The clock starts when the system goes live, not when the engagement starts. Before that there's a mapping stage: looking at the manual work in question and being honest about what a system can and cannot own, because a system that only covers part of a job only pays back on that part. Once it's live and running on its own, payback comes from two places: hours a person no longer has to spend on the task, and things that start happening that weren't happening before because there was never enough time for them, like following up every lead instead of just the ones someone got to.
Frequency matters more than size. A system that runs against something continuous, checking inbound messages all day, publishing on a schedule, watching a feed for something worth reacting to, starts contributing the day it goes live, because it's doing that work every day from day one. A system built for an occasional decision, or a full app that needs its own users or traffic before it produces anything, takes longer to pay back, not because it's weaker but because the underlying work happens less often or needs time to build up. Precipitate runs on this same pattern internally: over a hundred scheduled jobs operating around the clock across dozens of live integrations, some producing content in multiple languages every day, others checking something once a week. The daily ones earn their keep fast. The occasional ones take longer, by design.
It's a poor fit for work that happens rarely or leans on a relationship a system can't stand in for: a handful of high-stakes calls a quarter, judgment calls that need context nobody has written down, anything where the volume is too thin for automation to ever catch up to what it took to build. Before starting, look at how often the task actually happens and how much of it repeats the same way each time. A task done occasionally by one person doesn't clear that bar. A task happening daily, weekly, or across many instances at once usually does.