Most owners I talk to want to start automating back office tasks by buying a big platform and pointing it at the whole company. That is exactly how automation projects stall. The tools are fine. The problem is picking the wrong first job, one that is either too messy to encode or too rare to bother with.
The better move is to treat your first automation like a pilot, not a transformation. Pick one narrow, boring, repetitive task. Prove it saves real hours. Then use that credibility to fund the next one. I have watched a five-person finance team go from skeptical to asking for more, all because the first automation returned obvious time within two weeks.
This piece gives you the selection rule I use, one worked example end to end, and the single mistake that quietly wastes the most money.
The Selection Rule: High Frequency, Low Judgment, Clear Rules
Not every task is worth automating. A good first candidate scores high on three things at once.
- High frequency. It happens daily or weekly, not once a quarter. Frequency is where the savings compound. Automating something that runs four times a year is rarely worth the setup effort.
- Low judgment. The task follows a decision anyone on the team would make the same way. If two staff members would handle it differently, you have judgment, and judgment resists automation.
- Clear rules. You can write the steps as an if-this-then-that list without hand-waving. "Send a reminder three days after the due date if unpaid" is a clear rule. "Follow up when it feels right" is not.
Run your candidate list through those three filters. The tasks that pass all three are your starting shortlist. Almost always the winner is something nobody enjoys: invoice reminders, copying data between two systems, or assembling the same weekly report.
Here is a quick scoring frame you can use in a spreadsheet:
| Task | Frequency | Judgment needed | Rules clear? | Good first pick? |
|---|---|---|---|---|
| Invoice payment reminders | Daily | Low | Yes | Strong |
| Re-typing orders into accounting | Daily | Low | Yes | Strong |
| Monthly board narrative | Monthly | High | No | Weak |
| Approving unusual refunds | Weekly | High | No | Weak |
One Example End to End: Invoice Reminders
Let me walk one candidate the whole way, because the details are where people get stuck.
A distributor in Bekasi had one admin staffer spending roughly two hours every day checking which invoices were overdue and sending WhatsApp and email reminders by hand. She cross-referenced a spreadsheet against the accounting system, wrote each message, and logged what she sent. Some days she skipped it entirely because she was covering the front desk.
The current process, written as rules:
- Every morning, list invoices past their due date and still unpaid.
- Group them by how overdue: 1 to 7 days, 8 to 30 days, over 30 days.
- Send a polite reminder for the first group, a firmer one for the second, and flag the third for a phone call.
- Log the date each reminder went out so nobody gets messaged twice in a day.
Every step there is low judgment and clearly ruled. That is why it automates cleanly.
The build. This did not need artificial intelligence or an expensive suite. A scheduled job pulled unpaid invoices from the accounting system each morning, applied the aging buckets, and sent templated messages through the existing messaging channel. Total setup was about three days of one developer's time, or you can assemble a rougher version with a low-code tool if you have no engineer handy. If that route interests you, I wrote a separate piece on how to build internal tools without hiring a developer.
The payoff. The two daily hours dropped to about fifteen minutes of the admin glancing at the flagged over-30-day cases. Reminders now go out every single day instead of "when there is time," and days-sales-outstanding improved because nobody slips through. At a fully loaded cost of maybe Rp 4 million a month for that admin's time, recovering roughly a third of her day pays for the build inside the first month, and every month after is pure gain.
That is the shape you want: modest effort, obvious and recurring payoff, no drama.
The Mistake That Wastes the Most Money
Here is the warning that matters more than any tool choice: do not automate a broken process.
If your invoice data is inconsistent, your reminders will go out with wrong amounts and wrong names, and you will have automated your embarrassment at scale. Automation is an amplifier. Point it at a clean process and it multiplies good work. Point it at a broken one and it multiplies mistakes faster than a human ever could, which is worse because nobody is watching each case anymore.
So before you automate, spend a day fixing the underlying process by hand. Standardize the data. Agree on the rules out loud. Only then encode it. This is the same reason I keep telling people to fix their data before buying AI: the tool is never the weak link, the input is.
A second, smaller trap is automating something you are about to change anyway. If the finance team is switching accounting software next quarter, do not spend three days wiring reminders into the old one. Sequence your automation behind your stable processes, not ahead of them.
The Practical Takeaway
Start small and specific. Score your repetitive tasks on frequency, judgment, and rule clarity, and pick the one that scores high on all three. Fix the underlying process by hand first, then encode it. Prove the time saved in the first month, and let that win fund the next automation.
You are not trying to automate the company. You are trying to win one clean, boring, recurring task and build trust from there. If you would rather have a second pair of eyes on which task to pick first, that is exactly the kind of scoping conversation I have with partners. Choose one task this week. That is the whole starting move.