Automate the process, not the mess

The trap

A property manager in Surrey has a spreadsheet. Every time a tenant applies, someone copies their name, phone, email, and lease start date into it by hand. Takes about four minutes per application. The manager thinks: if I automate this, I save four minutes per application, maybe thirty applications a month, that's two hours a week back.

So they build a tool that reads application emails and fills the spreadsheet automatically. For the first week, it works. Then a tenant submits an application with their name spelled two ways. The tool picks the wrong one. Another application has the lease start date in a different format. The tool misses it. Now someone has to check every entry the tool creates, which takes almost as long as doing it by hand. The automation that was supposed to save two hours a week now creates two hours of checking work.

This happens because the process itself was never clean. Nobody had written down the rule for which name spelling to use. Nobody had set a standard for date format. The process worked when a person did it because a person can see context and make judgment calls. Automation can't.

What breaks

When you automate a messy process, three things go wrong.

First, hidden steps come to light. A dental clinic in Kelowna automates appointment reminders. The system sends a text the day before each appointment. But the clinic's schedule changes constantly—the hygienist calls in sick, the dentist double-books by mistake, patients cancel at the last minute. The automation doesn't know about these changes and sends reminders for appointments that don't exist anymore. Patients get confused. Staff get angry calls.

Second, exceptions become visible. A plumber in Richmond automates invoice creation. Most jobs fit the template: labour, materials, tax. But some jobs involve warranty work, or customer credits, or materials that weren't used. The automation creates invoices that are wrong, and someone has to fix them manually after the fact. The tool has made the work visible but not solved it.

Third, decisions get buried. An accounting firm in Burnaby automates expense categorization. The tool reads receipts and sorts them into categories: meals, travel, office supplies. But sometimes a receipt is ambiguous. Is a coffee shop meal or a meeting space. The tool guesses. It guesses wrong about 15 percent of the time. Now an accountant has to review every categorization, which means they're not saving time—they're just doing the work in a different order.

How to fix it first

Before you automate, spend a week documenting the process as it actually happens. Not as you think it happens. Actually.

Pick one person who does the task regularly. Sit with them for an hour. Watch them do it three times. Write down every step, including the ones they do automatically and don't mention. Write down the decisions they make and why. Write down the exceptions they handle and how.

Then ask: Is this step necessary, or is it leftover from an old system? Can we combine two steps into one? Is there a decision here that could be a rule? Is there an exception that happens often enough to have its own rule?

A 12-person software company in Gastown automated their onboarding checklist. Before they built the tool, they watched a team lead onboard a new hire. The process had 47 steps. When they asked why, the lead said some were legacy—they used to matter but didn't anymore. Some were in the wrong order. Some were waiting on information that arrived late because nobody had asked for it at the start. After they cleaned it up, the process had 23 steps in a better order, with information requested upfront. Then automation made sense. It sent checklists, tracked progress, and reminded people of deadlines. The tool worked because the process was real.

Once you've documented the process, get agreement. Bring in the person who does the work, the person who receives the output, and the person who pays for it. Walk through the steps together. Agree on what the right answer looks like. Agree on what counts as an exception and how to handle it. Get this in writing, even if it's just a shared document.

Then automate the parts that are repetitive and have no exceptions. Keep a person in the loop for decisions and edge cases.

Where to start

• Pick one task that takes 30 minutes to 2 hours a week, happens the same way most of the time, and involves copying, checking, or filing. Watch someone do it twice and write down every step without editing.

• Ask the person doing the work: What parts always stay the same? What changes? What decisions do you make? Write down the answers and any rules they mention, even if they seem obvious.

• Meet with the person who does the work and the person who uses the output. Agree on what the process should be, write it down, and identify which steps are repetitive enough to automate and which need a human decision.