Every small business owner I talk to eventually asks the same question, worded two different ways.
The hopeful version: “How much time will this actually save me?” The suspicious version: “Is this going to be worth it?”
Both deserve an honest answer. And honest answers are hard to find online, because almost every automation ROI article is either selling something or pulling round numbers out of an industry average that has no relationship to your business.
So this post is the opposite. Two real projects I built this year, with the real hours saved, the real cost to build, and the real running cost. No composite clients, no rounded percentages, no “average business owner”. Just numbers from two small businesses who have the invoices to prove it.
Before we get into them, a bit of context on why this matters.
Those numbers come from Cin7’s 2026 multi-channel complexity report. They are the industry average. They are also the number nobody should base their own decision on, because your business is not the average. The only number that matters is your number.
Here are two of them.
A 20-person factory office that ran HR out of a Teams channel
The client is H.A Fashion, a women’s woven-wear manufacturer near Hanoi that produces for export markets across Europe, the US, and Australia. Their main factory runs 50 to 70 staff. The office in Hanoi is 20 people: merchandising, QA, accounting, and HR.
The office lives inside Microsoft Teams. Leave requests happened in one shared channel. Someone types “Hi boss, taking tomorrow afternoon off”, the manager replies “OK”, and that’s the record. It worked. Until HR tried to build a payroll report from it.
The before state (measured, not guessed)
We counted. Over 12 months, the channel had 416 messages. About 95% of them were leave-related, roughly 200 leave events a year across 20 people. Nobody was filing anything. Nobody was keeping a structured record. HR reconciled the whole thing manually every month.
Here is the monthly HR breakdown we wrote down at the start of the project:
| Monthly HR task | Hours |
|---|---|
| Scroll the channel to reconcile who took leave | 1 to 2 |
| Build the monthly leave summary report | 1 to 2 |
| Update timesheets and salary calculations | 3 to 4 |
| Total per month | 5 to 8 hours |
Call that roughly 6.5 hours a month on average. 78 hours a year on leave admin alone.
That is only one side of the cost. On the other side were the employees. The ones who tried to follow process wrote a formal confirmation email to HR after the Teams chat, 10 to 15 minutes per request. At 200 leave events a year, that was another ~50 hours. Most employees skipped it, which is how HR ended up reconciling the channel manually in the first place.
Total visible cost before automation: around 128 hours a year (78 HR + 50 employee paperwork), plus some number of missed or misfiled requests we could not measure.
The after state
The full project details live in the case study. The short version: an AI-powered automation reads the Teams channel every hour, recognizes when a conversation is a leave request with an approval, extracts the who-what-when-who-approved, and files it two ways. An email to HR, and a row in a live spreadsheet.
Nothing changed for employees or managers. They still chat. The automation runs underneath.
The math
About 100 hours a year back in the team. Split roughly 50 HR + 50 employees. The remaining ~28 hours are the irreducible timesheet and salary work, which still has to happen but now starts from clean, structured data rather than scrollback.
Build cost was a small fixed-price engagement. Runtime is $0.02 a day in AI inference. Everything else (n8n, Microsoft 365, Google Sheets) the client was already paying for.
Payback window depends entirely on how you value an hour of your team’s time. At $15/hour fully loaded (conservative for an office role in Vietnam), 100 hours is $1,500 a year. At the $25/hour typical for Western SMEs, it is $2,500 a year. Either way, the running cost of $7.30 a year is rounding error.
What this project is not: it is not a 10x productivity revolution. It is a quiet, boring, reliable 100 hours a year reclaimed from the same people doing the same work. Most real automation ROI looks like this. Slow compounding, not fireworks.
A multi-channel e-commerce team where the spreadsheet was lying
The client is Hy An, a small retail business selling across Facebook, TikTok Shop, and Shopee. Under ten people. The owner tracked revenue in an Excel file with one panel per bank account per month, every inflow summed into a single revenue column.
That setup works fine for a while. Then the business grows, internal transfers between company accounts start getting counted as revenue along with real sales, and suddenly the P&L is inflating because the spreadsheet cannot tell the difference between a customer paying for goods and the owner moving cash between his own accounts.
The before state
Three separate drains, measured at the start:
| Monthly task | Hours |
|---|---|
| Export Pancake orders and paste into the spreadsheet | 2 to 3 |
| Rebuild the monthly P&L by hand | 3 |
| Chase down miscounted revenue after the fact | 1 to 2 |
| Total per month | 6 to 8 hours |
Call it 7 hours a month. 84 hours a year of repetitive owner time.
But there was a second, bigger cost that doesn’t show up on the time sheet. The phantom revenue in the reports was distorting decisions. Marketing spend. Inventory bets. Cash planning. Over a year, this is the kind of thing that can easily cost more than the 84 hours themselves. Hard to measure, but worth naming out loud.
The after state
Every transaction in the new Google Sheet carries a tag: customer, internal, owner capital, or refund. The monthly P&L only counts customer rows. Internal transfers still exist, they are just structurally invisible to the report. A daily automation pulls the full order list from Pancake and writes it to the sheet. The monthly P&L rebuilds from a dropdown.
The math
84 hours a year of owner time is the number I’ll anchor to, because anything else is speculation. At a conservative $25/hour, that is $2,100 a year. The running cost of $60 a year is, again, rounding error.
The payback window on Hy An was faster than the math suggests because the phantom revenue was actively distorting decisions, not just costing time. Once the sheet started telling the truth, the cost of decisions-made-on-bad-data went to zero. I cannot put a dollar figure on that for you, and I am suspicious of anyone who claims they can.
What these numbers actually mean
Both projects fall in roughly the same ROI shape. About 80 to 100 hours a year reclaimed, project cost in the $2,000 to $5,000 range, running cost under $100 a year, payback inside 12 months.
If you have seen automation articles claiming 10x or 20x productivity gains, here is what the real math usually looks like:
| Claim you often see | What it actually means |
|---|---|
| "Saves 40 hours a week" | 40 hours saved across an entire 50-person organization, or all admin time replaced at a 1-person shop |
| "80% reduction in admin work" | The original 100% was five hours a week, so it is four hours a week saved, not forty |
| "Pays for itself in two weeks" | Assumes a $200/hour billing rate and ignores the setup cost |
| "Increases productivity by 10x" | Measured on one specific sub-task the automation was built for, not on total output |
None of this is a reason not to automate. Both of my clients above got real value, on real timelines, at real prices. It is just a reason to build your own math from the bottom up instead of trusting a vendor page.
The part most articles skip
If you want the honest version, here is what automation does not save you, ever.
The work itself. Somebody still has to do the work. Automation moves who does it from a person to a piece of software, but the work still happens. If you have a task that requires human judgment, automation does not make the judgment go away. It can prepare the options, but a person still picks.
Setup time. Every automation has a one-time cost: mine on a paid project, yours if you build it yourself. That cost is real. On a project that reclaims 100 hours a year, spending 60 hours on setup is a fine trade. Spending 300 hours on setup for the same 100 hours back per year is a losing trade for at least two years.
Maintenance. Automations drift. APIs change. Systems get upgraded. Accounts expire. Every workflow you build will, eventually, need to be touched again. Small n8n projects like the two above typically need 1 to 2 hours of maintenance a year. Bigger ones, more. Budget for it or the thing you built will quietly stop working and nobody will notice until it becomes an emergency.
Writing the process down. A lot of what looks like “automation time savings” is actually the time you saved by finally forcing yourself to write down how the process works. That’s real value, but you could have captured half of it with a checklist. Do not credit automation for what documentation was going to do anyway.
Rule of thumb I use with clients: if a process is in somebody's head, we write it down first and try running it by the checklist for two weeks. If the checklist alone makes 30% of the problem go away, we automate the rest. If the checklist changes nothing, the problem is genuinely mechanical and the automation case is strong.
How to estimate your own number in 10 minutes
If you want to sanity-check whether an automation is worth it for you, here is the simple version I walk clients through on the first call.
- Write down the task. Specific. “Importing Shopify orders into Google Sheets once a day” beats “accounting admin”.
- Measure the current cost in hours per week or month. Actually measure. Do not guess. Spend one week with a timer on your phone if you have to.
- Multiply by your fully-loaded rate. Not what you pay the person, what it costs you. Salary, benefits, equipment, overhead. For most SMEs that is 1.3x to 1.5x the base salary.
- Estimate the automation build cost. Get a fixed quote from whoever is going to build it. If you are building it yourself, estimate hours and multiply by the same rate.
- Estimate the automation running cost. Infrastructure, API, AI inference, subscription. Round up.
- Check the payback window. Annual time saved, times rate, minus running cost, divided by build cost. If the answer is under 12 months, most owners should pull the trigger. Over 24 months, think hard. Over 36 months, probably skip.
The one thing missing from this math is the decisions-you-make-differently value, like the phantom revenue case at Hy An. That one is real, it just is not quantifiable up front. Treat it as upside on a project that already penciled out on the hours alone.
Get your number
Both projects above took about a 30-minute first conversation to scope. Most of that was me asking for the time audit, because without real hours the math is useless.
If you want to run this on your own business, send me an email with a one-paragraph description of the process you are thinking about automating. I will ask for the current hours, estimate the build cost and the annual saving, and tell you honestly whether it is worth doing. If it is not, I will say so.
The first conversation is always free. You leave with real numbers either way.