Your Chatbot Did Half My Job This Week. Here Is the Half It Could Not Do.

What we learned when a business owner ran his key person insurance question through ChatGPT before he ever picked up the phone.

The request that landed in our inbox

It arrived on a Tuesday. A specialty manufacturer, roughly $8 million in revenue, twenty-two employees. The owner wanted to talk about key man insurance on his operations lead.

The email was unusually good. Organized. It named the person, described what he does, estimated what his departure would cost, and listed the documents the owner assumed we would want. Most first inquiries do not look like that. Most first inquiries look like "what does key man insurance cost?"

He had put his situation into ChatGPT first, worked through what it gave him, and cleaned it up before sending.

It is worth being clear about our reaction: we were glad. Not threatened — glad. And also, by the end of the meeting, very clear about where the line is.

What the AI got right

A fair amount, and it deserves credit.

•      It gave him vocabulary. Owner, beneficiary, insurable interest, notice and consent. He could follow the conversation from the first minute instead of the fifteenth.

•      It made him identify the actual key person. Not the highest-paid person. The person whose absence changes what the company can deliver.

•      It made him quantify — roughly. He arrived with an estimate of revenue that runs through this one employee. Rough is fine. Rough is a starting point.

•      It gave him a documented list of what we’d need. Financials, comp detail, estimated health and age, etc.  We did not have to spend a week collecting.

That is real value. It compressed our discovery meeting from ninety minutes to forty, and the forty minutes we did spend were better ones. The client was not learning terminology; he was thinking.

And the three rabbit holes

However, his “AI search” also spent his time on things that did not matter to his situation. Not because it was wrong — because it had no way to know which of the many possible outcomes and options applied to him.

Rabbit hole 1 — chasing the perfect valuation formula

He had spent hours comparing methods: multiple of salary, multiple of contributed profit, discounted future contribution. He came in ready to defend a methodology. That is precision applied to the wrong question.

The question is not "what formula values this employee?" It is "what does this business need cash to do, and for how long, if he is gone on Monday?" Answer that, and the number falls out of it.

Rabbit hole 2 — term versus permanent, before the purpose was settled

He arrived with a product opinion. But the product decision is the last decision, not the first. Duration, purpose, and who owns the policy come first. Decide that the need runs eight years and disappears, and the answer is one thing. Decide the need converts into a buyout obligation later, and it is a very different thing — and, as it turned out, that was his situation.

Rabbit hole 3 — worrying about a court case that did not apply to him

He had read about the Supreme Court's Connelly decision and was concerned that a business-owned policy would inflate the value of his company for estate tax purposes.

Traditional key person insurance, owned by and payable to the business with no buy-sell component attached, does not create a Connelly problem.

Here is the uncomfortable part. He did have a Connelly exposure. It was sitting in a stock redemption buy-sell agreement in a drawer, signed years ago, funded with company-owned insurance a document he never mentioned, because he had not asked the chatbot about it. The AI answered the question he brought. It could not ask him the question he did not know to bring.

The pattern worth noticing

An AI is superb at answering the question you asked. It has no way to notice the question you should have asked instead because it cannot see the folder in your drawer, hear the hesitation in your voice, or know that the person you are insuring will be at your Thanksgiving table.

The sentence that changed everything

Twenty minutes into our conversation, he mentioned that his operations lead is his son-in-law.

He did not say it as a disclosure. He said it as an aside, the way you mention a detail you assume is irrelevant. It was the most important sentence of the meeting.

Because now this is not a question about replacing an employee. It is a question about who owns this company in fifteen years, whether the owner's other two children expect a share, whether the son-in-law can fund a purchase, and what happens to family relationships if the transition is unfunded and improvised at the worst possible moment.

Key person coverage still belongs in the plan — the business does need to survive a sudden loss. But it is now just one piece of a structure, not the whole answer. We’re now starting conversations around buy-sell arrangements and future planning.  Conversations that should include is advisor, attorney and his CPA at the table, not a policy quote.

No chatbot was ever going to get there. Not because the model is not smart enough — because he was never going to type that sentence into it.

The detail that would have quietly broken it

One more thing. For an employer-owned life insurance policy to pay income tax-free to the business, specific notice and consent requirements must be satisfied before the policy is issued. The AI mentioned this. It was mentioned in the way a small footnote mentions things and gets disregarded.

Nobody executes a footnote. Miss that step and the death benefit - the entire reason the policy exists - can become taxable to the business at exactly the moment the business can least afford it.  Making sure that step actually happens, in writing, in the right order, is not advice. It is execution. It is what an advisor is for.