chevron-down Created with Sketch Beta.

ARTICLE

Artificial Intelligence: Risks and Opportunities in Mergers & Acquisitions

William Marquardt, Ken Mathieu, and Frank Dery

Summary

  • This article focuses on leveraging AI in the M&A due diligence process.
  • We delve into the associated risks and advantages of employing AI to heighten efficiency and capitalize on data located within virtual data rooms, as well as the risks of relying on AI as an integral part of assessing a target entity’s operations.
  • By addressing these elements, organizations can better manage AI and achieve their strategic M&A goals effectively.
Artificial Intelligence: Risks and Opportunities in Mergers & Acquisitions
gorodenkoff via Getty Images

By leveraging artificial intelligence (AI) technologies, merger and acquisition (M&A) professionals can enhance operational efficiency, mitigate risks, gain clearer insights for decision-making, and ultimately increase the likelihood of a successful M&A transaction. AI holds promise across the M&A landscape, including with target identification, enhanced due diligence processes, valuation modeling, and integration or synergy opportunities. This article focuses on leveraging AI in the M&A due diligence process, delving into the associated risks and advantages of employing AI to heighten efficiency and capitalize on data located within virtual data rooms, as well as the risks of relying on AI as an integral part of assessing a target entity’s operations.

Structured and Unstructured Data Analysis

Due diligence, which typically encompasses financial, legal, operational, and regulatory components, is a cornerstone of an effective M&A process, providing critical insights that can lead to a successful M&A transaction. (The basic components of an M&A process consist of strategic planning, due diligence, valuation/pricing, integration planning, regulatory compliance, communications, post-merger integration, and change management.) When considering how to best leverage AI tools, assessing the type of information within a virtual data room is crucial. Broadly speaking, there are two types of data: structured and unstructured.

Structured data are generally organized in a fixed format consisting of rows and columns of information and are typically stored in relational databases or exported to MS Excel spreadsheets. This information is easily searchable, and the use of structured data AI applications, including data mining, predictive analytics, and various machine-learning models focusing on pattern recognition, can be helpful. These robotic process automation models can be used to detect anomalies within a data set; they are regularly used by fraud examiners but can be readily adapted for M&A transactions. AI provides enhanced efficiency and accuracy in data analysis and, when properly implemented, can be tailored to address specific industry, regulatory, or company risks.

Structured Data

Aside from speeding up data analytics, the ability to review an entire data set and conduct detailed trending and ratio analysis provides a significant advantage in producing more accurate and reliable results by eliminating sampling errors and selection bias while identifying outliers or anomalies within the dataset. The following are examples of the types of anomalies that can be identified using robotic process automation AI:

  • Revenue recognition: Detect inconsistencies, irregularities, or outliers in how revenue is recognized compared with historical and expected patterns. For example, an acceleration or delay in recognizing revenue could be identified by evaluating the frequency and pattern of changes in a project’s estimate to complete or by detecting revenue potentially being recorded before goods are shipped or services are rendered.
  • Expense recognition: Identify unusual increases or decreases in expenses such as costs being accelerated or deferred to meet financial targets. AI can also identify inconsistencies in cost categorizations, including misclassifications between operating expenses; selling, general, and administrative expenses; and capital expenditures. In addition, AI can assess vendor spend activity and duplicate or recurrent payment activity, which could indicate fraud or accounting errors.
  • Unusual or one-time transactions: Identify transactions that deviate from the historical baseline in terms of volume, velocity, and counterparties.
  • Cash-flow anomalies: Detect irregularities in cash-flow patterns such as large, unexplained transfers between accounts or abnormal cash inflows and outflows that do not align with operational activities or historical trends.
  • Accounts receivable and payable anomalies: Identify patterns in accounts receivable and accounts payable that suggest issues like re-aging of receivables, uncollectable receivables, or delayed payments that could indicate cash-flow issues.
  • Inventory management: Identify anomalies in inventory levels, turnover rates, or valuation methods. AI can also identify anomalies with respect to trends in inventory obsolescence and slow-moving inventory.
  • Payroll and compensation: Identify potential payments to ghost employees or irregularities in bonus or commission payments that could be used to manipulate employee compensation reporting.
  • Related-party transactions: Identify transactions between related parties that may require additional scrutiny due to potential conflicts of interest or dealings that are not at arm’s length.

Unstructured Data

The term “unstructured data” refers to information that does not have a predefined format or organizational structure, making it more difficult to collect, process, and analyze. Within a virtual data room, these data can include customer and vendor contracts, lease agreements, industry and competitor data, internal policy and process documentation, management communications (e.g., monthly financial and operational reviews), text and email communications, etc. Outside of the virtual data room, additional unstructured data—such as social media postings that contain customer and company reviews (e.g., Indeed, Glassdoor, LinkedIn, Vault, and industry-specific websites)—can be collected and reviewed.

These types of AI-enhanced reviews use natural language processing, a subfield of AI that involves the development of algorithms and models that focus on evaluating language. In practice, natural language processing can identify key contract terms and conditions (regulatory contract clauses), highlight inconsistencies (payment terms), identify service-level agreement metrics (performance requirements), and more. Natural language processing can generate red flags such as the following:

  • Inconsistent terms and clauses: Detect inconsistencies or discrepancies in terms and clauses across different sections of a contract or among multiple contracts, including in the use of defined terms.
  • Missing contract clauses: Identify the absence of essential clauses such as confidentiality, indemnity, adherence to data protection laws, industry-specific regulations, regulatory compliance, or internal company policies.
  • Ambiguous language and outdated or risky terms: Recognize terms and conditions that are no longer valid, have been updated, or may pose a high risk to one party, such as unfavorable payment terms, excessive penalties, or one-sided obligations.
  • Detection of fraud indicators: Identify common fraud indicators in communications, such as references to off-the-books transactions, pressure to meet targets, or unusual financial arrangements.
  • Conflicts in internal communications: Review internal emails and communications for signs of conflict, pressure to manipulate figures, disagreements about financial practices, or application of generally accepted accounting principles.
  • Media/news information: Outside data
    • Negative sentiment in news and reports: Analyze the sentiment of news articles, press releases, blogs, and social media mentions, which can help identify trends or concerns about a company’s financial stability, its market reputation, or quality issues with a particular product or service.
    • Adverse mentions in regulatory filings: Analyze regulatory filings for mentions of investigations, penalties, or compliance issues that could affect a company’s financial status.
    • Increased litigation mentions: Detect increased mentions of litigation, lawsuits, or legal disputes in various documents, which could indicate financial or reputational risks.
    • Sentiment analysis of stakeholder feedback: Analyze feedback from customers, employees, and investors for sentiment or concerns about financial performance and stability.

Using AI to examine large volumes of structured and unstructured data allows M&A professionals to enhance data-review processes, leading to a faster synthesis of potentially critical issues earlier in the M&A process.

Caution Ahead: Areas of Concern

The current enthusiasm for quickly adopting and deploying the newest tools in M&A transactions to shorten the diligence process and improve access to actionable intelligence needs to be tempered by a realistic assessment of what AI tools can deliver in individual transactions.

Incorporating AI in M&A transactions presents technical, regulatory, ethical, and operational challenges. A major technical issue is model bias, as AI models rely heavily on the quality of their training data. Incomplete, biased, or inaccurate training data can result in flawed predictions and analyses. This is particularly relevant for M&A transactions, where the data can be complex, opaque, and often incomplete, making it difficult to interpret the model’s output.

Moreover, transaction models are trained on historical data, which may not accurately predict future outcomes, especially in dynamic or rapidly changing environments where most M&A activity occurs. The adage “garbage in, garbage out” applies, highlighting that the quality of output is determined by the quality of the input. If flawed, inaccurate, or poor-quality data are input, the resulting output will also be flawed, inaccurate, or of poor quality.

Regulatory challenges involve data privacy concerns and issues related to copyright and intellectual property infringement when modifying third-party models (e.g., adjusting model fit) for a specific industry or jurisdiction. In addition, data security and privacy are critical when processing large volumes of sensitive data, such as financial records, customer information, and intellectual property. Ensuring the security and privacy of such data is essential to prevent breaches, leaks, or misuse.

Operational considerations involve the level of reliance on AI for M&A transactions. There is a risk of overreliance as AI models become more sophisticated, neglecting human expertise and judgment, which can be dangerous. While AI can automate certain tasks and provide valuable insights, it should complement, not replace, human decision-making.

AI Washing in an M&A Transaction: Buyer Beware

Given the popularity of the term “AI” and potential future value associated with leveraging this technology (or loss of value from not leveraging it), buyers should heavily scrutinize a subject company’s representations regarding the current and future use of AI and the impact on financial projections and valuation. The Securities and Exchange Commission has drawn parallels between AI and ESG (environmental, social, and governance) representations by likening “greenwashing” to “AI washing.” See Press Release, U.S. Sec. & Exch. Comm’n, SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence (Mar. 18, 2024); Chair Gary Gensler on AI Washing (Mar. 18, 2024). Just as “greenwashing” refers to unfounded representations regarding environmental sustainability, “AI washing” refers to unfounded representations regarding, for example, cost cutting to improve efficiencies or revenue enhancing, such as selling products embedded with AI.

AI is a powerful tool that can enhance a business’s value in various ways, but it can also be a buzzword used to attract potential buyers by exploiting their fear of missing out, potentially leading to irrational decisions. For instance, when assessing the value of an entity’s proprietary code, it is crucial to examine the code to ensure a proper balance between AI- and developer-generated components. The ease with which AI can replicate the code could significantly influence the valuation of the target entity.

While there’s always a risk that the company may not achieve the expected financial performance post-acquisition, understanding this risk and pricing it appropriately are crucial for the acquisition’s overall success. Below are examples of potential AI-related representations:

  • Manufacturing: Creating a more efficient supply chain through AI-driven inventory and production planning leads to reductions in the investment in inventory and in labor costs.
  • Reciprocal synergies: Creating synergies among divisions increases the utilization of shared resources and increases sales through cross-selling products and services.
  • Technology: More efficient software development results in a need for fewer programmers, decreased time to market, and higher-quality results.
  • Healthcare: More efficient billing and collection processes lead to a decrease in labor and an improvement in the timing of the cash cycle (i.e., the time between the provision of the service and the collection of cash).
  • Transportation: Enhanced logistics planning and execution could result from AI.
  • Insurance: More precise underwriting models lead to improved product pricing, fewer claims, and greater market share.

In addition to seller representations and the risk of AI washing, there are risks related to the target company’s operations. Actual use of AI could present the following risks:

  • Intellectual property: ownership and validity (potential infringements) and the patent landscape (if applicable, the issue of rapid changes and copying of ideas or applications that are open source)
  • Regulatory/compliance: data privacy and AI-specific regulations (e.g., material issues that may affect the development, deployment, and use of the company’s technology)
  • Technology: algorithm bias, security, etc.
  • Operational: scalability, talent/expertise departure post-acquisition (nondisclosure and noncompete agreements, etc.), performance and accuracy of AI model, testing, validation, updates, etc.
  • Legal/contractual: licensing agreements, third-party dependencies, etc.
  • Market/competitive: customer contracts (change of control, exclusivity, etc.)
  • Ethical/reputational: data usage, transparency, etc.

Recommendations and Conclusion

To successfully leverage AI-driven activities in an M&A transaction, organizations should start by defining clear objectives and desired outcomes, beginning with small, manageable goals and gradually expanding. Selecting appropriate partners and technology is crucial, as the right expertise and tools can have significant impacts on the project’s success. Equally important is establishing robust data governance and ensuring data quality to maintain the integrity and reliability of AI outputs. Continuous monitoring and evaluation of outcomes help in assessing progress and making necessary adjustments. In addition, it is vital to understand how the seller defines AI within the context of the seller’s representations and how AI is employed to identify potential misrepresentation and risks associated with its application. By addressing these elements, organizations can better manage AI and achieve their strategic M&A goals effectively.

The views and opinions expressed in this article are those of the authors and do not necessarily reflect the opinions, position, or policy of Berkeley Research Group, LLC, or its other employees and affiliates.

    Authors