Pilot Before Plumbing

Run a process manually for 5 to 10 cycles and automate only the part that stays stable.

Your consulting practice just landed its fifth client this quarter, and you are itching to build the automated intake system: forms, routing rules, a welcome email sequence, maybe a client portal. You have blocked a full weekend for it. Meanwhile you have onboarded exactly four clients, each one differently, and you cannot yet say which parts of the process actually repeat.

The rule here is to run the process manually for five to 10 cycles before you build anything permanent. The cue is a tempting automation that has not yet survived real variation. Do it by hand, log what varies from case to case, and automate only the stable core that appears across every cycle. When you do build, require a kill switch, a named owner, a review trigger, and a manual fallback so a broken automation cannot silently run. What changes: instead of a weekend spent plumbing the wrong pipes, you automate one small piece that genuinely repeats, and it holds up. Skip the build entirely when the repeated benefit does not clearly exceed the setup and maintenance cost, which is the honest answer for most processes you run fewer than 10 times a year.