Business Valuation in Litigation: Maintaining Accuracy and Defensibility in the Age of AI

In litigation, the issue is whether the expert can explain the analysis, the data, and the judgment that produced the opinion. AI can accelerate parts of that process, but it does not remove the expert’s responsibility for the conclusion. Therefore, the question addressed in this article is where AI can assist without taking ownership away from the expert.

A valuation expert can arrive at the stand with years of experience, a detailed model, and a carefully prepared report. However, the expert may stumble when asked to explain an input that came from an AI tool and cannot explain why that input was selected. Or worse, data driven by AI use creates hallucinations that do not exist. In litigation, the issue is not whether the technology is sophisticated. The issue is whether the expert can explain the analysis, the data, and the judgment that produced the opinion.

Business valuation in litigation requires careful application of professional judgment. AI can accelerate parts of that process, but it does not remove the expert’s responsibility for the conclusion. The current question is not whether valuation professionals should use AI; it is where AI can assist without taking ownership away from the expert. That is what this article is all about.

The Courtroom Does Not Care How Sophisticated Your Model Is

Courtrooms rely on experts for independent insights into various critical yet technical matters that cannot easily be understood by laymen. For instance, if determining the quality of a car’s manufacturing is critical in a case, an automobile engineer may be called upon to provide independent insights. The expert’s experience, education, and/or training enable them to provide reliable assessments. However, if an expert relies on the judgment of other parties, their opinion will be called into question.

These requirements are formally set forth in the Federal Rule of Evidence 702, which states that an expert’s testimony must be based on sufficient facts or data and be formed using reliable principles and methodologies applied reliably. This means that a valuation analyst can use AI as a tool but not as a substitute for independent judgment.

In the Ferlito v. Harbor Freight Tools USA, Inc case, the court ruled that the expert’s testimony stands since they had transparently disclosed the use of AI and limited the use of AI to confirm conclusions they had already formed with their own judgment, without relying on AI.

The significance of Ferlito v. Harbor Freight Tools USA, Inc. is not that ChatGPT was accepted as an expert. The court accepted the expert’s testimony because their experience and independent opinion did not depend on ChatGPT. They used the tool after forming the opinion, used it to confirm conclusions, and disclosed that use. That distinction is important in litigation valuation because it separates technology that supports an opinion from technology that supplies the opinion.

The Defensibility Problem is Not AI; it is Losing Ownership of the Opinion

The previously discussed court decision to allow AI-assisted expert testimony demarcates the legal boundaries of reliance on AI for experts called on to testify. Essentially, experts are allowed to accelerate the mechanics of their analysis, but they cannot rely on AI to make judgment calls.

For instance, a valuation analyst may leverage AI to establish an initial peer company set. This involves extracting valuation-related details and screening irrelevant companies. However, screening companies require expert judgment, which should not be left completely up to AI. The problem is not using AI but relying on AI for judgment calls.

When ‘Tool-Assisted’ Becomes ‘Tool-Dependent’

Suppose you are a valuation analyst who used an AI model to extract valuation data regarding comparable companies. The AI model does not differentiate between stale data from three quarters ago and data from the last month. If you accept the numbers without checking the metadata classification, you will end up with a conclusion that you cannot defend under scrutiny, deposition, and cross-examination.

If the valuation data were segregated by date, you would see how industry trends moved over three quarters. However, if you include the data from three quarters ago in an analysis for the current period, it will lead to an inaccurate valuation because of an inaccurate valuation multiple. It will not reflect the present reality or even the market conditions from three quarters ago.

As the valuation analyst, it is your duty to filter out accurate data if you are required to testify regarding a recent valuation. Such judgment calls will always remain the responsibility of experts. So, when you use AI tools for valuations, you have to be very careful about how much cognitive offloading is happening. The research can be offloaded to AI, but any task that requires judgment must be retained by the human expert.

AI cannot make judgment calls, but let’s explore what AI can do.

Comparable Transaction Screening: Casting a Wider Net Than Any Analyst Could Manually

Comparable transaction screening is one area where AI can provide meaningful assistance. A valuation analyst may need to work through a large set of transactions and extract information that would take significant time to review manually. Suppose you want to establish the present valuation multiple for a U.S. company. Then, you will be looking at M&A data for over 5,000 transactions that happened in 2026. Relying on AI to import and organize this data will prove much faster than manually parsing through the data or relying on traditional tools.

AI can help identify potentially relevant transactions, organize valuation data, and narrow the population for further review. The final screening decision should still remain with the valuation analyst because relevance depends on the facts and purpose of the engagement.

Sensitivity Analysis and Scenario Modeling

AI can also assist with sensitivity analysis and scenario modeling. The analyst will need to assess which sets of assumptions are appropriate for the case in question. Then, the AI can run various permutations and combinations of the assumptions, and organize the resulting outputs.

This can make it easier to see how changes in assumptions affect the valuation. The value of this assistance is strongest when the analyst already knows which assumptions matter and can explain why the scenarios were selected. Technology can expand the mechanics of the analysis; it cannot decide which scenario is appropriate for the case.

The Landscape is Further Along Than You Think

The valuation profession is not approaching technology from a blank slate. The technology discussion has moved into formal valuation standards.

The IVSC has recognized advances in machine learning, deep learning, data sourcing, data processing, automated valuation models, and automated valuation reporting. The valuation standards effective January 31, 2025, address data and inputs as well as valuation models. They also recognize that specialists or service organizations, such as AI firms, may assist with parts of a valuation if the analyst lacks the necessary technical skills; if their involvement is agreed to and disclosed.

At the same time, the valuer remains responsible for selecting data, assumptions, and adjustments based on professional judgment and professional skepticism, as well as for the selection and use of valuation models. This is consistent with the broader direction of valuation practice.

Kroll’s Valuation Insights materials describe a market where valuation is increasingly a recurring process that must be consistent, timely, and defensible under scrutiny. Technology can support workflow automation, data management, model governance, permissions, audit trails, exception handling, and documentation. However, the final judgment cannot be delegated. The existence of better infrastructure does not change who owns the valuation conclusion.

Courts, standards bodies, and valuation practitioners are already addressing the role of technology. The remaining question is whether the valuation expert has a disciplined process for using AI.

Three Failure Patterns That Survive Even Rigorous QC Programs

The first failure pattern is the belief that reviewing the output is enough. If a material input, assumption, or classification was accepted because the AI output looked reasonable, the expert may still struggle to explain the basis for that judgment during deposition or testimony.

In the Kohls v. Ellison case, the judge excluded the testimony of a Stanford professor because an AI model added citations to two academic articles that do not exist. The irony is that the professor is a leading scholar on AI and misinformation; he knew the real citations, but AI mistakenly filled his placeholders with fake articles. What is more important to note is that the court did not accept an amended version of the professor’s declaration stating that the Federal Rule of Civil Procedure 11 imposes a ‘personal, nondelegable responsibility’ on the person submitting a document to court. Hence, review must involve the evidence behind the output; not only the output itself.

The second failure pattern is using AI-generated report narrative without attribution or independent verification. A narrative may read clearly and appear consistent with the valuation, but readability is not evidence of correctness. The expert remains responsible for ensuring that the explanation matches the underlying analysis and that the statements can be supported by the work performed.

The opposing counsel may ask the expert to explain which documents and what information were reviewed, what the expert’s observations were, and how they reached the conclusion presented in the report. If the expert relied too heavily on AI, they may struggle to credibly answer those questions.

This principle applies whether AI drafted one paragraph or an entire section. In fact, a recent case shows that the principle applies even when AI models a single number. In the Matter of Weber, when the accuracy of damage calculations came into question, the judge himself tested Copilot, the AI model used for these calculations. When the AI model returned three different values and admitted that expert review is required to confirm accuracy, the court held that, in New York, counsel must disclose AI assistance in preparing evidence and such evidence may require a Frye hearing before admission.

The third failure pattern is allowing AI-selected inputs to ignore the legal standard that governs the assignment. A data point can be accurate and still be inappropriate for the case. The same distinction becomes important when an AI tool selects data without understanding the specific purpose of the engagement. The relevant question is not only whether the input exists. It is whether the input belongs in this valuation.

The more a technology tool influences a material part of the valuation, the more clearly the expert should be able to show what the tool did, what the expert checked, and where the expert exercised judgment.

When to Use AI, When to Override, When to Exclude Entirely

A useful starting point is to classify AI use by the nature of the task. Tasks that are primarily mechanical and repetitive are generally the strongest candidates for AI assistance. Comparable data extraction, screening support, and scenario calculations can fit this description when the analyst retains control over the inputs and reviews the outputs. The documentation should make clear what was delegated to the tool and what was independently verified.

AI should be overridden when a task moves from processing information into selecting a material assumption or making a case-specific judgment. The analyst should take control whenever the tool cannot distinguish between relevant and irrelevant facts. AI should be excluded entirely when its use would prevent the expert from explaining a material part of the opinion or when the analyst cannot adequately validate the source and reasoning behind an output.

The same principle applies to confidentiality concerns. The ABA guidance emphasizes the importance of protecting client information when using generative AI. A useful tool is not useful enough if its use creates a separate professional problem.

The practical test is simple. At every material point in the valuation, the expert should be able to explain what information was used, why it was used, how it was checked, and why the resulting judgment is appropriate for the assignment. If the expert cannot answer those questions without referring back to the AI system, the task has moved too far toward tool dependence.

The Expert Who Can Explain Their Relationship with AI Will Win the Next Decade

The direction of valuation practice suggests that AI will become another part of the professional toolkit. The IVSC Agenda Consultation 2024 shows that valuation organizations are already responding to machine learning, data processing, automated valuation models, and automated reporting. NACVA has also established an Artificial Intelligence and Machine Learning Commission to address the implications of AI for valuation and financial forensics professionals.

The question for valuation experts is less about whether technology will be used and more about whether its use will be disciplined enough to withstand scrutiny. That discipline can become a professional advantage. An expert who refuses every technology tool may spend more time on mechanical work than necessary. An expert who accepts every output may create weaknesses that are difficult to defend. The stronger position is to use technology where it improves the process while keeping professional judgment at the center of the engagement.

An expert should be able to say: “I used these AI tools for these specific tasks, validated the inputs and outputs through independent steps, and my opinion rests on my professional judgment informed by technology.” That statement does more than explain how the work was completed; it identifies who owns the conclusion.

The real competitive risk for valuation experts is not replacement by AI; it is replacement by an expert who can articulate and defend their AI protocol better than you can.

Colin McCrea, CVA, EA, is the Founding Partner at Eqvista Inc., where he leads the valuation team and plays a pivotal role in guiding the firm’s strategic direction. With over seven years of experience in valuation, taxation, and audit for SMEs across the U.S. and Asia, he has developed a strong reputation for precision, compliance, and client-focused insights. He is currently managing company valuations and corporate taxation at Eqvista, overseeing complex engagements that ensure financial accuracy and strategic value for businesses of all sizes.

Over his career, Mr. McCrea has completed 1,000+ valuation projects, covering general business valuations, ESOP (409A) valuations, shareholder buy-outs, litigation consultation and expert testimony, and estate and gift reporting. His  professional approach blends technical rigor with strategic insight, ensuring that clients receive accurate, compliant, and value-driven financial analyses. His leadership at Eqvista continues to shape the firm’s commitment to quality and innovation in valuation and taxation services.

Mr. McCrea can be contacted at (702) 781-3710 or by e-mail to Colin@eqvista.com.