Where the time actually goes.
Start with one task that has a written output and a low cost of being wrong. Then do it again next week and see whether it still works.
Pick one task you do every week, that produces something written, where a wrong answer would be obvious. Do it with AI until it works properly rather than until it demos well. Then do it again the following week. The second week is where most firms find out whether they built a workflow or a party trick, and it is the step almost everyone skips.
Worth doing first.
Meeting prep, not meeting notes
Notetakers get the attention, but the expensive hour is the one before the meeting: pulling the CRM record, the last review, the portfolio drift and the open items into one page. That is a retrieval-and-assembly job with a fixed shape, which is exactly what these tools are good at. Make it cite where each fact came from, or you have moved the checking rather than removed it.
CRM hygiene, once
Duplicate households, inconsistent field values, dead email addresses, beneficiary fields that were never filled. This is the least glamorous project in any firm and the one that makes every later automation possible, because automation on bad data produces confident wrong answers faster.
The age-milestone monitor
A rule that watches dates of birth against the list below and raises a task. It is simple enough to build in a weekend and it catches the thing nobody catches: a client who turned 63 and nobody ran the IRMAA conversation.
Billing reconciliation
Quarterly, cyclical, tedious, and error-prone in both directions. Worth automating because the errors are symmetrical — you are as likely to have undercharged as overcharged, and both are conversations you would rather not have.
Ages that should raise a task.
Point a rule at date of birth and have it open a task ahead of each of these. Lead time matters more than the trigger — a task that fires on the birthday is already late for most of them.
Ages and thresholds move with legislation. Confirm the current rules before you wire them into anything that fires without a human reading it.
The invisible errors.
Automation is safe where a wrong answer is visible and cheap. A drafted email gets read before it goes out, so a bad draft costs nothing. A mis-parsed beneficiary designation does not get read by anyone, and it surfaces years later at the worst possible moment.
Sort candidate tasks by that question before you sort them by time saved. The tasks worth automating first are rarely the ones that take longest — they are the ones where being wrong announces itself.
Time your own baseline.
Published claims about advisor time savings range from three hours a week to twenty. That spread tells you the studies are measuring different things — some count drafting only, few subtract the review time that AI output creates.
Time one task for two weeks before you change anything, then time it again after. One number you produced beats every number you can cite, and it is the only one that survives someone asking how you know.
Get the next one.
Field notes from real deployments, written the way this one was. No schedule promised and nothing sold to you. Unsubscribe whenever.
Before you point this at a client file.
Read client data and consent first. Which tool the data goes into is a decision worth making before the workflow exists, not after.
Keep reading
Build with us in Milwaukee.
Thirty minutes, one real constraint, and a written description of what could be built. No pitch decks, no small talk.