.png)
Can AI Review a Valuation Report? Yes, but not the way a CBV does
- Vince Osbourne
- 1 day ago
- 8 min read
Insights from Vince Osbourne, CMA (US), CBV, ABV
Article Written July 13, 2026 | Posted 2026-07-13| Business Valuation
Artificial intelligence (“AI”) tools are increasingly becoming part of valuation practice. Practitioners are using large language models to research comparable transactions, extract data from inconsistently formatted financial statements, summarize due diligence materials and increasingly perform a first-pass quality review of valuation reports against professional standards.
The technology is moving faster than many professional workflows built around it. That creates both opportunity and risk. Used carefully, AI can improve consistency and free practitioners to focus on issues requiring judgment rather than repetitive and simple tasks. Used carelessly, it can create false confidence, introduce unsupported content, or obscure where professional responsibility lies.
The question this article addresses is specific: Whether an AI model can meaningfully review a valuation report for compliance with the CBV Institute Practice Standards. In brief, the answer to that question depends on what the reviewer is actually asking and how the question is asked. AI handles "is this addressed?" reasonably well. It is much weaker on "is this supportable?" In valuation work, the latter is usually the more challenging question. CBVs who use these tools without appreciating that distinction are taking on professional risk.
What AI Does Well in a Valuation Context
Large language models (“LLMs”) are, at their core, systems trained to predict patterns in text. As a result, they have real and useful capabilities for valuation work.
AI can be effective at checking whether a report contains required elements pursuant to the Practice Standards such as the scope of work, intended purpose and users, valuation date, standard of value, valuation approaches, key assumptions, restrictions, and required disclosures. This is mechanical checking, and AI does it quickly and consistently. It may catch omissions that a human reviewer reading the same long document might miss.
AI is well-suited to flagging contradictions within a document. Some examples include a discount rate stated as one figure in the narrative and another in the schedule; a maintainable earnings figure that does not reconcile across sections; a defined term used inconsistently; or an adjustment described in the report but not reflected in the calculation. In the author’s experience, these are the kinds of errors that can easily creep into long reports assembled under time pressure.
AI can also compare report content against the text of professional standards. This requires proper configuration: grounding the model in the current CBV Institute Practice Standards through retrieval-augmented generation¹ (“RAG”), or supplying that text directly in a large context window². It can then determine whether the report addresses the required matters and flag areas where the language may be incomplete or inconsistent.
Pulling financial data from inconsistently formatted statements or documents such as general ledgers, summarizing lengthy materials such as agreements and organizing comparable transaction data are tasks where AI can deliver genuine efficiency. These are time-consuming, error-prone manual tasks, and automating the first pass can improve both speed and consistency.
These are first-pass functions. The judgment calls remain with the reviewer.
Where AI Fails
The most important questions in a valuation review are not structural. They are judgmental. Is the selected discount rate reasonable given this company’s risk profile? Does the underlying evidence adequately support the maintainable earnings figure? Are the redundant asset adjustments defensible? Are the selected valuation approaches appropriate for the business, the purpose, and the available evidence? Is the selected industry code the correct one?
An AI model can tell you whether a report discusses the discount rate, but it cannot tell you whether the rate is professionally supportable in the circumstances. It can flag that a maintainable EBITDA figure changed between sections, but it cannot tell you whether the final number is supportable.
The updated Practice Standards place significant emphasis on credible, properly supported valuation conclusions³. The level of work and corroboration expected varies depending on the nature of the valuation conclusion. Still, the central principle is clear: important inputs and assumptions must be supported by appropriate evidence.
LLMs generate plausible-sounding text, and that fluency can mask fabrication⁴. In valuation work, this risk is not theoretical⁵. A model may invent comparable transactions, misstate a standard, cite a source that does not exist, or produce confident commentary on company-specific risk with nothing behind it. The text sounds right. That is the problem.
In a quality control context, both types of errors matter. Missing a real issue is a problem. Flagging compliant work as deficient creates unnecessary rework. Either way, the professional consequences fall on the valuator.
AI models may not reliably know the most current standards, guidance, market data, or professional requirements unless those materials are specifically provided and the model is instructed to use them. This is especially important where standards have recently changed. A model asked to review a valuation report against current CBV Institute Practice Standards may confidently apply outdated requirements if it is relying on older training data or incomplete source material.
Practice Standards 100 and 120
The revised Practice Standards effective for engagements beginning on or after January 1, 2026, provide an important framework for thinking about AI-assisted valuation work.
Practice Standard No. 100 establishes a principles-based foundation for valuation conclusions. It emphasizes independence, objectivity, professional skepticism, competence, and informed professional judgment. Each of these concepts has direct implications for AI-assisted work.
Professional skepticism is an attitude, not a checklist. It involves questioning, probing, and declining to accept assumptions at face value. Just as it cannot be delegated to a junior with no valuations experience or training, it cannot be delegated to a model that lacks independent awareness, professional accountability, and a stake in the conclusion. An output is a probabilistic response. A judgment is a reasoned professional conclusion.
Practice Standard No. 120 is particularly relevant because it addresses the scope of work, quality review, and the reliability of external sources and tools⁷. A valuation practice using AI in report preparation or review should treat the tool as part of the engagement process that requires professional oversight. The valuator must consider whether the tool is appropriate for the purpose, whether its outputs are reliable, and whether its limitations are understood.
Practice Standards Nos. 110 and 130 are also relevant. AI may help identify whether required report disclosures appear to be present and whether file documentation appears organized, but the presence of disclosure language or supporting documents does not, by itself, establish that the valuation conclusion is credible or properly supported.
Competence includes understanding the limitations of the tools you use. A practitioner who treats a clean AI review as assurance of compliance is not meeting that standard. The bar does not go down because a model ran first. It goes up. You now need to understand the valuation conclusion, the supporting documents and reasoning behind the conclusion as well as the technology.
Confidentiality and Client Data
Valuation reports contain confidential financial information, tax-planning details, litigation strategy, management representations, payroll information, customer data, and other sensitive information. Before uploading client information to any AI tool, a firm should understand whether the tool stores prompts, uses inputs for model training, permits administrative review, retains files, or transfers data outside approved environments.
For professional valuation practices, this is a governance issue. Firms should distinguish between public AI tools, enterprise-approved AI tools, and internally controlled systems. They should also identify which types of client information should never be entered into an AI system unless specific safeguards and approvals are in place. Firms must establish policies and procedures in this regard, communicate them to staff, and provide relevant training thereon.
The question is not simply: “Can the AI model review the report?” The question is also: “Can this particular tool be used on this particular client's information in this particular context?”
Kalex Partners Inc. welcomes the opportunity to answer your questions and provide support on business valuation matters, including those related to topics covered in this article.
What Responsible AI-Assisted Review Looks Like
AI is a triage tool. It surfaces gaps, inconsistencies, and missing disclosures. It is not a sign-off/approval tool. Assessing reasonableness, weighing evidence, and standing behind the conclusion are still the reviewer's job.
Running a model first does not replace human review. It just changes where the reviewer should focus his or her efforts. Is the conclusion supported? Are the assumptions reasonable? Has contradictory evidence or possible positions been addressed? Those are still the reviewer's questions to answer.
In practical terms, a responsible workflow might involve running an AI-assisted review against a controlled copy of the report and source materials, requiring the model to identify the specific report section or source document supporting each issue it raises. The reviewer would then assess each item and classify it as accepted, rejected, or requiring follow-up. The AI output should become part of the review process, not the conclusion itself.
At the firm level, that type of workflow should be supported by clear policy and documentation. A valuation practice using AI should have clear guidance addressing:
• which AI tools are approved for professional work;
• what types of client information may or may not be entered into those tools;
• what AI may be used for, such as drafting, extraction, summarization, or first-pass review;
• what AI may not be used for, including final professional judgment or unsupported standards conclusions;
• how AI outputs must be verified;
• when AI use should be documented in the file;
• how hallucinated, unsupported, or unverifiable outputs should be handled; and
• who remains accountable for the final report.
That answer should be the same in every file: the CBV who prepares or signs the valuation conclusion is accountable.
The CBV Institute has begun addressing AI through existing standards and general guidance⁸ . Detailed AI-specific protocols are not yet in place. That gap has begun to close: on July 6, 2026, the CBV Institute released an Exposure Draft of a revised Code of Ethics that would make these obligations explicit. Members would remain responsible for work quality regardless of the technology used, including AI, and would be required to assess whether reliance on AI tools is appropriate⁹.
For users of valuation reports, AI-assisted review should be understood as a quality-control aid, not an assurance mechanism. A report reviewed by AI may still contain unsupported assumptions, incomplete corroboration, or judgmental issues that require professional assessment.
The Takeaway
AI has an important place in valuation practice. The question is not whether to use it, but whether the reviewer and report preparer are clear on what job it is actually doing.
Professional judgment is not pattern recognition. It means reading the file skeptically, challenging the evidence, and being accountable for the conclusion. An AI model cannot currently provide this.
The technology is here. The professional judgment about how to use it responsibly remains ours to exercise.
Footnotes
¹ Retrieval-augmented generation is a method of configuring an AI model to search a specified body of documents and incorporate retrieved content into its responses, rather than drawing solely on its training data. This allows the model to be grounded in current, practitioner-supplied materials such as the applicable practice standards.
² A context window is the maximum volume of text an AI model can process in a single session. Where a model's context window is sufficiently large, the full text of applicable standards or reference materials can be supplied directly in the prompt, allowing the model to reference them without a separate retrieval mechanism.
³ Source: CBV Institute (2025, November). Countdown to 2026: Scope of work – Credible and properly supported valuation conclusions. https://cbvinstitute.com/countdown-to-2026-scope-of-work-credible-and-properly-supported-valuation-conclusions/
⁴ Source: Dahl, M., Magesh, V., Suzgun, M., & Ho, D. E. (2024). Large legal fictions: Profiling legal hallucinations in large language models. Journal of Legal Analysis. https://arxiv.org/abs/2401.01301
⁵ Source: Ernst & Young (2026, January). Managing hallucination risk in LLM deployments at the EY organization. https://www.ey.com/en_gl/technical/enterprise-solution-guides-technology-leaders/managing-hallucination-risk-in-llm-deployments-at-the-ey-organization
⁶ Source: CBV Institute (2025). Countdown to 2026: PS 100 – The new foundation. https://cbvinstitute.com/countdown-to-2026-ps-100-the-new-foundation/
⁷ Source: CBV Institute (2026, January). CBV Institute’s updated Valuation Practice Standards now in effect. https://cbvinstitute.com/news_article/cbv-institutes-updated-valuation-practice-standards-now-in-effect/
⁸ Source: CBV Institute (2024, June). Primer on artificial intelligence: Essential considerations for CBVs on the responsible use of AI. https://cbvinstitute.com/wp-content/uploads/2024/06/AI-Primer-June-2024-Final-EN.pdf
⁹ Source: CBV Institute (2026, July). Revised Code of Ethics – Exposure Draft. https://cbvinstitute.com/wp-content/uploads/2026/07/Exposure-Draft_Code-of-Ethics_June-10-2026-1.pdf
About the Author
Vince Osbourne, CMA (US), CBV, ABV is a Director at Kalex Partners Inc. with over 20 years of experience in business valuation, corporate finance, financial planning and analysis, and strategic financial leadership. Most recently, he served as Chief Financial Officer and Corporate Secretary of a mineral exploration company. Vince is a Chartered Business Valuator, holds the Accredited in Business Valuation designation, and earned a Bachelor of Economics from York University.
Comments