AI The Impact of
on ESOP Transactions and Administration
COREY ROSEN
CONTENTS Introduction......................................................................................... 1 AI and Professional Services Generally.............................................. 2 AI and Legal Services......................................................................................................... 2 AI and Valuation.................................................................................................................. 2 AI and Software.................................................................................................................. 3
AI and ESOPs...................................................................................... 4 Feasibility Studies............................................................................................................... 4 Financing............................................................................................................................. 5 Plan Structure and ESOP Legal Advice.............................................................................. 5 Trustee Issues..................................................................................................................... 5 Appraisal/Fairness Opinion................................................................................................ 5 Plan Administration............................................................................................................ 5 Repurchase Analyses.......................................................................................................... 5 Look for Domains Where 90% Accuracy Is OK.................................................................. 5 Irreducibly Human Inputs................................................................................................... 6 The Bottom Line................................................................................................................. 7
About the NCEO and the Author....................................................... 8
The National Center for Employee Ownership 510-208-1300 customerservice@nceo.org www.nceo.org Copyright © 2026 by the National Center for Employee Ownership and the author. All rights reserved. This publication is designed to provide accurate and authoritative information in regard to the subject matter covered. It is sold with the understanding that the publisher is not engaged in rendering legal, accounting, or other professional service. If legal advice or other expert assistance is required, the services of a competent professional person should be sought.
INTRODUCTION Generative AI is seen as a potentially transformative technology in part because it promises to reduce the time and expense needed for service-intensive projects. Selling a business to an ESOP is certainly service-intensive, and finding ways to make the process quicker, easier, and less costly would be a real boon for selling business owners and would-be employee owners. As the NCEO explored in our paper Why Selling to an ESOP Costs Less than Selling to Another Buyer: Breaking Down the Cost Differences, ESOP transactions are typically much less costly and often no more complicated than selling to another buyer; still, relative to setting up another retirement plan, ESOPs are complex and expensive, and this can be a barrier for companies that would otherwise be good ESOP candidates. Once it is set
up, the annual costs of running an ESOP are modest, but there are still many important administrative tasks that take time and careful attention. How much of a difference can AI make for ESOPs? As is true for most of the promised benefits of AI at this point, it can be difficult to separate hype from real hope. Predicting the future of a new technology in any detail is impossible, and AI is moving especially fast; trying to foresee AI’s impact on ESOPs or anything else would be an act of arrogance more than insight. Still, we can describe where the consensus of opinion lies right now. We believe the most likely scenario is that AI will speed up and improve some tasks but will not significantly reduce costs or the time required for ESOPs.
THE IMPACT OF AI ON ESOP TRANSACTIONS AND ADMINISTRATION
1
AI AND PROFESSIONAL SERVICES GENERALLY Before looking specifically at ESOP issues, it is useful to look at how AI is beginning to affect the broader domain of professional services in law and finance. Any ESOP-specific impacts will likely be secondary to these wider economic effects.
AI and Legal Services There is now a considerable literature on the expected impact of artificial intelligence on the legal profession, including the fees that will be charged. Much of this literatureis written by people in the field who are clearly concerned about the possibility that AI could reduce the need for legal services and/or reduce legal fees. So far, there is no evidence that AI is reducing legal fees, and it may be increasing them. Law firms may use AI to increase efficiency and quality but may end up charging more for their work product. Axiom, a company that provides contract legal researchers, found in a 2026 study of over 600 law firms in eight countries that 79% of law firms are using AI to provide services, “but only 6% pass those savings to clients—while 34% actually charge premium rates for AI-enhanced work.” Fifty-eight percent have not reduced their rates when using AI assistance. The study concluded that “rather than viewing these efficiency gains as an opportunity to provide better value to clients, many law firms appear to be treating AI as a profit multiplier. The technology allows them to complete work faster while maintaining—or even increasing—their billing rates.” Another legal services company, Thomson Reuters, found in its 2025 Future of Professionals report that 74% of firms reported using AI. But the authors went on to argue that AI will not reduce legal fees for non-routine matters:
General-purpose AI tools, however capable they have become, are not sufficient for fiduciary-grade work. Where outputs influence legal judgments, financial disclosures, regulatory filings, or client advice, ‘almost right’ is simply not good enough. In the moments that matter, this means AI must be built on authoritative, domain specific content; rigorous privacy and security; subject-matter expertise; outputs that are transparent and verifiable; and access to real-time human support.
In a 2026 article in Bloomberg Law, Justin Daniels and Linda Orton argue that legal services that are “repetitive, high-volume matters increasingly will compress toward efficiency. AI will make that work faster, cheaper, and more automated, and clients should expect to capture much of that value. Bespoke legal judgment is different. Complex negotiations, AI governance, regulatory ambiguity, strategic transactions, cyber risk, and novel legal issues often depend on the ability to synthesize incomplete information, identify second-order consequences, and exercise experienced judgment under uncertainty.” AI will improve quality, but not costs, on these matters. Close to good enough is not good enough in legal matters. There is an entire dataset of “legal decisions in cases where generative AI produced hallucinated content”; as of mid-August 2026, there are 1,313 cases in the US alone. Because therisks are so high, considerable time needs to be spent verifying what AI produces.
AI and Valuation Another professional service likely to be heavily affected by AI is valuation. Appraisers must gather and analyze often a very large volume of data from multiple sources. This is the sort of task that AI is well suited to perform with both speed and accuracy. AI
THE IMPACT OF AI ON ESOP TRANSACTIONS AND ADMINISTRATION
2
also promises to reduce errors that could be introduced as a result of human judgment in selecting comparable companies and relevant data comparisons. AI will also be helpful in trend analysis. As with all AI, however, the usefulness of AI will depend on how good the prompts are and how well appraisers are able to identify when information is inapplicable or inappropriate for the valuation. Finally, AI can quickly and accurately review written material for inconsistencies and errors. While all of these tools will speed up some aspects of business appraisals in general, much of the work of valuation is based on judgment and conversations with company leaders. As in law, AI is not currently expected to significantly reduce appraisal costs for complex valuations.
AI and Software One inarguable impact of AI is that it has made software easier and cheaper to create. More people without a background in software development are using AI to make custom tools that would have required an expensive team and lengthy timelines in the past. The jury is out on the pitfalls of the “vibe-coding” approach to software and how large the net overall cost savings will be. AI can generate
impressive prototypes and small tools at astonishing speed, but creating enterprise-grade, maintainable, secure software still requires expert humans. Initial savings on development may bring increased maintenance costs later. Nonetheless, AI seems to have increased the productivity of professional human developers and is certainly causing a rapid increase in the quantity of software. This means that any aspect of ESOPs amenable to software solutions will likely be explored by new market entrants lured by the possibility of providing more efficient and hence cheaper services than incumbents. Some companies setting up an ESOP may find success creating their own custom software to facilitate aspects of the process, such as gathering data for a feasibility study or appraisal, making financial models, or crafting employee communications, something some ESOP companies are already doing, both for written and visual presentations. To some extent, AI is changing the nature of what is possible in software by allowing custom-coded tools to be created and iterated quickly by people who are domain experts rather than software experts. It’s too early to foresee whether this “software on demand” vision will cash out in real productivity gains, but it is a possibility.
THE IMPACT OF AI ON ESOP TRANSACTIONS AND ADMINISTRATION
3
AI AND ESOPS It is not difficult to imagine a near future in which ESOP providers will run the details of the client’s scenario through an AI program that will quickly generate a deal structure. It will still need to be reviewed by the professionals, but the process could save a considerable amount of time. At this point in AI, however, reviewing and modifying these documents for the almost inevitable errors and oversights may take as much time as if the professionals generated the material themselves. This is especially true in complex transactions. That may not affect fees, however. Companies may pay for the establishment of an ESOP by paying an hourly fee for services, but fees are more commonly set on a project basis. If that is the case, it seems highly likely at this point that project fees will not be affected by AI for most ESOP services. The firms providing these services, however, may be able to do the work more efficiently, but as the example of what has happened in the legal industry suggests, that appears unlikely at this point to lower the fees that they charge. In fact, firms may argue that AI enhances the quality of their product. Looking at ESOP tasks component by component can clarify where AI can make a difference.
Feasibility Studies Detailed feasibility studies often involve a week or more of consultant time. Elements of the study include: •
How can the deal be financed? (Exploring and modeling options for bank debt, seller notes, warrants, cash contributions to the plan, mezzanine debt, and SBA backing)
•
S vs. C issues: 1042 or not, tax issues on conversion, and how each choice affects financing, especially if the ESOP is not buying 100%
•
How much should (or can) the ESOP buy?
•
Multi-stage transactions vs. 100% transactions
•
Financial feasibility, including balance sheet and loan covenant impacts (modeling the impact of different possible financial futures)
•
Seller note issues (rate, whether it would be repaid after senior debt, and term)
•
Impact on family members, if any
•
Plan rule alternatives
•
Comparison to a non-ESOP sale
•
Equity rights or direct equity for key people
•
Preliminary valuation?
•
Impact on other benefits/compensation, including what ESOP accounts might look like over time
•
Payroll issues
•
Anti-abuse issues, if applicable
•
Section 1042 investment options, if applicable
Feasibility studies typically cost between $30,000 and $60,000. AI can be especially helpful at the prefeasibility stage. AI programs can ask for input from the seller and come up with a reasonable first pass on whether an ESOP is worth considering. AI could also make data gathering faster. We expect that there will be increasingly sophisticated interactive platforms that allow business owners to do prefeasibility assessments with detailed results. These assessments, however, will not substitute for expert advice, both because AI systems are still likely to miss some key issues and because getting advice from experts will still be necessary because each case is different. Moreover, existing non-AI technology already allows sellers to use spreadsheet models to do a preliminary financial assessment of feasibility. The NCEO, for instance, has an online ESOP calculator. The most intensive work of a feasibility study, however, involves multiple conversations with
THE IMPACT OF AI ON ESOP TRANSACTIONS AND ADMINISTRATION
4
owners, the CFO, and often other officers of the company to consider multiple scenarios. AI could help with data gathering and analysis, and write a draft final report, but much of this is already done by making adjustments to boilerplate forms and language.
Financing Structuring and (if a seller note does not completely finance the deal) finding financing adds another layer of time and costs, including bank loan negotiations, seller note structuring (such as how much should be in warrants versus interest payments), looking for non-bank financing alternatives, and preparing lending documents. AI could help generate the documents, but not much else.
Plan Structure and ESOP Legal Advice Writing the plan document will not be significantly impacted by AI because these documents already are based on a law firm’s readily adaptable templates. Most of the work is conversations with the relevant players to decide what plan features to adopt.
Trustee Issues Trustees have to read and review the appraisal report and use that as a basis for negotiating the final terms of a sale. These are all matters of judgment and conversations AI is not likely to make any easier, although it could help the trustee assess the validity of some of the data in the report and comparable company transactions.
Appraisal/Fairness Opinion AI systems could help ensure that all inputs are properly assessed and may be able to quickly analyze large datasets to look for such things as comparable companies, as well as potential inconsistencies or errors in reports and documents. This may have more of an impact on accuracy than costs.
Plan Administration Much of this work is already automated, with most of the time spent entering data. AI could help write draft reports or forms, such as sophisticated communications material for employees. It is likely that AI will improve the quality and sometimes speed of these processes, but the costs are already not significant for most ESOP companies.
Repurchase Analyses This is the one area where AI has already made an impact. Commercially available software can be purchased for companies to do their own analyses, and some ESOP providers have developed their own AI models for their clients. While these programs provide a great deal of user flexibility to run multiple analyses, they are not cheaper than traditional assessments. Over time, there is a possibility that there will be more readily available AI programs, although companies would still be well advised to get professional consultation on how to enter the inputs and assess what to do with the numbers.
Look for Domains Where 90% Accuracy Is OK Much has been made of AI “hallucinations,” fabricated facts or faulty conclusions from premises that make the AI’s output incorrect, and which, as noted, have already caused many embarrassing incidents in legal proceedings. AI by itself should never be relied on for any decision with meaningful stakes, much less decisions with legal implications, of which there are many in creating or maintaining an ESOP. Even if hallucinations continue to decrease in quantity and severity with new AI models, it’s likely to remain the case that an expert human will need to carefully review AI outputs to guard against fabrication or error. This significantly reduces, if not eliminates, the prospect of using AI to replace human subject matter experts. However, even though AI-generated text can never be assumed to be fully correct, there are plenty of domains where this actually doesn’t matter. In the
THE IMPACT OF AI ON ESOP TRANSACTIONS AND ADMINISTRATION
5
right context, perfect accuracy can be less important than the ability to process huge amounts of information and make many good-enough judgments, provided they can be easily verified or disproven. For example, take sales prospecting. A team of AI agents can research thousands of companies to identify those that may be good fits for a particular product, winding up with a short list of a few dozen leads. Even if the error rates are substantial, this can be extremely useful. False positives (companies the AI flagged as a lead but that in fact are bad fits) will be quickly identified by the human salesperson acting on the leads, while false negatives (great fits that were missed by the AI) may be found later through some other process, and their absence doesn’t cancel out the benefit from the high-quality leads the AI did find. There are many other domains that have analogous properties, such as idea generation. AI can come up with a hundred ideas in seconds; if one or two are valuable and can be identified as such, it doesn’t matter that the rest were poor, or that not every possible good idea was found. In these examples, making some or mostly correct calls, acting on a high volume of information, is what matters. Note that in both examples, a critical element is that a human can discern what parts of the AI output to discard. We asked Claude to think about examples of this “quantity beats quality” property of AI in an ESOP context. Here is one of its several thoughts: Before a company even commits to an ESOP, advisors often need to review years of cap tables, prior valuations, corporate governance documents, and existing benefit plans to spot issues (e.g., prior IRC §409A problems, existing plan documents that might conflict with ESOP qualification, related-party transactions that will complicate the trustee’s fiduciary analysis). An AI can be turned loose on hundreds of pages of historical documents to flag “here are 40 things that might be relevant to feasibility” — a false positive costs a lawyer five minutes of “no, that’s fine”; a false negative might get caught at the next review pass, and the cost of missing something is bounded by the fact this is a preliminary scan, not the final compliance determination.
Readers can decide for themselves if this is persuasive and what that says about LLMs either way.
Irreducibly Human Inputs Setting up an ESOP, like any business deal, requires a large number of decisions. AI can play a role in gathering data to inform these decisions, and AI personas will happily offer “their” advice if asked, but ultimately AI cannot reduce the number of decisions that must be made, or, put another way, the number of bits of information that must be supplied to determine the final shape of the transaction and the plan. Barring a wholesale transformation of the legal system, these decisions must be supplied by humans. AI can shrink the cost of gathering information, but it cannot shrink the cost of deciding what to do with it, nor can it play the crucial role of humans as the holders of accountability for those decisions. For example, take a relatively basic decision: What vesting schedule the ESOP should use. Under ERISA, the plan document must provide for three-year cliff vesting, six-year graded vesting, or a schedule more generous than either. At the end of the day, the humans responsible for approving the final plan document will need to make this decision, weighing factors such as workforce demographics and turnover, forfeiture timing dynamics, and administrative convenience. AI may be able to synthesize the available data and provide a recommendation to inform the decision, but that recommendation will ultimately just be one more data point for the human decision-makers to consider. It is possible that the more people feel they must rely on AI advice to make these decisions optimally, the more they will feel the need to invest in infrastructure needed to supply the AI with as much relevant context as possible, which could actually increase the total cost and complexity of the process. In fiduciary decisions, this limitation on AI’s role becomes even sharper, because a key purpose of the fiduciary role is to provide a locus of legal responsibility so that a suit can be brought if the decision is contested. For example, even if useful roles can be found for AI in creating appraisal reports, fiduciaries will still be required to prove that they followed a prudent process in overseeing the valuation. To take another example, companies decide how to fund the repurchase obligation and how conservatively
THE IMPACT OF AI ON ESOP TRANSACTIONS AND ADMINISTRATION
6
to invest for it. AI can help create and refine modeling software to explore demographic and valuation scenarios effectively, and a business can certainly attempt to feed enough financial and workforce data into an AI to let it weigh in on the decision. But the ultimate decision is a judgment call about risk appetite that only the board and the company leaders can make. If there is a failure to fund the obligation, they and the trustee will be sued. “The AI said this was OK” is unlikely to become an acceptable legal defense anytime soon.
between the relevant parties or decisions that only humans can make. AI can help provide the inputs, but much of this is already automated through existing technologies and templates. AI may be very useful for companies in running their own scenario planning and in producing software, documents, videos, graphics, podcasts, and other communication materials, however. We are well-advised to heed Yogi Berra’s sage advice that “it’s tough to make predictions, especially about the future.”
The Bottom Line The impact of AI is likely to be limited for a few key reasons. Much of the work involves conversations
THE IMPACT OF AI ON ESOP TRANSACTIONS AND ADMINISTRATION
7
About the NCEO and the Author We are the National Center for Employee Ownership (NCEO), a nonprofit organization that has been supporting the employee ownership community since 1981. We have thousands of members because we help people make smart decisions about employee ownership, with everything from reliable information on technical issues to inspiration to help companies reach the full potential of employee ownership. We generate original research, facilitate the exchange of best practices at our live and online events, feature the best and most current writing by experts in our publications, and help employee ownership companies build ownership cultures where employees think and act like owners. Whether you are considering employee ownership, managing an existing plan, or advising clients, we can help. Our members have access to all of our online resources, and we are committed to providing extensive materials for anyone interested in learning more about employee ownership, from people considering employee ownership to decision-makers at employee-owned companies to ESOP participants to journalists to stock plan administrators and other service providers. Our staff covers the nation from multiple locations in the US, and our board includes representatives from employee-owned companies and the professional advisors who serve them. We are supported almost entirely through membership fees and our activities, but we do accept donations, which are tax-deductible (we are qualified as a 501(c)(3) nonprofit charitable organization).
Corey Rosen Corey Rosen is the NCEO’s founder and former executive director and now is its senior staff member. Corey has spoken on various subjects related to employee ownership all over the world with government, business, and union leaders, and he is regularly quoted in leading magazines and newspapers. He has appeared on national television and radio programs and also has authored four books on employee ownership, plus more than 100 articles for various business, academic, and professional publications. He has authored or coauthored several of the NCEO’s practical and research publications. He holds a PhD in political science from Cornell University.
The National Center for Employee Ownership 510-208-1300 customerservice@nceo.org www.nceo.org