Zapier earned its reputation. So did Make and n8n. For small businesses with a handful of straightforward integrations, they're a bargain, automation that would take an engineer two weeks done with a few clicks instead.
The trap is that they keep working long after they've stopped being the right answer. The Zapier stack that was $40/month two years ago is $850/month now, fails silently every fortnight, and nobody fully understands it. At that point, custom code is the cheaper option, by a lot.
Five warning signs you've crossed the line
- Your Zapier bill is over $400/month. The maths starts to favour custom code at around this point, you're paying $5k a year for something that runs in a Lambda function for $20.
- You have at least one Zap that's failed twice this month. Silent failures in revenue-touching automation are eating money you can't see.
- Nobody on your team can confidently explain how the most important Zap works. Knowledge has drifted out the door. The Zap is now load-bearing and unmaintained.
- You've added "manual checks" to compensate for unreliable automation. The cost of those checks is hidden but real.
- You're stacking Zaps inside Zaps because logic isn't expressible directly. The complexity has outgrown the tool.
Two of these = pay attention. Three or more = it's time to graduate.
Where Zapier still wins
Don't read this as "Zapier is bad." It's not. Zapier is brilliant for:
- Connecting two SaaS tools with simple field-mapping.
- Automations that fire fewer than 1,000 times a month.
- Workflows that change frequently (it's faster to edit a Zap than push code).
- Companies without an engineering team or technical operations person.
The mistake is using Zapier for high-volume, business-critical, complex flows, not Zapier itself.
What custom code actually replaces
The pattern that works is a small middleware service, usually 200–800 lines of Node, Python, or Go, that:
- Receives webhooks from your tools (HubSpot, Stripe, Xero, etc.).
- Applies your business logic in real code.
- Calls APIs to update other tools.
- Logs every step, retries on failure, alerts on persistent errors.
It runs for $20–$80/month on Render or Vercel or AWS Lambda. It doesn't fail silently. It can be tested. It's documented in code, not in a UI.
Zapier is a great way to learn what your automations should do. Custom code is the cheaper way to actually run them.
The cost comparison
For a typical mid-stage Australian business with 10–20 integrations:
- Zapier: $400–$1,500/month. Plus 1–2 hours/week of engineer or ops time fixing failures. Annual: $7k–$22k including labour.
- Custom middleware: $9k–$16k build. $30–$80/month running. 0–3 hours/month maintenance. Annual: $1k–$2k after the initial build.
The break-even is usually 9–14 months. After that, custom is cheaper forever.
The hybrid that often wins
You don't need to rip out Zapier entirely. The pattern I see most often:
- Keep Zapier for simple, low-volume, infrequently-changing integrations.
- Move the high-volume, business-critical, complex ones to custom code.
- Use n8n (open-source, self-hosted) for the middle ground, programmable but visual.
Most healthy automation stacks I see at $1m–$10m revenue businesses use all three.
What to do this week
- List every Zap and Make scenario you have. Almost nobody has this list; you'll be surprised.
- Score each on volume × business impact × failure tolerance. Anything high on all three is a candidate for code.
- Cancel Zaps that have run zero times in three months. Likely 20–40% of your stack.
- Document the top five. If you got hit by a bus tomorrow, could a colleague figure them out?
- Move the worst-performing one to code. Often pays for itself in months.
The bottom line
Zapier is brilliant until you've outgrown it. The signs are clear: rising bills, silent failures, lost knowledge, manual workarounds. When two or more apply, custom code is usually cheaper, more reliable, and easier to maintain.
If you want three Australian developers to audit your stack and quote a custom replacement, FindDevs gets you those quotes. Free, three real numbers, in 24 hours.