Zapier Review

The connector graph is the whole moat

Roughly 7,000 integrations. Strip away the marketing and that single number is the entire argument for Zapier, and it holds because automation value is gated by the worst-supported tool in your chain, not the best. A platform that covers 1,000 apps deeply still fails the workflow that needs app number 1,001. Zapier almost never makes you fall back to a raw HTTP call, and for a non-technical operator that difference is the difference between shipping today and filing a ticket with engineering.

That is the position this review defends: Zapier wins on coverage and time-to-first-working-Zap, loses hard on cost the moment volume or branching enters the picture, and the crossover is arithmetic you can compute before you sign.

What it actually does well

Coverage is the headline, but the load-bearing detail underneath it is trigger quality. Most platforms list an integration; the question is whether it exposes a real-time webhook trigger or only a poll. Zapier's better-supported apps push instant triggers, so a Zap fires on the event instead of on a 1-to-15-minute polling cycle. For ticket routing or lead capture, that latency gap is the product.

The 2026 AI actions are the genuine new capability, not a checkbox. An AI action runs a GPT or Claude call as a native step with the model output typed into downstream fields, so classification, extraction, and drafting happen inside the Zap with no separate service to host or bill. Tables and Interfaces extend the same logic: a built-in store and a form builder mean a small internal tool stays on one invoice instead of pulling in Airtable plus a form vendor.

Where it breaks: the pricing mechanism, stated exactly

The Achilles heel is not "Zapier is expensive." It is the specific shape of how it gets expensive. A task is billed per successful action step, per record. The cost of a Zap is therefore (steps minus trigger) times records processed, not records processed. A 5-step Zap over 100 records is 500 tasks. The same 100 records through a 9-step Zap is 900 tasks. Your bill scales with the complexity of your logic multiplied by your volume, which is precisely the two axes that grow as a team matures.

Make bills the same module-run unit, so this is not a free lunch elsewhere. The decisive difference is the base price the multiplier sits on. Zapier Professional is 69.99 USD per month for 2,000 tasks. Make Pro is 16 USD per month for 10,000 operations. The per-unit rate and the included volume both move in Make's favor, so the multiplier that hurts on Zapier hurts roughly a quarter as much per dollar on Make. We have the internal number to anchor this: rebuilding our own Pondero automation stack (eight production scenarios) off Zapier Professional onto Make in March 2026 moved the same trigger volume and the same destinations from 69.95 USD per month to 18.82 USD per month plus a 2.82 USD operations top-up. Same work, roughly a third of the cost. The full migration math is in the Zapier vs Make comparison.

Error handling is the second structural gap and it comes from the same root cause: the builder is linear. Make models a failure as a separate error route you can branch into, retry, and aggregate. Zapier's linear path gives you per-step retry and an autoreplay, which is enough for a 3-step Zap and visibly thin for a workflow with conditional branches. Complex branching is technically possible through Paths, but each path you add is more boxes stacked in one column, and past two or three branches the design fights the canvas instead of the canvas serving the design.

When the recommendation flips

Use Zapier when the deciding factors are connector coverage and a builder a non-technical teammate can own without help, and your task math stays modest: low step counts, low-to-moderate record volume, mostly linear logic. That describes a large share of real ops work, and for it Zapier is the correct call and the fastest one.

It flips on two triggers, either one sufficient. First, volume: once monthly tasks (steps times records) clear the point where Professional's 2,000 included tasks turn into recurring overage, the per-dollar gap to Make compounds every month and the migration pays back inside a billing cycle or two. Second, shape: once a workflow needs more than two real branches or genuine failure routing, Make's router and error-route primitives stop being a nicety and start being the reason the thing is maintainable. If you are self-hosting-capable and volume is the whole story, n8n removes the per-run meter entirely and wins on cost outright; the cost is running and patching it yourself.

The candid read: Zapier earned default status and keeps it for the simple-and-broad case. Compute steps times records against your real volume before you renew. If that number is large or your logic branches, the renewal is the moment to reprice on Make, and on n8n if you can host.