Why Zapier can't save you (and when it actually can).

Before we go further: Zapier is good. Make is good. n8n is good. We use all three inside our own systems. This isn't a takedown — it's a map of where they stop.

Drag-and-drop automation handles two things excellently: stateless triggers, and structured handoffs between two tools. If your problem looks like "when X happens in tool A, do Y in tool B," buy Zapier and stop reading.

Where it stops working

Four places, in roughly the order our clients hit them.

• 01Exceptions that need judgement — "if the invoice is over 10K, route to Sarah, unless it's a known supplier, unless it's end of quarter."
• 02Unstructured inputs — emails, PDFs, free-text customer notes, voicemail transcripts.
• 03Multi-system reconciliation — three tools that all think they're the source of truth.
• 04Long-running processes with rollback — anything that needs to survive a failed step three days in.

Each of these is solvable in Zapier. None of them is solvable cleanly. You end up with a Zap with twenty-eight steps, three filter branches, two delay paths, and a manual review queue that the operator forgets to check on Thursdays.

“The painful 40% of automation is exception handling. Zapier can solve it. The maintenance bill is what kills you.”

The maintenance asymmetry

Here's the trap. A complex Zap is cheap to build and expensive to maintain. A custom system is expensive to build and cheap to maintain. The crossover is roughly six months of operations.

If you're running a process for less than six months — campaign launch, holiday spike, one-off migration — the Zap is cheaper. Use it. If the process is core and runs forever, the maintenance compounds, and you've effectively built a custom system in the worst possible substrate.

A diagnostic in three questions

When a prospect asks us "should we just use Zapier?", we ask:

• 01Does any step require a judgement call that depends on context outside this process?
• 02Does the input arrive as text written by a human?
• 03Does the process need to survive a failure that happens hours after the trigger?

One yes — Zapier with effort. Two yes — Zapier with pain. Three yes — call us.