Technology & AI
What to automate first, and what to leave alone
Firms automate what is annoying. The better order is what is frequent, understood, and cheap to get wrong.
Published
Three tests, in order
How often does it happen. How well do you understand it. What does an error cost.
Frequent, well-understood, and low-cost-to-get-wrong is where to start. Rare, ambiguous, and expensive is where firms usually start, because that is what hurts.
Frequency is the multiplier
A weekly task automated saves fifty-two times. A quarterly one saves four. Founders reliably prioritize the quarterly task because it is more painful each time it occurs.
Understanding is the precondition
Automating a process nobody can describe produces a system that does something nobody can describe, faster.
If the process cannot be written down in a page, write it down first. That step alone frequently removes most of the pain and sometimes removes the need to automate.
Automating a process you do not understand is how you get a machine that is confidently wrong at scale.
Error cost sets the review requirement
Generating an internal summary and sending a client communication have different consequences when wrong. The second needs review in the loop; the first does not.
This determines the shape of the automation, not whether to do it.
A reasonable first list
Meeting preparation and follow-up. Data collection into a single place. Scheduled reporting. Reminders and the compliance calendar. Onboarding document collection.
All frequent, all well understood, all low cost to get wrong.
What to leave alone for now
Anything producing something that looks like advice. Anything where the rule differs per client. Anything you do twice a year.
The last one is not because it is hard. It is because it will never repay the effort, and the project will quietly consume a quarter.