Your second automated step: 4 checks
In the previous lesson your monthly review produced numbers telling you the first automated step has settled. This is where the common mistake starts: an owner who has seen the result wants to automate everything at once, and ends up with three half-finished steps instead of one that works well. Real expansion moves one step at a time, and the question is not what could be automated, but which one deserves to be next.
A real-world example
An air conditioning maintenance company switched on its first step three months ago: a customer sends a WhatsApp message such as "the living room unit is not cooling", the system opens a ticket and replies with its number and the next available slot. The step settled and corrections became rare. The owner now has three candidates for step two: issuing the invoice after a ticket closes, scheduling technicians and assigning tickets to them, or reminding customers when their periodic service is due.
The emotional pick is scheduling, because it drains him most. But it also depends most on information that lives nowhere in the system: which technician is near that district, who knows that model, who called in sick today. The invoice, by contrast, has a clear and complete input: a ticket closed, a part replaced, labour hours recorded. He picked the invoice, not because it matters most but because it is the most ready.
Four checks before you choose
- The first step has genuinely settled: corrections dropped two months in a row. Building on a shaky step multiplies the faults and hides where they came from.
- One clear input, one output: you can write it in a single line — if this arrives, that comes out. If you need a paragraph of branching conditions, the step is not ripe yet.
- The data is in the system, not in people's heads: every piece of information the step needs is already recorded. What one supervisor knows from experience is not a valid input for an automated step.
- Errors are cheap and reversible: a draft invoice corrected before sending costs far less than a message that already reached the customer. If the error is expensive, make the step stop and wait for your approval.
Apply the four checks to the three candidates and the gap shows immediately:
| Candidate step | Clear input | Data ready | Reversible error |
|---|---|---|---|
| Invoice after ticket closes | Yes | Yes | Yes, draft reviewed before sending |
| Technician scheduling | No, branching conditions | No, technician skills unrecorded | No, an appointment already promised |
| Periodic service reminders | Yes | Partly, contract dates incomplete | No, a message already sent out |
The table did not kill scheduling, it postponed it. It becomes valid the day technician skills and districts are recorded, and that groundwork can start today without any automation.
Two weeks running in parallel
- Start with one type only: residential tickets, and leave corporate contracts for later.
- Run the step alongside the manual way for two weeks: the system prepares the draft, the accountant reviews it before sending, and logs every time they disagreed with it.
- Set one success measure and a number for it before you begin, such as 8 out of every 10 drafts passing with no edit.
- If the measure is met, widen by one type at a time. If it is not, the fault is usually in the rule rather than the idea: read the disagreement cases, fix them, then repeat the two weeks.
Do not switch off the manual step on the first day the measure is met. Keep it one extra week so a busy week passes through it, because problems appear under pressure, not on quiet days.
When to postpone expansion
Three situations mean this is not your moment: the first step still produces corrections every week, your team still works around the system in the old way, or the candidate step itself is not documented as a clear manual procedure. The third is the most deceptive: what you cannot write down as a written procedure cannot become automatic, because automation executes a rule, it does not invent one.
And when the second step succeeds, connect it to the one before it as a single flow instead of leaving it as an island: a ticket opens, closes, and its invoice goes out and lands in the accounts, as in WhatsApp invoicing and running your business from WhatsApp.