potential areas of overlap in the merged companies’ operations.⁶

AI has the ability to transform the way M&A is conducted. While AI is on the cusp of disrupting the tried-and-true M&A playbook, users of AI should also be cognizant of potential risks.

Cautions

• Reliability: Many AI systems have limitations and risks that need to be resolved before they can be relied upon to make ethical decisions, accurately interpret documents in specific contexts and provide valuable negotiation tools.

Predictive AI output may be biased or flawed if the data provided to train the AI model is biased or if the system’s method of analyzing data is flawed.⁷ Human judgment must always be applied when using AI as relying on the results of AI without additional research may lead to incorrect or irrelevant advice.

A specific issue that often comes up when using generative AI is “hallucinations.”⁸ Hallucinations occur when generative AI produces a confident response that is not true. These hallucinations are common in many current generative AI tools, particularly where they are not paired with information from reliable sources. This reduces their reliability and means work done by generative AI still has to be reviewed by a human in order to meet professional standards. Take the recent, real-life example where a licensed lawyer in New York used ChatGPT to conduct research and did not confirm the validity of the cases ChatGPT quoted.⁹ The citing at least six fictitious cases in a filed brief, which were hallucinated by generative AI, led to a sanction hearing for counsel.¹⁰ We expect hallucinations in legal tools to be mitigated by supplementing models with reliable data, such as precedent language or a trusted electronic research database.

• Lack of contextual understanding: The difficulty of providing AI with adequate context may limit its utility. For example, the industries of the parties involved may greatly impact the material issues in the transaction. In a M&A transaction involving a target company in the business of providing virtual online courses, the licensing arrangements would differ vastly between a provider who licenses the virtual courses for use on an institution’s learning management system, as opposed to a provider who allows users to access the virtual courses directly through a webpage hosted and owned by the provider; a generally trained generative AI model may not be aware of the differences between these licensing arrangements or capable of assessing the unique “red flags” under each arrangement. Jurisdictional differences in legislation also pose challenges for generative AI. We expect to see significant work to attempt to overcome these challenges.
• Data security / confidentiality: Leaked information can be extremely damaging to a proposed deal and to the interests of clients as it undermines their bargaining position. One of the key principles of the legal profession is the duty of confidentiality a lawyer owes to their client.

Users need to be aware that data submitted to some AI tools may be stored indefinitely and may be shared with third parties or used for AI development and improvement. In law firms, no confidential information, privileged, or personal information should be submitted to AI tools unless the user’s organization has approved the specific tool for use.

Conclusion

It will be vital for lawyers working on M&A transactions to be able to apply AI when working on deals. The efficiencies offered are just too great to ignore. However, practitioners will have to make sure to use AI ethically, responsibly and with proper due diligence.

If you have any questions about the AI in the M&A context or corporate law related issues in general, please contact co-authors Danika Kotylak, David Mikhaiel, editors Andrea Johnson, Eric Lung, and Kimberly Burns or another member of Dentons’ Corporate group.

1. Crisstene Gonzalez-Wertz, Paul Price, Chrisophe Begue, and Bruce Anderson, “Stronger M&A strategies through AI-driven processes” (December 2019), online: <https://www.ibm.com/downloads/cas/KYBQMBBW>. ↩
2. No client information was submitted to ChatGPT in the described test case. ↩
3. Adam Boostrom et al, “How AI will guide future M&A deals” (January 4, 2023), online: Tata Consulting Services <www.tcs.com/what-we-do/services/consulting/white-paper/ai-algorithms-mergers-and-acquisition-deals>.↩
4. Ibid.↩
5. Ibid.↩
6. Allen Clark, “How AI and data analysis can help your business provide the personalized service customers demand” (March 8, 2023), online: Zendesk Blog <www.zendesk.com/blog/ai-data-analysis/>. ↩
7. Jake Silberg and James Manyika, “Tackling bias in artificial intelligence (and in humans)” (June 6, 2019), online: McKinsey & Company < https://www.mckinsey.com/featured-insights/artificial-intelligence/tackling-bias-in-artificial-intelligence-and-in-humans> ↩
8. Craig S. Smith, “Hallucinations Could Blunt ChatGPT’s Success” (March 13, 2023), online: Institute of Electrical and Electronics Engineers https://spectrum.ieee.org/ai-hallucination↩
9. Ramishah Maruf, “Lawyer apologizes for fake court citations from ChatGPT” (May 28, 2023), online: CNN https://www.cnn.com/2023/05/27/business/chat-gpt-avianca-mata-lawyers/index.html↩
10. Ibid. ↩