In Episode 4, we compared four tools — n8n, Make, Zapier, and kintone — which likely gave you a rough picture of which tool fits which company size. But once you're standing at the CEO's office door with the actual approval papers in hand, the first question they throw at you is almost always the same. "So how much is this whole thing going to cost me?"
That's exactly what happened last month when I sat down with an operations manager at a logistics company in Busan. He had brought a thick binder comparing tools, but the budget spreadsheet column was empty. "I get that n8n is cheaper. But where do I put the self-hosting server fees, how do I account for developer labor, and how many months should I estimate for our ERP integration…" His predicament is where this article starts.
The Real Cost of an Automation Project — Think in Three Layers
The reason it's hard to answer "How much does automation cost?" in one line is simple: the cost stacks in three layers.
The first layer is the initial cost. This includes software licenses, development and build labor, data preparation, and staff training. The second layer is operating cost — monthly subscriptions, LLM API usage fees, and the time of whoever handles ongoing maintenance. The third layer, often left out, is maintenance cost. RPA in particular is fragile: whenever the UI of an integrated system changes, the bot breaks, so you have to budget separately for annual rework.
The three layers together are called TCO (Total Cost of Ownership). That's why you should never tell an approver, "It starts from KRW 200,000 per month." Multiply by twelve months, add development labor, layer on the maintenance capacity, and only then does the number become honest.
Estimate Ranges by Company Size — Where Does Yours Fit?
Here's a rough map of what estimates actually look like in the Korean market as of 2026. Treat these as benchmarks only — depending on task complexity and the number of connected systems, expect the numbers to swing by around 30% in either direction.
| Company Size | Scope | Initial Build | Monthly Ops |
|---|---|---|---|
| Small (10–50 people) | 1–2 simple macro/Excel automations | KRW 1M–5M | KRW 100K–500K |
| Small-to-mid (50–150 people) | Single-ERP integration, one workflow | KRW 5M–15M | KRW 300K–1M |
| Mid-market (150–300 people) | 3–10 RPA bots, multi-department rollout | KRW 20M–50M | KRW 800K–3M |
| AI Agent PoC | Partial adoption experiment | KRW 2M–8M | LLM API extra, from KRW 300K |
If the table makes you wonder, "Our estimate is 5M won, but somewhere else quoted 30M — why?" — that's the correct reaction. The gap almost always comes down to how many existing systems need to be integrated and how clean your data already is. The moment you're bridging ERP, groupware, and an in-house custom management system, the estimate gains another digit.
Monthly Subscription vs. One-Time Build — When Does TCO Flip?
This is the fork most practitioners face first. Go SaaS with Zapier or Make, or go n8n self-hosted or fully custom-built?
Zapier bills per step. If your workflow has 5 steps and runs 30 times a day, you're at 4,500 tasks a month. Add a couple more workflows and you're pushed straight into a higher-tier paid plan. n8n Cloud, on the other hand, counts by workflow execution. Whether the workflow has 5 steps or 15, a single run counts as one.
The breakeven point observed in practice tends to look like this: once a Zapier subscription crosses the $200 (roughly KRW 280,000) monthly threshold, n8n self-hosting or partial custom development takes the lead on TCO. Even amortizing an 8M–15M won upfront development spend across the next 24 months still lands you below the cumulative SaaS bill from that point on.
One trap, though. Going self-hosted means picking up a new line item: developer upkeep. If nobody in-house can operate it, you outsource — and that contract cost has to sit inside the TCO to keep the number honest. This is where Episode 4's tool comparison becomes a useful reference again. Tool selection ultimately comes down to where your internal resources cross the budget curve.
Budget Allocation Principles That Break Even in Year One
There's a field-tested benchmark that keeps showing up in consulting work: 30% license / 40% development and build / 20% operations and training / 10% contingency. Stray far from this ratio in your initial plan and something usually starts wobbling later.
The most common failure pattern is cutting the 10% contingency. It gets trimmed because the budget feels tight, but in reality, unexpected spend inevitably surfaces during the data preparation stage. "The Excel formats differ across every department." "Client codes are duplicated in three separate systems." — that's how two unplanned days of developer time end up on the invoice.
Start with subscription-only and you can flip to positive ROI in 4–6 months. Low upfront pressure, but the ceiling on scalability arrives sooner. One-time development takes 6–12 months to recoup, but marginal cost drops sharply after that, and the curve tips in your favor from year two. The answer most often chosen in the field is a hybrid: put stable, repetitive tasks on SaaS quickly, and reserve partial custom development for company-specific processes. For SMBs balancing quick payback with room to grow, it's currently the most practical combination.
Estimates That Approvers Trust — Three Common Traps
After reviewing dozens of estimates, three traps keep showing up.
First, misreading the licensing model. Whether you're billed by number of users, number of bots, or number of transactions can shift the one-year TCO by tens of millions of won. It's common to sign on the strength of an "unlimited users" line, only to discover later that per-execution fees are billed separately.
Second, omitting the maintenance line entirely. Inflate the upfront build cost, mark "no further expenses expected," and approvers actually get suspicious. Being upfront about a 15–20% annual maintenance line earns more trust.
Third, hiding data preparation labor inside the development cost. This should sit on its own line, both for the approver's clarity and for the practitioner's safety. It's a different kind of spend.
Execution Checklist
- Lock the list of systems to integrate up front — the count drives the estimate
- Base monthly execution volume on three months of actual measurement data
- If the SaaS subscription is trending toward KRW 200,000/month or higher, get parallel quotes for self-hosting and custom development
- Always reserve a 10% contingency — for data preparation risk
- State the 15–20% annual maintenance line explicitly in the estimate
- Cap the PoC at KRW 5M or under, and set the pilot-expansion decision at the three-month mark
At 5years+, we've worked with small and mid-sized companies in Korea and Japan on this kind of budget design and estimate review. Share the outline of your situation and we'll come back with a rough estimate within 24 hours. If it helps, feel free to reach us through the contact page.
In Episode 6, we'll cover the topics you can't avoid the moment you layer LLMs on top of automation — PII masking, audit logging, guardrails, and the rest of the security and governance work. Once the budget picture is set, the next gate is inevitably the question, "Is our personal data leaking out through this thing?"