> ## 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.

# Three AI failures the customer saw first
- URL: https://intelligence.artificialrealitymedia.com/three-ai-failures-the-customer-saw-first/
- Published: 2026-08-06T09:00:00.000Z
- Updated: 2026-08-06T09:00:00.000Z
- Description: A concierge that could not name a restaurant, an answer circled in pen, and a $5 million valuation of a business earning $125,000. One shared mechanism.
- Author: Galina Fendikevich
- Tags: Cross industry, #For AI Buyers, #For AI Companies, #What's not working, #Import 2026-09-03 04:53

Three AI failures reported across Artificial Reality's first season share one mechanism: the output reached a customer with no expert in between. A hotel concierge that could not name a nearby restaurant. A yacht charter client told he could anchor in 25 knots of wind. And a $5 million valuation, brought to M&A broker Christine McDannell, of a business earning $125,000 in net profit.

None is a story about model quality. All three are stories about placement.

## The concierge that could not recommend a restaurant

*Provenance note: this case was described by Galina Fendikevich to Sofiane Ghorbel during his episode, from a case study she had read. Artificial Reality has not independently sourced it, and reports it here as it was discussed on air.*

A major hotel chain launched an AI concierge to handle guest recommendations. A reporter tested it by asking for restaurants nearby, having landed late and wanting to eat. It could not answer.

What happened next is the instructive part. By the same account, the company pivoted the technology to revenue management, demand forecasting and back-end operations, and recovered what the chatbot had cost. Same organisation, same technology. One deployment became a story about AI failing in public. The other stopped being a story at all.

Sofiane Ghorbel, twenty-five years in luxury hospitality, gives the reason his industry treats this asymmetrically: you do not sell rooms, cruises or flights, you sell trust. One wrong recommendation can undo a relationship built over years. In his framing an AI mistake is not a bug, it is relational, and it costs the client. His formula: the winners will be "the companies that combine high tech plus high touch," because "technology create\[s\] efficiencies, humans create memories."

## The answer the client circled in pen

A yacht charter client asked [ChatGPT](https://chatgpt.com/?ref=intelligence.artificialrealitymedia.com) whether it was safe to anchor in 25 knots of wind. It said yes. The captain, who had sailed those waters for years, said absolutely not. The client screenshotted the answer, circled it with a pen, and called his broker, furious.

The AI had no access to the anchorage, the seabed, the boat or the forecast, and gave a confident answer anyway. What made that a business problem rather than a bad answer is that it reached a customer with no expert between the two.

Jennifer Kerum's own AI deployment, in the same business, sits on the other side of the line for a structural reason: her translation happens inside her own inbox, in a message she composes and sends, not in a channel the guest queries.

## The $5 million valuation

A husband-and-wife team brought an AI-produced valuation to an M&A brokerage: $5 million. Their actual net profit was $125,000.

Christine McDannell's assessment, delivered before the meeting so she would not have to say it to their faces, was a three-to-four-times multiple on net profit. On $125,000, that is roughly $375,000 to $500,000 — an order of magnitude below what they had been told.

The mechanism matters more than the number:

> "It was taking every single department in their company and it was putting a value from the payroll... It assigned a value of every department, even the accounting department, like 200,000 or something. That's how they got to this five million number."

The AI had counted expenses as assets. Her diagnosis of why it produced a wrong answer rather than no answer:

> "It wants to make you happy."

*Note on the figures: the $5 million valuation and the $125,000 net profit are stated by McDannell. The three-to-four-times multiple is her stated guidance to the clients; the resulting range is arithmetic. The recap episode's "closer to $400,000" falls inside that range but is not a figure she states — treat it as the host's rounding rather than the broker's number.*

This is a front-of-house failure, not a tool failure: McDannell uses the same class of tool constantly and successfully. She cut payroll by 83%, and on pricing: "When business owners complain about the $100 a month thing, I'm like, legit, this will cost me $40,000." The difference is that she is a trained M&A professional reading the output. The couple were not, and nobody told them the answer might be wrong.

## Why these three are different from ordinary tool failures

- The recipient could not tell a reliable answer from an unreliable one.
- The AI expressed no uncertainty in any of the three.
- The cost landed on a relationship or a decision, not on a subscription.
- The same technology worked for the same buyer once it was pointed at the back office.

## AI tools named in this report

| Tool             | Named by                                   | Verdict                                          | Used for                                               |
| ---------------- | ------------------------------------------ | ------------------------------------------------ | ------------------------------------------------------ |
| ChatGPT          | Jennifer Kerum's client, yacht charter     | Didn't work — confidently wrong                  | Whether it was safe to anchor in 25 knots              |
| No product named | Christine McDannell, M&A broker            | Didn't work — $5M against $125,000 net profit    | A valuation produced for the owners                    |
| No product named | Reported on air, not independently sourced | Failed in the lobby, paid off in the back office | A hotel chain's concierge, then its revenue management |

*Tools named by operators on the record. Inclusion is reporting, not endorsement.*

## What this means if you are building

The demonstrations are front of house. The concierge, the assistant, the chat interface: visible, impressive, easy to put in a keynote. That is where the marketing budget goes, and by the account of six operators across seven industries, it is where the money gets lost.

So ship the advisory answer to the professional the customer already pays, not to the customer. The valuation, the recommendation and the safety call each had a competent reader one seat away, and each product skipped that seat.

## Where this comes from

S1E8: Why 99% of AI products fail — the full interview with Galina Fendikevich. Listen or watch: [YouTube](https://www.youtube.com/watch?v=WLzJ7U6inpw&ref=intelligence.artificialrealitymedia.com), [Spotify](https://open.spotify.com/episode/1DI6RkvekF4NF7HfpuzEqm?ref=intelligence.artificialrealitymedia.com) or [Apple Podcasts](https://podcasts.apple.com/us/podcast/why-99-of-ai-products-fail-the-gap-between/id6780699025?i=1000780319385&ref=intelligence.artificialrealitymedia.com).