Compliance
AO-41: An AI Tool Can Draft Your 409A. It Can't Be the Appraiser.
The Appraisal Standards Board says a tool cannot comply with USPAP, only an appraiser can. What AO-41 means for founders buying an AI-only 409A valuation.
By 409.AI Team - 2026-08-21
# AO-41: An AI Tool Can Draft Your 409A. It Can't Be the Appraiser.
The quote reads like a bargain. Fully automated 409A, no calls, no intake questionnaire, a report in about an hour, and $0 on the entry tier. Then comes the sensible question: is it actually defensible?
Until recently that was hard to answer with anything sturdier than professional instinct. Then the body that writes the appraisal profession's standards answered a version of it in writing.
On April 23, 2026, the Appraisal Standards Board of the Appraisal Foundation adopted [Advisory Opinion 41, Use of Technology in an Appraisal or Appraisal Review Assignment](https://appraisalfoundation.org/pages/uspap). It covers automated valuation models, regression and statistical software, and generative AI, across real property, personal property, and intangible property. That last category is where your 409A lives.
The line at the center of it is short. A tool cannot comply with USPAP. Only the appraiser can.
What AO-41 actually is, and what it isn't
AO-41 doesn't create new obligations. It explains how the ones already in the Uniform Standards of Professional Appraisal Practice apply when an appraiser reaches for software, and it was written to consolidate guidance that had been split across two older advisories: AO-18 on automated valuation models and AO-37 on computer-assisted valuation tools ([Texas Appraiser Licensing and Certification Board](https://www.talcb.texas.gov/public/news-articles/asb-releases-second-exposure-draft-technology-use-appraisal-assignments)). Both predate generative models entirely, which is why the refresh was needed.
It is also, importantly, not a prohibition. AO-41 assumes appraisers use these tools and tells them how to do it without breaking anything. Nothing in it says a model can't run a backsolve, screen comparables, or draft narrative sections. What it says is that the model's output is information, not a conclusion, and it stays information until a named human verifies it, applies judgment, and decides whether relying on it produces a credible result.
That distinction is the one founders are groping toward when they compare valuation providers and can't articulate what feels off about the cheapest quote.
Why an appraisal standard matters to a tax question
Section 409A is Treasury's rule, not the Appraisal Foundation's, and no IRS examiner is going to open a notice by citing AO-41. So the connection needs explaining.
The most common route to 409A safe harbor is the independent appraisal. Treasury Regulation [§1.409A-1(b)(5)(iv)(B)(1)](https://www.govinfo.gov/content/pkg/CFR-2025-title26-vol6/pdf/CFR-2025-title26-vol6-sec1-409A-1.pdf) points to an appraisal meeting the requirements of [IRC §401(a)(28)(C)](https://www.govinfo.gov/content/pkg/USCODE-2024-title26/html/USCODE-2024-title26-subtitleA-chap1-subchapD-partI-subpartA-sec401.htm), which defines an independent appraiser as one meeting requirements similar to those in the regulations under section 170(a)(1). Follow that one more step and you land in [26 CFR §1.170A-17](https://www.law.cornell.edu/cfr/text/26/1.170A-17), which sets out what a qualified appraiser is: an individual with verifiable education and experience in valuing that type of property, qualifying either through coursework plus two or more years of relevant experience, or through a designation awarded by a generally recognized professional appraiser organization on the basis of demonstrated competency.
The same regulation contains a definition most people never read. Generally accepted appraisal standards, it says, means "the substance and principles of the Uniform Standards of Professional Appraisal Practice, as developed by the Appraisal Standards Board of the Appraisal Foundation."
So Treasury has already named USPAP as its benchmark for what competent appraisal work looks like, in the regulation the 409A safe harbor reaches back to. And the designation route to qualified-appraiser status runs through professional bodies whose members are bound to USPAP as a condition of holding the credential. USPAP Standards 9 and 10 are the ones that govern the development and reporting of business and intangible asset appraisals, which is the work a 409A is.
None of that makes AO-41 binding tax law. It does mean that when the standard-setter tells credentialed business appraisers what they are personally answerable for, it is describing the obligations of whoever signs your report. Our breakdown of [what actually earns 409A safe harbor](https://409.ai/articles/409a-safe-harbor-price-vs-qualified-appraiser) covers the regulatory conditions in full; AO-41 fills in the professional side of the same picture. If you're newer to this, start with [what a 409A valuation is](https://409.ai/articles/what-is-a-409a-valuation-a-comprehensive-guide).
The output is a number. The judgment is the product.
Consider what a 409A actually involves once the cap table is loaded.
Say you closed a $22 million post-money Series A six weeks ago. A backsolve solves for the total equity value at which the new preferred prices out to exactly what investors paid, then an option pricing model allocates that value across the classes. To run it, someone has to choose an expected time to liquidity, a volatility estimate drawn from a peer set, and a discount for lack of marketability. Change the term from 2.1 years to 3.5, the volatility from 55% to 70%, or the [DLOM](https://409.ai/articles/discount-lack-marketability-dlom-409a-valuation) from 22% to 32%, and the common stock price moves materially. There is no arithmetic that produces those three inputs. There is only defensible reasoning about a specific company.
The same is true one level up. Choosing [between an OPM, a PWERM, and a backsolve](https://409.ai/articles/409a-allocation-methods-opm-pwerm-backsolve) is a judgment about how much visibility the company has into its exit paths, and a wrong choice is not a rounding error.
AO-41's framing handles this cleanly. Assignment results are the appraiser's opinions and conclusions. A tool's output isn't one until the appraiser has decided it should be. Whether that decision happened, and on what basis, is the difference between a valuation and a calculation with a logo on it.
Four obligations that don't transfer to software
AO-41 works through the existing USPAP rules that come under pressure when a tool enters the workflow. Four are worth knowing as a buyer.
Competency. An appraiser is not expected to reverse-engineer a model's internals. They are expected to know enough about a tool to interpret its output, recognize where it is likely to be wrong, and judge whether relying on it is appropriate in this assignment. Using something capable without understanding its limits is itself a competency problem.
Scope of work. If the tools available can't support a credible result, the appraiser's options are to change the scope, renegotiate, or walk away. Proceeding anyway is not one of them. This is quietly the sharpest test you can apply to a vendor, and the next section turns it into a question.
Record keeping. When a conclusion rests on a tool's output, the workfile needs to show it: the output, the data behind it, the prompts or instructions given, and enough evidence to reconstruct how it fed the conclusion. That is the same file an examiner asks for. Our piece on [what the IRS requests in a 409A exam](https://409.ai/articles/irs-audit-409a-valuation-document-request) lists the documents that actually get produced, and "the platform generated it" has never been an entry on that list.
Ethics and confidentiality. This is the one founders underrate. A 409A intake means handing over your cap table, your term sheets, your board deck, and often your financial model. If that material is typed into a third-party system, the appraiser has to know what the system retains and what it may transmit, and get client consent before it goes in. AO-41 treats using a system you know may leak confidential information as a failure of due care, not a technicality. If a vendor can't tell you where your data goes, that isn't a privacy footnote. It's a standards problem for the person who signed.
The question a fully automated pipeline can't answer
Vendor evaluation gets easier when you stop asking about capability and start asking about refusal.
Every experienced appraiser has declined work. The financials didn't reconcile, the founder wanted a number before the analysis started, the company's structure fell outside their competency, the eligibility conditions didn't hold. AO-41 codifies that instinct: when the work can't be done credibly, you decline, withdraw, or change the scope.
A pipeline that guarantees a report in an hour has no path for that. It has no state for "this engagement shouldn't be completed." Which means the only place a decline can originate is a human with the authority and the incentive to stop the process, and that human's willingness to say no is the actual thing you're buying.
So the useful questions to a prospective provider are narrow ones:
What happens when your reviewer disagrees with the model's output? Ask how often it has happened and what changed as a result. A review that can only approve isn't a review.
What sits in the workfile if this gets examined in three years: the inputs, the model version, the prompts, the reasoning?
Where does my cap table go once I upload it, what retains a copy, and does anyone need my consent before it is sent anywhere?
Under what circumstances would you decline this engagement? Listen for a real answer with real conditions in it.
What to do with a report you already have
If a past valuation came out of a process with no identifiable appraiser behind it, the exposure isn't theoretical. A grant priced below defensible fair market value stops being an exempt option and becomes deferred compensation under §409A, and the [IRS Nonqualified Deferred Compensation Audit Technique Guide](https://www.irs.gov/pub/irs-pdf/p5528.pdf) is explicit that the resulting additional taxes fall on the employee rather than the company. Nobody on your team chose the appraiser, and none of them will be consulted about who pays.
That is fixable, and much cheaper to fix before anyone asks. Our guide to [correcting an incorrect 409A](https://409.ai/articles/dealing-with-incorrect-409a-valuations) walks through the options. It's also worth reading alongside the [AICPA's first full rewrite of the cheap stock guide since 2013](https://409.ai/articles/aicpa-cheap-stock-guide-2026-update-409a-valuation), because the two developments point the same direction: the profession is tightening its expectations about documented judgment at exactly the moment the tooling is getting good enough to make skipping it tempting.
The takeaway
AO-41 didn't rule on whether AI belongs in valuation work. It ruled on who answers for the output, and the answer is a person with a name, a credential, and something to lose. Software can shorten every step between intake and draft, and it should. It cannot hold the responsibility, because responsibility is not a feature that ships.
When you next compare 409A quotes, find the signature page before you find the price. If there's a qualified individual there who reviewed the inputs, chose the method, could have refused the engagement, and keeps a workfile that shows their work, the automation underneath is a speed advantage. If there isn't, you didn't buy an appraisal. You bought a number, and the person who pays for the difference is on your payroll.
If you want to see how that division of labor is supposed to look, our [409A valuation service](https://409.ai/products/409a) pairs an AI-drafted analysis with a credentialed reviewer who signs it.