Why your Zapier or Make bill keeps growing (and how to cut it)

Short answer: Automation bills grow because usage is counted per action, per item, not per workflow, so one busy trigger or a loop over many rows multiplies quickly. On Zapier, only successful action steps count as tasks, and filters, paths and several built-in utilities are free; on Make, every module run is an operation (now billed as credits), including each scheduled check for new data. Filter early, switch polling to instant triggers where possible, process in batches, and delete steps that do nothing useful.

Before changing tools, find out which workflow is eating your allowance. Both Zapier and Make show usage per workflow, and in most small businesses one or two automations account for most of it.

How each platform counts

Zapier: tasks

Zapier's help center (as of November 2026) says a task is a successful action. What counts:

each successful action step, once per run
successful steps in an error-handler path
steps that run again when you replay a whole Zap run
steps inside sub-Zaps, plus the call and return steps

What does not count:

trigger steps
Filter and Paths steps
testing in the Zap editor
actions that error or halt, and steps that never run because a filter stopped them
Formatter, Delay, Looping, Digest, Storage and a few other Zapier utility apps
Zapier Tables and Zapier Forms steps

Some add-ons are counted differently (Zapier lists its Lead Router and MCP tool calls at several tasks each), so check the page if you use them.

Make: operations and credits

Make's help center defines an operation as "a single module run to process data or check for new data". Most modules run once per bundle (one email, one row, one file), so sending five emails is five operations. Trigger modules are the exception: they run once per check, whatever they return. Make now bills in credits; for non-AI apps one operation equals one credit, while AI modules can also consume credits for tokens.

The consequence: a scenario scheduled every few minutes uses an operation on every check, even when nothing new arrived.

n8n: executions

n8n's cloud plans are priced on monthly workflow executions "regardless of complexity", so a 20-step workflow costs the same as a two-step one. Its free Community edition can be self-hosted, which changes the cost question from usage to server and maintenance time.

The cost drivers

Driver	What happens	Where it hurts most
Filtering late	Actions run before the step that decides the item didn't matter	Zapier and Make
Frequent polling	Each scheduled check counts, even with no new data	Make
Looping over many items	Each item through each action counts	Zapier and Make
One workflow per tiny job	Repeated lookups and updates across several workflows	Both
Replays and retries	Re-running a whole failed run re-counts successful steps	Zapier
Noisy triggers	"Updated record" fires on every small edit	Both
AI steps	Extra credits for tokens on some modules	Make (check AI pricing elsewhere too)

Fixes, roughly in order of payoff

Put the filter right after the trigger. On Zapier, a Filter step is free and stops everything after it. On Make, a filter between modules means later modules don't run for that bundle. Ask: which items should never get past step one?
Use instant triggers instead of polling. Make's webhook triggers fire when data arrives instead of checking on a timer. If the app offers an instant trigger (often labeled "INSTANT" in Make), use it.
Slow the schedule down. If a scheduled scenario only needs results by morning, run it once a day rather than every 15 minutes.
Pick a narrower trigger. "New paid order" instead of "order updated". "New row in this sheet" instead of "any change". Fewer trigger events means fewer runs downstream.
Batch instead of looping. Many apps accept bulk actions: add 50 rows in one call, send one summary email with all new leads. Zapier's Digest app collects items and releases them together, and Digest steps don't count as tasks.
Search before you create, once. Duplicate "find or create" steps across several workflows add up. Combine related workflows so the lookup happens once.
Replay only the failed step. Where the tool lets you, rerun from the failing step rather than replaying the entire run.
Delete dead steps. Old notification steps nobody reads, logging to a sheet no one opens, duplicated Slack messages.
Check AI steps. If an AI step classifies every email, try a cheap keyword filter first and send only the unclear ones to AI.

When self-hosting makes sense

Self-hosting n8n's Community edition removes per-run billing, but you take on hosting, updates, backups, security and fixing things when they break. It tends to make sense when:

you run high volumes of simple, stable workflows
someone on your team, or a contractor, can maintain a server
you need workflows to run on your own infrastructure

It usually doesn't when you have a handful of workflows, nobody technical in-house, or depend on niche app integrations that are mature on Zapier or Make but thin elsewhere.

A middle path: keep business-critical, low-volume workflows on a hosted platform and move the high-volume data shuffling to a script or a self-hosted tool.

FAQ

Do failed Zaps use up tasks?

According to Zapier's help page, action steps that error or halt don't count. Steps that succeeded before the error do, and replaying the whole run can count them again.

Why does Make use operations when nothing happened?

A trigger module's scheduled check is itself an operation. Use an instant trigger or a less frequent schedule if most checks find nothing.

Is moving from Zapier to Make (or the other way) the fix?

Sometimes, but the counting rules differ, so model your actual workflows first. Fixing filters, triggers and batching often saves more than switching.

How often should I review usage?

Monthly is usually enough. Look at the top three workflows by usage and ask whether each step earns its place.

If your automation bill has crept up and you want someone to trace which workflows cause it, send us a list of your busiest ones.