RMD season is the true test of an AI prompt

Come December, clients usually see the same required-minimum-distribution reminder from three institutions before their advisor's lands. By then it reads like a fourth notice for a deadline nobody was going to miss. The message is not the problem; the list is. Fifty clients are getting the same one.

Seasonal campaigns such as RMDs, tax-loss harvesting windows, and annual beneficiary reviews share a shape that daily client emails don't. They have one core message, one hard deadline, and a recipient list that may run from twelve households to twelve hundred. An advisor with a working prompt for a single client reply often assumes it scales. It doesn't, not until the prompt itself is rebuilt.

The fourth installment of WealthTech Today's AI Prompt Lab, published this week, takes up that batch problem. The series follows prompts advisors actually run. Earlier installments covered the daily briefing, client emails that sound like you, and quarterly review letters. This one is about a message that has to go to fifty clients at once without reading like a form letter.

The piece draws one line between one-to-one email prompts and batch prompts. A one-to-one prompt tunes tone to a specific relationship. A batch prompt must hold a fixed skeleton — compliance language, call to action, the firm's voice — plus three or four per-client variables: account type, distribution amount, deadline, prior-year action. The starter prompt in the piece is built around that split. It takes a CSV export from whatever CRM or portfolio system you use and runs the same in ChatGPT, Claude, or other models.

The risks are concrete. Send a Roth conversion reminder to a client who has no Roth and the relationship takes a hit. Get the tone wrong and fifty people receive a form letter with their name pasted at the top — a reputational cost the article argues is worse than sending nothing. The fixed-skeleton structure addresses both. What never changes stays locked, what changes per client stays in the variable slots.

Strip away the prompt and this is a practice-management lesson. Well-run client service has always used a fixed skeleton, variable slots, and a compliance rail. Artificial intelligence does not change the logic; it changes what the logic costs to run. A firm that treats AI as a one-off email generator misses the batch use case where AI delivers the most value and carries the most compliance risk.

Each installment — daily briefing, client emails, quarterly reviews, seasonal batch — is a master prompt structure rather than a single prompt. The seasonal one is the most demanding because it combines a deadline, a distribution list, and a compliance guardrail at once. Get the separation right once, and the same structure runs three tax seasons in a row with only the input data changing.

The catch is maintenance. A fixed skeleton is only as good as the compliance language inside it, and that language changes. Set this up once, forget it, and a regulator's mid-season disclosure update will force a rewrite. The data side has the same problem: a CSV export pulls today's numbers, not last year's, and the prompt will confidently send a stale distribution amount to fifty households if the input is stale.