The best tasks to automate aren't necessarily the most annoying ones. They're the ones that are repetitive, predictable and based on clear rules.
I normally think about business automation in three stages:
1. Automate collection.
Stop asking people to repeatedly copy information between systems. Forms, APIs, webhooks and integrations can move structured information automatically.
2. Automate detection.
This is where automation becomes much more valuable. Instead of someone constantly checking for something, let the system identify when attention is required: a deadline is approaching, a number crosses a threshold, a payment changes, or an expected event doesn't happen.
3. Keep humans for decisions.
Not every detected event should trigger an automatic action. Cancelling a supplier, approving significant spending or changing an important client process often deserves human review.
Recurring software costs are a simple example. We're building LemSubs around this kind of workflow: structured subscription data can generate events through webhooks or be queried by AI agents. Instead of manually checking every subscription, an operator can focus on questions such as which important renewals are approaching or what changed in a client's software stack.
My rule is simple: automate observation before automating judgment.
Removing repetitive data collection and monitoring usually saves time with relatively little risk. Automating every decision just because the technology allows it can create entirely new problems.
