Pick the process that happens most often, has a correct outcome you can describe in one sentence, costs real money or hours when it goes wrong, and lands on the people closest to your customers or revenue. Score a shortlist against those four tests and the first system usually picks itself. How loudly someone complains about a process in a meeting is not one of the tests, and it shouldn’t be.
How do you choose the first process to automate?
Most teams start with the wrong question, which is “what annoys us most?” The answer to that is usually whatever went wrong last week, and it tends to be a rare, dramatic process rather than the expensive one. We run the same short test with every client instead: frequency, clarity, cost of error, and who feels the pain.
The mechanics are simple. Sit down for an hour with the people who actually do the work and list every repetitive process you can think of, usually somewhere between eight and fifteen. Put them in a Google Sheet, score each one from 1 to 5 on the four criteria, and add up the totals. It’s the same approach an industrial engineer takes to a production line: measure where the time goes before you decide what to change, because the bottleneck is rarely where people say it is.
The first process to automate is rarely the one that hurts most in the meeting. It’s the one that quietly costs you every single day.
Frequency: why a daily task beats a painful quarterly one
Frequency comes first because it multiplies everything else. A quarterly report that takes two people three days to assemble feels terrible, but it works out to about 48 hours a quarter, or 16 hours a month. Compare that with copying orders from WhatsApp into a spreadsheet or ERP: at 15 minutes per order and 20 orders a day, across 22 working days, that’s 110 hours a month. Nobody holds a meeting about it because each order only takes a few minutes, which is exactly why it survives.
Frequent processes are also better first projects for a practical reason. They give you feedback fast. If you automate something that runs fifty times a day, you’ll know within a week whether the logic holds up, where the edge cases are, and whether the team trusts it. Automate a quarterly process and you wait three months to find out whether it worked.
Clarity: can you describe the correct outcome in one sentence?
The second test is whether you can write down what “done correctly” looks like without arguing about it. Compare these two descriptions. “Every lead from the website form is in Pipedrive within five minutes, assigned to the right rep by region, with a WhatsApp confirmation sent to the lead.” That is a process you can automate this month. “Approve discounts that make sense for the account.” That one depends on who you ask, and automating it would simply hard-code one person’s opinion.
Tools like n8n, Make and Zapier execute rules very reliably, but they don’t invent them. Models like Claude or ChatGPT can handle the judgement steps inside a process, such as classifying an incoming email or extracting fields from an invoice, yet they still need a clear definition of a good answer to be checked against. If a process fails the clarity test, the first job is to define it and give it an owner. That’s valuable work, but it isn’t your first automation.
Cost of error and who feels the pain
Put the cost of error in dollars or hours, not in a shrug. An invoice that goes into Xero or QuickBooks with the wrong amount means a credit note, an awkward call and a late payment. A website lead that sits untouched for two days is very often a lost deal. A formatting mistake in an internal report costs almost nothing. A high cost of error makes a process more valuable to automate, because people make mistakes on repetitive work when they’re tired or rushed, but it also means you design in a check. For a first build, that usually means a human approval step in Slack before anything reaches a customer.
The last test is about who carries the load. The process that costs the business most is not always the one the loudest person in the room is complaining about. Salespeople retyping leads lose selling time. Customers waiting a day for a routine answer don’t complain, they just leave. In one of our projects, a customer-service portal that consumed hours of manual work now resolves routine requests automatically, which returns about 30 hours a month to the team.
FAQ
What if the most frequent process is also the messiest?
Then clarity is your real first project. Spend a week or two writing down the rules, agreeing on the correct outcome and naming an owner, and run it manually that way. Once the process behaves consistently by hand, automating it becomes straightforward.
Should the first automation be small or ambitious?
Small enough to ship and prove within a few weeks. The first system’s job is to earn the team’s trust and give you a clean before and after. You can connect more steps once the first one runs without anyone babysitting it.
Do I need AI for my first automation?
No. Many of the best first automations are plain rules moving data between tools you already use, such as a form, a CRM and a spreadsheet. Add an AI step only where a human currently reads something and makes a judgement call.
How do I know if the first automation worked?
Measure the baseline before you build: hours spent per week, error count, and response time. Measure the same three numbers a month after launch. If you didn’t write down the baseline, you’ll be arguing from memory.
Where to start this week
Block one hour with the people who do the repetitive work, list every process that runs at least weekly, and score each one on frequency, clarity, cost of error and who feels the pain. Take the top scorer, write its correct outcome in one sentence, and record how many hours it takes today. That gives you a scope and a baseline. If you want a second pair of eyes on the shortlist, this scoring session is the first step of every JARA engagement.