Every week someone tells me they automated a process and saved a fortune. When I ask how they measured it, the answer is almost always a feeling, not a number. Measuring automation roi honestly is the step most owners skip, and it is exactly the step that separates a real win from an expensive toy.
The uncomfortable truth is that most automations look worse once you count them properly. That is not a reason to avoid the math. It is the entire reason to do it. If you only automate the things that survive an honest calculation, you will spend far less and keep far more.
Here is the actual formula I use with clients, filled in with a plausible example, so you can run your own numbers before you build anything.
The Formula Nobody Writes Down
Automation ROI is not "it feels faster." It is a subtraction:
Annual value = (hours saved per year x loaded hourly cost) + (error cost avoided per year)
Annual cost = build cost (amortized) + monthly maintenance x 12
ROI = Annual value minus Annual cost
Three of those five numbers are the ones owners forget. Loaded hourly cost is not just salary, it includes BPJS, THR, and overhead, usually 1.3 to 1.5 times the base wage. Error cost avoided is the money you lose today from mistakes the automation prevents. And maintenance is the recurring tax that quietly eats the savings, because software breaks when an API changes, a form moves, or a vendor updates something.
Skip any of those three and your ROI is fiction.
A Filled Example: Automating Invoice Data Entry
Take a small trading company where one admin manually keys supplier invoices into a spreadsheet. The owner wants to automate the extraction and entry.
Hours saved. The admin spends about 2 hours a day on this, so roughly 40 hours a month, or 480 hours a year.
Loaded hourly cost. The admin earns Rp 5,000,000 a month. Loaded at 1.4x, that is Rp 7,000,000, or about Rp 40,000 per hour across a 175-hour work month.
- Hours saved value: 480 x Rp 40,000 = Rp 19,200,000 per year
Error cost avoided. Manual entry produced maybe two costly mistakes a year, a wrong payment and a missed discount, averaging Rp 3,000,000 each.
- Error cost avoided: Rp 6,000,000 per year
Annual value = Rp 25,200,000
Now the costs:
Build cost. A custom extraction workflow costs Rp 30,000,000 to build. Amortized over 3 years, that is Rp 10,000,000 per year.
Maintenance. Vendor tools, monitoring, and the occasional fix run Rp 1,500,000 a month, so Rp 18,000,000 per year.
Annual cost = Rp 10,000,000 + Rp 18,000,000 = Rp 28,000,000
ROI = Rp 25,200,000 minus Rp 28,000,000 = negative Rp 2,800,000
Read that again. The automation everyone would have called a success actually loses money in year one, driven entirely by maintenance. That is the number the feeling never shows you.
What the Honest Number Reveals
The example is not an argument against automation. It is an argument for counting. A few things fall out immediately once you see the math:
- Maintenance dominates. In this case it was larger than the build cost and larger than the error savings combined. Any automation with fragile integrations carries a heavy recurring tax, and that tax is where "cheap to build" projects go to die. I have seen this pattern in legacy data migration work too, where the ongoing upkeep dwarfs the one-time cost.
- The admin does not disappear. Saving 40 hours a month only pays off if that person redirects those hours to revenue-generating work. If they just have a lighter day, the "savings" never hit the P&L, and your value column is inflated.
- Cheaper approaches change the answer. If a low-maintenance tool cut monthly upkeep to Rp 400,000, the ROI flips strongly positive. The build-versus-buy decision, not the automation itself, was the real lever.
This is also why I am skeptical of a lot of automation marketing. The demos count the hours saved and stay silent on the maintenance. For a broader take on which promises to ignore, see AI hype vs reality for small businesses.
A Simple Worksheet You Can Run Today
Before you approve any automation, fill in six lines:
- Hours saved per year (be conservative, measure a real week)
- Loaded hourly cost (base wage x 1.4)
- Error cost avoided per year (real incidents, not hypotheticals)
- Build cost, divided by the years you will realistically use it
- Monthly maintenance x 12 (ask the vendor to commit to a number)
- Subtract lines 4 and 5 from the sum of lines 1x2 and 3
If the result is negative or thin, either find a lower-maintenance approach or leave the process manual. There is no shame in a manual process that costs less than its automation.
The Takeaway
The reason to count automation ROI properly is not to kill automation. It is to kill the automations that only look good in a slide. Hours saved is the seductive number, maintenance is the honest one, and the gap between them is where budgets quietly leak.
Run the six-line worksheet on your next idea before you spend a rupiah building it. Most will not survive, and the few that do will be genuinely worth it. If you want a second set of eyes on the numbers before you commit, that kind of honest technical assessment is exactly what I bring to a partnership.