"So, How Much Did We Actually Make?"
Last month I spent about two hours with the CEO of a mid-sized manufacturer. His finance manager, he told me, had frozen for a full thirty seconds when the CEO threw that exact question at him in a briefing. The automation project had been humming along for three months, the operators kept saying "things got easier," but no one had prepared an answer in numbers.
Honestly, this isn't just his company's story. Businesses that follow the concepts, categories, and roadmap I laid out in Episode 1 of this series and make it through adoption in one piece tend to get stuck at exactly this point. Today's piece is the next chapter — how to prove, after the fact, how much money automation actually made you.
The Real Reason Measurement Fails — There's No Baseline
Why can't companies calculate ROI? Not because the formula is hard. It's because no one wrote down what the world looked like before automation.
There are four baselines you absolutely have to capture before a PoC begins. First, the average time to handle a single instance of the task. Second, monthly volume. Third, the error or rework rate. Fourth, the associated labor cost — the operator's hourly wage multiplied by time spent. Miss these four and no dashboard, however pretty, will answer the question "how much better is this than before?"
If you haven't yet read the previous episode on the differences between RPA, workflow automation, and AI agents, I'd start there. The tool you picked subtly changes what your baseline items should look like. AI agents in particular force you to define the very line between success and failure by hand.
The Four Core Metrics — At a Minimum, Watch These
The industry has dozens of metrics on offer, but if you want to keep the approval chain awake, four is plenty. Any more and you end up with metrics for the sake of metrics.
| Metric | Definition | Calculation |
|---|---|---|
| Execution Count | Monthly volume processed by the automation | Log count (aggregated daily) |
| Success Rate | Share of runs that completed normally | Successful runs ÷ Total runs × 100 |
| Time Saved | Processing time saved per task | (Baseline time − Post-automation time) × Execution count |
| Cost | Savings vs. operating cost | Time saved × Hourly wage − (Operations + license fees) |
Stack these four every week in the same format and you won't get caught flat-footed at a CFO briefing three months later. A 30% success rate may sound small, but on a task that runs a thousand times a month, that's three hundred cases no human has to touch.
Calculating ROI — Simple Formula, Devil in the Details
The basic formula is straightforward.
ROI (%) = (Annual net gain − Initial investment) ÷ Initial investment × 100
There are industry benchmarks worth citing. Aggregating RPA case studies published by Automation Anywhere, SS&C Blue Prism, and others, well-designed projects tend to land at 100–250% ROI over the first 12–18 months, with the top tier clearing 380%, and payback periods of six to nine months. Tasks that are high-frequency and narrow in scope sometimes recover the principal in three to four months.
That said, there are two traps I'd flag for Korean and Japanese SME executives.
The first is the company-wide integration approach. "Automation lifted our total revenue by X" is a claim that's nearly impossible to prove causally. Invoice processing in finance, lead registration in sales, response handling in CS — calculating ROI per task and summing at the end is far easier to push through an approval chain.
The second is mixing qualitative effects into the base ROI. "Employee satisfaction went up" or "decisions got faster" are real gains, but the moment you try to force them into a number and stuff them into the ROI figure, the credibility of the entire report starts to wobble. Keep base ROI to "direct time saved × hourly wage + reduction in rework cost from errors," and describe the softer effects qualitatively in a separate section. It's safer.
The Weekly Report — What Keeps Metrics Alive
However good your dashboard is, if no one looks at it, the numbers are dead. What we recommend to clients is a single-page weekly report. Every Friday afternoon, the automation lead emails it to the CEO, CFO, and business unit heads.
It carries about six items. Execution count this week and the change vs. last week. Success rate and the top three failure reasons. Cumulative time saved, converted into currency. Any newly discovered exception cases. Improvements queued for next week. And finally — the single sentence the CFO actually wants. That last line is, honestly, the whole thing.
"Time saved by automation this week converts to roughly ₩4.7M in labor cost." Stack a line like that every week for six months and the case for expanding the automation budget writes itself.
In 2026, One More Metric Is Becoming Necessary
As AI agents built on top of LLMs move seriously into business automation, two new metrics have joined the classic four. One is Decision Accuracy. The other is Exception Handling Capability. In rule-based RPA, "did it run or not" was the whole game. Agents that exercise judgment demand a separate question: "did it judge correctly?"
How you measure that depends heavily on the character of the tool you picked. So in the next episode I'll compare actual tools — n8n, Make, Zapier, and kintone — from an SME lens. Which one clears the approval chain in which situation, and how to calculate license cost accurately. See you then.
Wrapping Up
Measuring business automation ROI isn't rocket science. Capture four baselines before the PoC. Stack four core metrics weekly in the same format after adoption. Calculate ROI per task and sum at the end, not top-down. Hold to these three rules and you won't get stuck for words in front of the board.
At 5years+, we've walked Korean and Japanese SMEs through this end to end — automation PoC design, baseline measurement, weekly report templates. We share our ROI calculation Excel template (baseline sheet + four-metric formulas built in) with anyone who asks. Send a quick note here and someone will get back within a day. Feel free to start with something as light as "which metrics would even fit our company?" — that's a fine place to begin.