I have a report that runs every Monday morning. Pulls last week’s numbers, formats them, drops them in a doc I can read over coffee. I used to tell people it was automated. It wasn’t. Somewhere between “pulls the numbers” and “drops them in a doc” there was me, sitting with a spreadsheet open, manually matching customer IDs from two systems that refused to agree on how to spell a customer’s name.
That’s the manual step in the middle. Almost every small operation has one. The ends of the pipeline look automated: data comes in on its own, the report goes out on its own. The middle is where a person quietly does the thing that makes the two ends line up.
What this step usually looks like
It’s rarely dramatic. It’s small, repetitive, and just annoying enough that you keep meaning to fix it and never do. A few common versions:
- Exporting a CSV from your payment processor and a CSV from your CRM, then pasting one into the other to figure out which cancellations were actual churn versus failed card retries that resolved a day later.
- Manually tagging which signups came from an ad versus organic search because your attribution setup only catches some of the traffic.
- Recalculating monthly recurring revenue by hand because a plan change halfway through the month doesn’t map cleanly to your billing tool’s built-in MRR number.
- Copy-pasting funnel numbers from three different dashboards into one sheet because none of the tools talk to each other and you don’t trust any single one to have the full picture.
None of these are hard, exactly. They’re just a person, doing a small judgment-heavy task, standing between two automated systems.
Why it’s there in the first place
The honest answer is usually data quality, not laziness. Systems disagree. A refund shows up as negative revenue in your accounting tool but as an unresolved dispute in your payment processor. A customer who downgrades their plan looks like partial churn to one system and a normal transaction to another. Someone has to look at the edge case and decide what it actually means for the business.
That’s a real function, not a chore you failed to eliminate. The problem isn’t that a human is involved. The problem is when that human involvement is invisible, undocumented, and living entirely in your head, which means the report only exists as long as you personally sit down and do it.
What it actually costs you
Set aside the time for a second, that’s the obvious cost. The less obvious one is fragility. A report that depends on you manually reconciling two exports is a report that doesn’t run if you’re sick, traveling, or just busy that week. It’s also a report where the definition of “churn” or “MRR” can quietly drift, because the rule for handling an edge case lives in your judgment that week, not in a written rule anyone could check.
I’ve had weeks where I handled a refund one way in the spreadsheet and a similar refund a different way three weeks later, because I didn’t remember exactly how I’d classified it the first time. Nobody caught it because nobody but me ever looked at the raw numbers. That’s not a metric problem. That’s a process problem wearing a metric’s clothes.
Finding your own manual step
Before you try to automate anything, spend one reporting cycle just watching yourself. Literally write down every action between “pull the data” and “send the report.” Most people find two or three of these steps, not one. Some examples of what shows up on that list:
- Deciding whether a customer counts as active or churned when they’ve paused but not cancelled.
- Deduplicating leads that came in through two different forms.
- Manually excluding test accounts or your own team’s usage from conversion numbers.
- Recomputing a cohort’s retention because your analytics tool buckets by signup week and you actually care about signup month.
Write the list down before you decide what to automate. If you skip this step, you’ll end up automating the easy 80% and leaving the annoying 20% exactly where it was, which is often the part costing you the most time.
Deciding what’s worth automating
Not every manual step is worth fixing. Some judgment calls genuinely need a human, and trying to force them into a rule just moves the error somewhere less visible. A rough way to sort your list:
Automate it if the rule is consistent. If you always classify a paused subscription the same way, that’s not judgment, that’s a rule you haven’t written down yet. Write the rule, then automate it, whether that’s a spreadsheet formula, a script, or a filter inside your existing tools.
Leave it manual if the rule keeps changing. If how you handle a refund genuinely depends on context, an angry customer, a billing error, a legitimate return, forcing that into an automated bucket just hides the judgment instead of removing the need for it. Keep it manual, but write down what you decided and why, so it’s consistent even without automation.
Flag it, don’t fix it, if you’re not sure yet. Some steps need a few more cycles of doing them by hand before the pattern is clear enough to automate correctly. That’s fine. Automating a rule you don’t understand yet just bakes an early guess into your numbers permanently.
Tools people reach for, and their limits
For a one-person operation, the usual toolkit is some mix of spreadsheet formulas, connector tools like Zapier or Make that move data between apps without code, and small Python or SQL scripts for anything a connector can’t handle cleanly. Which one fits depends entirely on the shape of your manual step, how many systems it touches, and how comfortable you are maintaining a script versus a no-code workflow.
I won’t tell you which one is “best,” because I haven’t run a controlled comparison and I’m not going to pretend I have. What I can tell you from running my own reports is that the tool matters a lot less than the rule. A well-defined rule for handling a paused subscription is easy to automate in almost any tool. A vague one will produce inconsistent results no matter what you build it in.
What automation genuinely can’t do here
Automating the mechanical part of a step, the copying, the matching, the recalculating, doesn’t remove the need for judgment on the parts that actually require it. A script can flag every refund over a certain size for you to review. It can’t decide, on its own, whether that refund reflects a real product problem or a one-off. That decision is still yours, and it should stay yours. The goal isn’t to remove yourself from the report. It’s to remove yourself from the parts where you’re not actually adding anything, so you have time left for the parts where you are.
It’s also worth saying plainly: automating a manual report process doesn’t guarantee it becomes more accurate. If the underlying rule was wrong, automating it just makes the wrong answer show up faster and more consistently. Fix the rule first. Automate second.
Where to start this week
Pick the one manual step that annoys you most, not the one that seems technically interesting to automate. Write down the rule you actually follow when you do it by hand, including the exceptions. If the rule holds up across a few past examples, that’s your candidate for automation. If it doesn’t, that’s useful information too, it tells you the definition needs work before the process does.
If you want more on how I track the numbers behind these reports, and where the manual steps still sit in my own setup, you can find the rest of what I write at Data Research Analysis Collection.
Get new guides and videos first — join the Telegram channel.