I build automations as code, so read this with that in mind. I also tell people on scoping calls when what they need is three steps in Zapier, because building it custom would waste both our money. The choice comes down to three questions: what the workflow touches, what happens when it breaks, and what you own at the end.
When n8n or Zapier is the right answer
- Every system involved already has a connector in the platform.
- The workflow is a handful of steps with little branching.
- If it fails for a day, someone notices and nothing important is lost.
- The person maintaining it is comfortable in the platform's editor.
Our own cost calculator says it plainly: if your answers put you under $6,000, the honest advice is usually that you don't need a custom build yet.
When custom code is worth it
- No connector exists. The data lives on websites, portals or legacy systems that a platform cannot reach without scraping.
- The rules are the product. Deduplication, enrichment, qualification and required fields decide whether the output is useful.
- Silent failure is expensive. An empty spreadsheet delivered on time is worse than an alert.
- You need to own it. You want the system in your repository, on your infrastructure, maintainable by any engineer you hire.
A worked example: dozens of county sites, checked by nobody
A real-estate investment client was checking dozens of US county court and sheriff sites by hand every morning. None of those sites has a connector in any automation platform. The system that replaced the manual work is a daily Playwright pipeline with proxy rotation that harvests new filings, dedupes them, enriches them through skip-trace APIs and pushes only qualified leads into GoHighLevel. It sends a daily summary and a failure alert, so a broken site is noticed the same morning. It runs fully unattended.
That workflow ends in a CRM, which a platform can talk to. Everything before that point is scraping, matching and judgment rules, which is where code is the better tool. More on this kind of build on the web scraping and data pipeline services page.
What happens when it breaks
Sources change. A form gets a new field, a site changes its layout, an API starts returning something different. In a GUI platform, there is nothing to debug except nodes to inspect and re-drag. In code, there is a stack trace and a line number, and a test that can be added so the same break is caught next time.
A reliable automation checks more than whether a job finished. Required fields, duplicate rules and delivery checks get defined before the build, and if a source stops returning usable records the workflow flags it instead of quietly delivering nothing.
What you own in each case
| No-code platform | Custom code | |
|---|---|---|
| Where the logic lives | Inside the platform's editor | A repository you own |
| Licence underneath | Yes, the platform's | None beyond your hosting and APIs |
| Strongest ownership claim | Exporting the workflow | Source code, README and runbook |
| Who can maintain it | Someone who knows that platform | Any engineer, because it is just code |
| When it fails | Nodes to inspect | A stack trace and a line number |
n8n can run on your own server, which helps with data residency, but the workflow still lives inside n8n's editor and still needs someone who knows it. For a full list of what to collect at the end of any build, use the AI software handover checklist.
Why automation projects fail after go-live
The most common complaint from people who have bought automation work is not that it never worked. It is that when it stopped working, nobody was still responsible. Whichever tool you pick, fix that in writing before work starts: a named warranty period, an acceptance window, and a written update every week of the build. Fixed price vs hourly AI development covers how those terms fit into a contract.
What a custom automation costs
The Data & Automation Engine starts at $6,000 and typically takes three to six weeks, covering collection, cleaning, deduplication, enrichment, scheduling with summaries and failure alerts, and delivery into your CRM, database or dashboard. If your workflow is document-heavy, invoice automation: OCR is only the first step shows how to scope the review and export steps.
Frequently asked questions
Is n8n better than custom code for automation?
For a few steps between tools that already have connectors, a platform like n8n, Zapier or Make is usually the right answer. Custom code is worth it when sources have no connector, the dedup and enrichment rules matter, silent failure is expensive, or you want to own the system outright.
Can I own an n8n or Zapier workflow?
You can usually export the workflow, but it still runs on the platform and under its licence. With custom code you own the repository, documentation and runbook, and any engineer can maintain it.
How much does a custom automation cost?
At Kortex Labs the Data & Automation Engine starts at $6,000 and typically takes three to six weeks. If a cost estimate comes in under $6,000, the honest advice is usually that a custom build is not needed yet.
What happens when a website or source changes?
A well-built pipeline defines required fields, duplicate rules and delivery checks up front, so when a source stops returning usable records it raises an alert the same day instead of delivering an empty result.