Starting from a messy spreadsheet
The first time I tried to automate my back office work, it started with one ugly spreadsheet. I had tracking columns for invoices, checkboxes for whether they were paid, and little notes I told myself I would “clean up later” but never did. Every Monday I was manually emailing clients, updating payment status, and copying numbers into my bookkeeping app. It felt endless. When I realized I was spending more time maintaining the spreadsheet than doing the actual work, that is when I started looking at automation systems.
The first mistake I made was assuming Google Sheets plus Zapier on top would instantly solve everything. What actually happened was Zapier kept firing emails whenever I edited the wrong cell. I learned quickly that if you do not set up precise triggers, you end up spamming people. That was embarrassing the first week. Lesson learned: only trigger on a very specific event like “a row is added to this exact tab” instead of any edit across the file. I had to backtrack after my accountant got six copies of the same notification 😛
Connecting invoices without breaking sync
When I moved payments into an invoicing platform, I thought this would simplify everything. But the invoices never synced back properly. I had invoices showing up in two different states. Paid in one system, overdue in the other. No error message, just out of sync. What really helped here was adding a middle layer with Airtable. Airtable became the source of truth. The invoicing tool updated Airtable, and Airtable then informed my bookkeeping app. If something failed, it was much easier to see it in one place instead of hunting across three dashboards.
At one point Airtable stopped receiving updates from the invoice platform for a whole week. Why? Because I had a filter that excluded “archived” items, and apparently the platform marked everything as archived the day after issuing. The fix was just removing that filter, but it makes me laugh now that I lost days just to a hidden filter option ¯\\_(ツ)_/¯
Automating expense receipts the hard way
Scanning receipts through email should have been simple. Snap a photo with my phone, email it in, Zapier grabs the attachment, pushes it to cloud storage. Except Gmail started marking my own receipt emails as spam. No notification, nothing. Just vanished into the spam box. I only caught it when I noticed my cloud folder was suspiciously empty after a week.
I solved it by switching to forwarding rules instead of emailing myself directly. So the system now works: I forward to a special Gmail address, that address pushes the attachment to Drive, and the filename gets rewritten using the subject line. Once that part finally worked, I added an automation that pulled the total amount using OCR so I did not have to type it manually. That was surprisingly accurate most of the time. Where it failed most often: handwritten restaurant receipts. Those always looked like they were written during an earthquake.
Keeping track of employee requests
My small team constantly dropped requests in Slack, and those were getting lost in the daily chatter. So I created a workflow where a certain Slack reaction emoji pushed that message into a Trello board. It looked magical when it worked: React with a checkmark, and the message became a new card instantly. The problem was that sometimes Zapier skipped tasks randomly. No error, no explanation, just skipped. I eventually realized it was because the Slack message contained a hyperlink with unusual characters, and Zapier silently failed if it could not parse it.
The workaround was weird. I had to insert a Formatter step to strip out links before moving text to Trello. That solved the skip issue, but it also made some messages less useful because the links were gone. To deal with that, I added another field specifically for “attachments” and shoved the raw data in there. It is not elegant, but everything lands in Trello now and I do not miss important requests.
Building approval flows in email
I wasted a lot of time trying to get a manager approval flow to work in Slack. Notifications would often hit while someone was away from their desk, and then nobody remembered to scroll back. I tried adding reminders, but they stacked up in a really annoying way. Eventually I switched approval flows to email instead, even though that felt old fashioned.
The automation sends a structured email with two links: approve or deny. Clicking one link updates Airtable. The trickiest part here was handling repeated clicks. People would sometimes double tap just because they were unsure if it registered. Without protections, this created multiple approval records. I fixed it by using an automation step that checked whether the record already had a status before allowing an update. Now if somebody clicks multiple times, only the first one counts. That small conditional saved me from a lot of headaches.
Why error notifications matter
For months I thought everything was working fine until I realized that one small Zap had been off the entire time. No red warning, no message, nothing. It was a webhook that silently failed because the receiving app had changed its authentication rules. I had hundreds of tasks that never went through. That was the moment I decided every single automation must have an error notification, whether that means sending me an email or dropping an alert in Slack.
The uncomfortable part is that error messages often do not say much. “Task halted” with no explanation is common. To avoid chasing ghosts, I built in test runs once a week. Even though that sounds manual, it is still easier than being blindsided months later. Testing feels boring when nothing breaks, but the time it saves me when something does fail is huge. Like I once had an invoice generator stop working only for international clients, and without a test system, I never would have discovered it until one of them complained 🙂
Documenting for my future self
I keep a private doc where I note tiny gotchas like “do not trigger this automation after 10pm because the timezone conversion makes it fire tomorrow.” It is not pretty, but it keeps me sane. What I learned is that the system only feels automatic if you have already solved the limits and quirks. Otherwise, you are babysitting your tools constantly. Every time something breaks, I add a rough sketch of what went wrong, what I tried, and what finally worked. Half the time future me thanks past me for remembering the weird edge cases.
At some point I realized the doc is basically a survival manual for my own back office automations. Nobody else would be able to follow it, but it is what keeps me afloat when I open too many browser tabs and forget what I was even fixing.