Last month, in a meeting room in Seongsu, Seoul, the CEO of a logistics startup pointed to the whiteboard and asked me: "We''ve locked down the metrics — execution count, success rate, hours saved. But which tool do we pick to actually produce those numbers?" On the whiteboard were twelve process lines, with four candidate tool names beside them: n8n, Make, Zapier, kintone.
Defining metrics and picking a tool look adjacent, but they''re different problems. Once you have the ROI framework we covered last episode in hand, the next wall is always the same: which tool fits your organization''s scale, regulatory environment, and staffing mix. Tool selection isn''t a problem you close with a single feature comparison sheet. You have to hold it up against the picture of your organization three years from now before an answer appears.
The four tools were born to solve different problems
People often line up these four in the same category and compare them feature by feature, but their origins differ. Skim the features without understanding where each tool came from and you''ll regret it in three months — reliably.
Zapier — the shortest on-ramp for non-developers
A US-based SaaS that launched in 2011. With more than 7,000 integrations, it holds the largest catalog in the industry. Its natural home is simple, two- or three-step trigger-action flows. Marketing and operations teams can ship their first automation in a two-hour afternoon without a developer. The catch: pricing is per-task, so as workflows grow more complex and execution counts climb, the cost curve gets steep fast.
Make — a visual middle ground for handling complexity
Formerly Integromat. Scenario elements like routers, iterators, and aggregators let you draw branching, parallel, and iterative logic visually. Features that force a higher Zapier tier land on Make''s baseline plans. Industry data suggests that for the same scenario, Make can come in around 60% cheaper than Zapier. It fits teams who aren''t developers but can understand logic when it''s drawn as a diagram.
n8n — an architecture for self-hosting and AI agents
The only self-hostable option among the four. Deploy it on an internal server and your data never leaves the perimeter. For organizations under healthcare, finance, public sector, or personal-data-residency requirements, it''s effectively the only choice. Since the January 2026 release of n8n 2.0, LangChain-native integration, roughly 70 AI nodes, agent memory, and connections to self-hosted LLMs have all become standard. If you''re seriously building AI agents or an in-house RAG stack, n8n''s architectural advantage is striking. The tradeoff: choose self-hosting and your organization carries firewalls, SSL, backups, and account management on its own shoulders.
kintone — not a connector, but a container for work
A no-code database and app builder from Cybozu. Where the previous three are "connectors between tools," kintone is "a container for the work itself." Approval workflows, procurement forms, customer records, field reports — you turn the organization''s actual work streams into apps and accumulate them in a central database. On the ground, SMBs across Korea and Japan have logged visible results. The tradeoff: no self-hosting, and connections to external SaaS are typically handled by running Zapier or Make alongside it.
Pricing, scalability, security, integrations — a comparison across four lenses
A single table won''t settle the question, but it does give the conversation a starting point. Below is a summary drawn from each service''s official documentation and pricing pages as of H1 2026.
| Lens | n8n | Make | Zapier | kintone |
|---|---|---|---|---|
| Pricing model | Per workflow execution; server cost only when self-hosted | Per operation; branching and parallel flows included | Per task; costs spike at high volumes | Monthly subscription per user |
| Extensibility | Custom nodes, REST, SDK — effectively unlimited | HTTP module covers most cases; custom work is limited | Small scripts via Code by Zapier | Plugins and JavaScript customization; Cybozu partner ecosystem |
| Security & data residency | Self-hosting keeps data on an internal server | EU/US regions; no self-hosting | US region only; hard for regulated industries | Cybozu Cloud; domestic data center via partner options |
| Integration count | 1,000+ official plus many community nodes | Around 1,500+ official | 7,000+ official — the industry leader | Focused on native apps, not connections; external links via Zapier or Make |
| Fit for AI agents | AI nodes and agent memory as native features | OpenAI module for simple calls | AI calls possible inside Zaps; weaker for autonomous-decision architecture | LLM integration is custom-development territory |
Read only the table and n8n looks like a universal answer, but the field isn''t that simple. Self-hosting takes at least a 0.3-FTE server engineer. Zapier''s per-task pricing sounds terrifying, yet for small teams running under 2,000 executions per month there''s a range where it''s actually cheaper than the alternatives. Extending the episode 3 argument — judgment should rest on execution count, success rate, and hours saved — you have to estimate your organization''s actual execution volume before the table comes alive.
Honestly, when I first started consulting I used to recommend n8n unconditionally. The architecture is free, and the cost is predictable. That thinking changed after I handed a self-hosted n8n to a five-person retail company with no dedicated admin, and they missed a single certificate renewal — automation was down for three days. Tools are a function of organizational capability.
Five typical workflows — which tool fits each
The picture gets much sharper when we walk through the five scenarios I encounter most often in SMB consulting.
First: order and CS automation for online stores. Connecting Cafe24, Naver Smart Store, and KakaoTalk consultations into one flow is a common request. If integration count is the decisive factor, Zapier wins; if the branching logic is complex, Make has the edge.
Second: approvals and procurement workflows. The real problem here isn''t connecting tools — it''s the internal approval chain itself. kintone fits naturally. A common hybrid is to layer Zapier on top for outbound notifications.
Third: internal RAG or an AI consultation bot. Feeding internal documents, quotes, and product manuals into an LLM so it can answer. The requirement combines data-leak concerns with agent architecture, which makes self-hosted n8n the most stable base.
Fourth: sensor data collection and alerting in manufacturing. In most cases, data coming out of factory PLCs or gateways can''t leave for external SaaS. Self-hostable n8n is close to the only viable candidate.
Fifth: customer data handling in regulated industries. Hospitals, accounting firms, law firms. Zapier and Make can hit hard blockers over server jurisdiction. What remains: self-hosted n8n, or kintone with a partner''s domestic data center option.
Self-hosting vs. SaaS — three axes for the decision-maker
From the seat of the person signing off, the view isn''t about features — it''s about risk. Three axes are enough.
The first axis is data residency. Can your organization''s data legally sit on external servers? Are there data-persistence clauses in your contracts? If not, SaaS is far easier. If regulatory requirements bind you, self-hosting is effectively mandatory.
The second axis is operational headcount. Choose self-hosting and server certificates, backups, access rights, and log management fall on the organization. Can you allocate even 20% of one developer''s time? If not, SaaS is the answer. If you haven''t already priced in the management cost beneath that $80-a-month server bill, the real total cost can easily exceed SaaS.
The third axis is the three-year picture. Will automation stop at ten workflows, or grow to a hundred? Once workflows cross into double digits, SaaS per-task pricing gets heavy fast, and self-hosting''s fixed-cost appeal grows. Conversely, if you''ll always sit at 5 to 10 workflows, SaaS''s operational convenience wins.
Closing — a tool is a hypothesis, not a decision
Episode 4 confirmed three things, briefly stated. The four tools were born to solve different problems. Judgment doesn''t come from one comparison table — it needs your organization''s execution volume, regulatory posture, and staffing. And the self-hosting question is a matter of risk and the three-year picture, not features.
At 5years+, we''ve walked through this kind of tool selection alongside 60 to 70 SMBs across Korea and Japan over the past three years. The lesson I keep coming back to: what mattered wasn''t which tool they picked at the start — it was how honestly they conducted the first review three months in. Tool selection is a hypothesis, not a decision. When the organizational context shifts, the tool shifts with it.
Once the tool is chosen, the next wall is budget. In the next episode, episode 5, I''ll walk through pricing by organization size and unpack the total cost of ownership for the first three months and the first year in actual numbers. If you finished this episode without a clear sense of where your organization fits, I''d recommend starting with a tool-selection consultation — one hour of looking at your situation together.