> ## Content Index > Fetch the complete content index at: https://intelligence.artificialrealitymedia.com/llms.txt > Use this file to discover other available public pages before exploring further. # The $5 million valuation on $125,000 of net profit - URL: https://intelligence.artificialrealitymedia.com/ai-valuation-5-million-ma-mistake/ - Published: 2026-09-02T12:02:00.000Z - Updated: 2026-09-02T12:02:00.000Z - Description: An AI report valued a business at $5 million on $125,000 of net profit by counting expenses as assets. The broker's guidance was a 3-4x multiple. - Author: Galina Fendikevich - Tags: M&A advisory, #For AI Buyers, #What's not working, #Import 2026-09-03 13:12 A couple walked into an M&A brokerage holding an AI-generated valuation of $5 million for a business earning $125,000 in net profit, with no intellectual property and no real estate. The model had counted their expenses as assets. Christine McDannell, who owns The Magnolia Firm, a sell-side M&A brokerage in Austin, told them her own guidance is a three-to-four-times multiple on net profit. On $125,000 of net profit, that multiple is $375,000 to $500,000 — Artificial Reality's arithmetic from the range she gave. It is roughly an order of magnitude below the report they arrived with. ## How did an AI tool reach $5 million? By adding up departments. McDannell read the report and found it had assigned a dollar value to each department in the business, apparently derived from payroll, and totalled them. > "It was taking every single department in their company and it was putting a value from the payroll, I guess, of the company. It assigned a value \[to\] every department, even the accounting department, like 200,000 or something. That's how they got to this five million number. And I was like: these departments cost — these are expenses, these are value adds. It did some weird stuff." That inverts the underlying accounting. An accounting department is a cost the business carries, not an asset a buyer purchases. Artificial Reality records that McDannell attributes this particular report to [Claude](https://claude.ai/?ref=intelligence.artificialrealitymedia.com) Cowork, per what the sellers told her. ## How did the broker catch it? With one question, before the meeting. She checked the only input that could plausibly justify a number that size. > "I wrote them right back and I'm like: do you guys own the real estate for your business? And they were like, no." Then she prepared them rather than correcting them live: > "I emailed them immediately before the Zoom meeting, because I'm like — oh my God, I don't want to say this to their face. I need to prepare them. And I was like: hey, I'm gonna be honest, you're looking at like a three to four X multiple on your net profit." She has seen this more than once, and notes the sellers themselves laughed about it on the call. ## Why do AI valuations come out too high? Because the model is agreeable, in her diagnosis, not because it is stupid. > "It wants to make you happy. It wants to make you happy — your company's worth five million. Like, yeah. That's the problem." Her fix, which she says she has not applied to her own setup, is instruction-level: > "The trying-to-make-you-happy thing is you have to push back on it. Or you have to put in your global instructions, whatever LLM you're using ... push back on me, be contrarian." Hers is "just normal," because she trusts her own realism as the check: "If it kicks me a crazy number, then I'm just like: okay, no." Galina Fendikevich, who hosts Artificial Reality, the AI buyer-evidence podcast, offered a more optimistic read in the same conversation — that she has noticed models becoming much more direct in response to user complaints about sycophancy. The $5 million report suggests that improvement has not reached the place where it does the most damage. ## Why is this failure harder to spot than most? Because nothing in the output looks wrong. The report had departments, line items and a total. It read as structured and reasoned. A seller with no M&A background has no way to know that summing the payroll cost of departments is not a valuation method. The format and the confidence are the danger, not a visible mistake. And the tool is most dangerous for exactly the people most likely to reach for it — owners without an advisor, trying to find out what their life's work is worth. The same category of tool works in McDannell's own hands, on the same underlying models, because she supplies what the couple could not: "It does a good job if you have the right input. And you have to know the industry to know the inputs in the prompts." ## Would an M&A-specific model fix it? She advises against building one now. > "There's some people making them right now. I feel like they're advancing so fast ... I would not build one right now, to be honest, if it was me personally, because it's advancing so quickly, and it's scraping the same data. I just don't know why it's not compiling it correctly. I think it's gonna get there too soon, too quick." Her own approach is the opposite of a new model: encode the expertise into a detailed skill running on a general-purpose one. ## What can an owner check before trusting a number? Three things this case exposes, none of which require an advisor to run: - **Ask the tool for its method, not its number.** If the answer is a sum of departmental values, the arithmetic has already gone wrong. - **Check what it treated as an asset.** Expenses counted as value adds is the specific error McDannell found. - **Instruct it to push back.** Her stated remedy is a global instruction to be contrarian, which she has not needed because she can spot a crazy number herself. Artificial Reality reports what happened in this exchange; none of it is valuation or financial advice. But the failure mode generalises past M&A, because a model trained to please will flatter any high-stakes, once-in-a-lifetime number an ordinary person asks it for. ## Where this comes from S1E5: How she cut her payroll by 83% using one AI agent — the full interview with Christine McDannell. Listen or watch: [YouTube](https://www.youtube.com/watch?v=1VO7nWtzNEU&ref=intelligence.artificialrealitymedia.com), [Spotify](https://open.spotify.com/episode/6E4sAUyfk4X5gUqEtzhL8o?ref=intelligence.artificialrealitymedia.com) or [Apple Podcasts](https://podcasts.apple.com/us/podcast/how-she-cut-her-payroll-by-83-using-this-one-ai-agent/id6780699025?i=1000776953676&ref=intelligence.artificialrealitymedia.com).