The Job You Hate Most May Be the Wrong One to Automate First

The job you hate most is a fine place to start looking and a poor way to choose. How a job feels at the end of a long day tells you nothing about whether software can do it, whether the information it needs is written down anywhere, or what one wrong answer would cost. Five questions do. If you are already comparing tools, looking for something cheaper than Zapier, park that for a week: the right tool depends on the process, and so does whether you need a new tool at all.

Ask five questions of every job on your list

• How many hours a week does it take? Count them for one ordinary week with tally marks instead of guessing.
• Does it repeat, the same way each time? Software is good at the hundredth identical case and poor at the one that is different every time.
• Is it rules or judgment? Prices, opening hours and free slots are rules. A complaint, a refund or a health question is judgment, and judgment stays with a person.
• Is the data already there? If the information lives in a calendar, a price list or the invoicing tool, software can read it. If it lives in one person's head, writing it down comes first.
• What does a mistake cost? A wrong reminder costs an apology. A wrong payment or a wrong tax filing costs far more, so those come later, with a person approving each one.

The best first candidate has many hours, many repeats, clear rules, data already written down, and mistakes that are cheap to catch.

Four example businesses, run through the test

These come from our demos. The businesses are invented and every figure below is an example, worked out on its demo page.

• A clinic front desk (example). Every new patient fills in the same forms on paper and someone types them in later: 40 new patients a week at 9 minutes each is 6 hours a week. It scores high on hours and repeats, and the data starts with the patient, so the patient enters it once, on a tablet. The cost of a mistake shapes the build: the invoice waits until every consent it needs is signed. See the clinic demo.
• A salon booking on WhatsApp (example). The same questions about prices and free slots, typed by hand between clients: about 60 messages a day at 2 minutes each is 2 hours a day. The rules are the price list and the free slots, so an assistant can answer and book, and a complaint or a health question goes straight to a person. See the salon demo.
• Invoicing (example). Every order typed again into the invoicing tool, then each invoice uploaded to Romania's e-Factura system by hand: about 100 invoices a month at 4 minutes each is about 7 hours a month. The order already holds the data, and a mistyped price costs a credit note, a new invoice and a new upload, so the build takes the price from the order. In your own setup, keep a person approving each upload until it has run clean. See the invoicing demo.
• Document intake (example). An accounting office types supplier invoices into its register field by field, about 6 minutes each. Read automatically, the same invoice takes about 20 seconds, and a person checks only the fields the system is unsure about. That person is the answer to the cost-of-a-mistake question. See the intake demo.

Pick one, and keep a person on it

Start with the candidate that does best on all five, and automate that one thing. For the first weeks, a person reads what it did every day, and the automation reports what it changed on every run. We wrote about catching the automation that says it worked when it did nothing. Once the first one has run clean for a month, pick the next.

To run the test on your own list, use our free Owner's Automation Worksheet: a week of tally marks, a score from 1 to 3 on each of the five questions, and a check for silent failures in what already runs.