Most writing on how to automate client onboarding counts the time saved and not the time spent keeping it working, which is why so many firms have an automation nobody dares change. This page is about the second number: which rules age badly, what a change costs once several are chained together, and which parts are worth deliberately leaving manual because a person is cheaper than a rule. It assumes you have already decided to automate something.
The rules that age badly
Anything encoding a person, a tool or a price. A rule that routes to a named colleague breaks when they leave; one that posts into a specific channel breaks when the team reorganises; one that branches on a service tier breaks when you rename the tiers. These are the rules that quietly stop firing, and a rule that stops firing is worse than no rule because everyone still believes it is working.
What a change costs later
More than it did the first time, and non-linearly, because chained automations have to be traced before they can be edited. A firm with four linked rules changes them in an afternoon; a firm with twenty schedules the work. That curve is the argument for keeping the number of rules small and the steps between them explicit.
What to leave manual on purpose
The accept decision, the kickoff, and any step whose failure is expensive and silent. Silence is the key word: an automated step that fails loudly is fine, one that fails by simply not happening is how a client goes three weeks without the thing you promised while your dashboard reads green.
Measure before and after, in days
Calendar days from signature to first delivered work, per client, before and after. Minutes saved is the number vendors quote and the one nobody feels; days is the one the client experiences. The free worksheet here gives you the before figure from your own steps and volumes.
Questions people ask about automate client onboarding
How many automation rules is too many?
When somebody has to draw a diagram to change one. That is the point where the maintenance cost has overtaken the saving.
Should automation ever email the client directly?
Only for things you would be comfortable sending unread, which in practice means requests and reminders, not answers.
What breaks first?
Rules that name a person. Staff change more often than processes do.