Technology

Four things it does well.
Three it does not.

Most of what advisors are sold as AI is a general model with a prompt in front of it. The value is in what you use it for.

Where it genuinely helps today

Most of what advisors are sold as artificial intelligence is a general model with an industry prompt in front of it. That does not mean the underlying tools are useless — it means the value is in what you use them for, not in who wrapped them.

Here is where practitioners are getting real returns, and where they are not.

The four that work

Drafting things you already know how to write. Meeting follow-ups, client explanations, the paragraph explaining a rebalance. You are not asking it what to think; you are asking it to produce a first draft of something you would have written more slowly.

Making complex things plain. Turning your explanation of a concentrated stock strategy into something a client's spouse can follow. This is genuinely one of the highest-value uses and almost nobody does it deliberately.

Summarising long documents you have. A plan document, a trust, a lengthy prospectus. With the crucial caveat that you verify anything you rely on, because summarisation errors are confident and invisible.

Preparing for meetings. Give it the situation and ask what questions a thoughtful client would raise. It is a rehearsal partner, not an answer.

The three that do not

Anything requiring current facts. Rates, thresholds, contribution limits, tax brackets. Models state outdated figures with complete confidence and that is a client-facing error waiting to happen.

Recommendations. Not because the output is bad, but because the reasoning is not yours and you are the one with the fiduciary obligation. You cannot defend a recommendation you did not make.

Anything client-identifying, in a tool you have not assessed. This is the one that becomes a disclosure problem rather than a quality problem.

Use it for the work you could do yourself but slower. Do not use it for the work you are paid to be accountable for.

The discipline that separates useful from dangerous

Give it context rather than questions. A vague prompt produces a plausible generic answer; a prompt containing the actual situation, constraints and audience produces something usable.

And review every word before it reaches a client. Not because the output is usually wrong, but because the failures are fluent, and fluent errors are the hardest kind to catch on a second reading.

The question that comes before all of it

Where does the data go, and have you assessed it? Every tool is another boundary to establish and keep current, which is why the number of tools matters as much as the choice of them.

Writing the policy →

Questions this did not answer? Ask them directly — that is what the twenty minutes is for.

Book 20 minutes with Kyle