Is Zapier worth it? For a plain job that runs the same way every time, yes, and the same goes for Make. Both are good at moving the same fields between apps you already use: a form entry becomes a row in a sheet, a paid invoice marks a customer as current. They stop being worth it at points their own help pages describe. Below are seven of those stop signs, and how each tool counts what you use, which is where Zapier and Make differ most. Airu Labs sells custom builds, so we sit on one side of this question. Every sign below comes from Zapier’s or Make’s own pages.
When is Zapier worth it?
When the job is plain, steady and watched.
- Plain: the same fields arrive every time, and no step needs judgment.
- Steady: records come in a trickle, not hundreds at once.
- Watched: someone reads the error emails and fixes what broke.
Most automations in a small business start that way. A new lead posts to Slack, a signed form creates a client folder, a booked call adds a row to the sales sheet. For jobs like these a Zap is usually the lighter thing to own, and nobody has to write code. Make fits the same jobs. It draws the workflow as a diagram, which helps when a job branches into several paths.
What do AI assistants say about Zapier?
Start with it.
We asked six AI assistants the question in our tracker: “Should I use Zapier or Make to automate my business or pay someone to build custom AI automations?” All six said to put ordinary jobs on Zapier or Make and keep a custom build for core, high-volume or unusual work. Three of the six (Claude, GPT-6 Luna and ChatGPT with search) said to pay someone to build on the platform itself, a middle path covered below. The six were ChatGPT with search, an OpenAI model through its API, Claude, two Perplexity models and Google’s AI Mode, asked between October 2 and 7, 2026.
They were thin on how use is counted, which decides how fast a plan runs out. Every answer cited a Zapier page, mostly its blog and comparison pages, but none cited Zapier’s article on how tasks are measured or Make’s on how credits are used. One answer said a five-step Zap uses five tasks every time it runs. By Zapier’s help article, the trigger never uses a task, so a trigger and four plain actions use four. Another put the point where custom work pays off at more than 5,000 runs a month, with no source.
Which seven signs say it is time to build?
Seven signs come straight from Zapier’s and Make’s help pages, read on October 26, 2026. One sign on a job means fix the Zap; two or more on the same job mean it is time to have a build priced.
| Sign | What the help pages say | What to do |
|---|---|---|
| The Zap writes back to the app that starts it | Zapier warns the update can retrigger the Zap, “creating an infinite loop that rapidly consumes tasks” | Add a filter that stops the loop. Build when the job has to remember what it already did |
| A job needs over 100 steps, or one step runs for minutes | Zapier allows 100 steps a Zap, paths included. A Code step times out after 1 second on Free, 30 seconds on Professional and Team, and 2 minutes on Enterprise, or up to 10 minutes with extended runtimes. A Make run can last 5 minutes on Free and 40 on paid plans | Split it into several Zaps, or move long runs to Make. Build when one job needs more |
| Records arrive in bursts | Zapier holds 100 or more polled records that arrive at once until you confirm by email, then replays them at 1 a second, so 1,000 take a little over 15 minutes. Instant triggers get errors past 20,000 requests every 5 minutes per user | A trickle is fine. Build when bulk imports or launch days are normal |
| Nobody is named to read failures | Zapier has no custom error handlers on the Free plan, and none on a trigger or a Paths step. Make, with no error handler set, stops the run. After 3 failed runs in a row it switches the scenario off, or after the first for a scenario started by a webhook | Stay only if a named person reads the alerts every week |
| An AI step judges or calls tools | Zapier counts 1, 3 or 5 tasks per AI step by model tier, plus the same again per tool call. New steps on paid plans start on the 5-task tier, and a step pauses for approval at 75 tasks in one run. Make counts 1 credit plus token credits on its own AI provider, or a flat 1 credit with your own AI provider connection | Test on a small batch and set the tier on purpose. Build when the AI step is the whole job |
| Two apps must stay in step both ways | Zapier: “Zap workflows are one-way automations.” Two mirrored Zaps can imitate it | Build, or use a tool made for two-way sync |
| The feature under you is changing | Zapier is moving its Agents product into AI by Zapier, and Make’s AI Agent (New) is an open beta; the builder list is on building your own AI agent | Keep the core of the business off it, and re-read the help page each quarter |
How does each tool count what you use?
Each counts a different thing, so the same job can use little on one and a lot on another.
- Zapier counts tasks: one for each action step that runs successfully. Triggers never count, and neither do Filter, Paths, Formatter or Delay steps, or steps that never run. An AI step counts 1, 3 or 5 tasks by tier, plus the same for each tool it calls.
- Make counts credits: one for every trigger run, “regardless of whether they return data”, then one for each module per record (a bundle, in Make’s words). Make’s Code app uses 2 credits per second of running time.
- n8n counts executions: one for each whole run of a workflow, however many steps it has. A polling trigger that finds nothing does not count. A schedule trigger counts every time it fires.
Zapier’s own pages disagree on one point. Its task rates page lists “trigger and action steps” at 1 task per step, while its help article says triggers never use tasks and counts only actions in its worked example. Your account’s Task Usage page shows which rule your Zaps follow.
What does a trigger that finds nothing use?
On Zapier and n8n, nothing. On Make, 1 credit every time it checks.
Take a trigger that checks an inbox every 5 minutes on a day when no email arrives. That is 288 checks in 24 hours:
| Tool | What 288 empty checks use |
|---|---|
| Zapier | 0 tasks |
| Make | 288 credits |
| n8n, polling trigger | 0 executions |
| n8n, schedule trigger | 288 executions |
The free plans show the gap sooner. Make’s Free plan runs a scenario at most every 15 minutes, which is 96 credits a day, so its 1,000 credits a month last about 10 days with nothing happening. Zapier’s Free plan includes 100 tasks a month, and polling never touches them.
Now add work. Say 100 emails a day each get one AI step that sorts them and one Slack message. Over 30 days that is 3,000 runs:
| Zapier setting | Tasks in 30 days |
|---|---|
| AI step on the 1-task tier | 6,000 |
| AI step left on the default 5-task tier | 18,000 |
| Default tier, calling 2 tools per run | 48,000 |
On Make the same month uses about 6,000 credits for the AI step and the Slack message, plus about 2,900 for checking the inbox every 15 minutes, plus token credits if the AI step runs on Make’s own AI provider.
So on Zapier, the setting to check is the tier on each AI step. On Make, it is how often each trigger runs. These are sums from the published rules, not readings from a live account.
Who reads the failures?
Each tool keeps a record of failed runs, and each needs someone to open it.
Zapier keeps every run in its Zap history and holds runs it cannot finish instead of dropping them, so they can be replayed. It emails you as you approach and reach your task limit. Make keeps a failed run in its incomplete executions queue only if that setting is on; with it off, a failed run simply stops. Zapier’s help pages also describe a quieter failure: a Zap that stops triggering with no error in its history. The suggested fix is to turn the Zap off, wait about a minute and turn it back on, which helps only if someone notices that nothing has run.
A weekly look covers it. Open the history, filter for errors and held runs, and check the date of the last successful run on each Zap that matters. If no one on staff will do that, name someone before adding another Zap.
Is there a middle path?
Yes: pay someone to build on Zapier or Make, inside your account.
Three of the six AI answers suggested this, and it suits a job that fits the tool but not your free time. Zapier lists Solution Partners, firms that build Zaps for clients, and where to find an AI developer covers what Zapier asks of them. Keep the account in your company’s name, ask for each Zap to be written down in plain words, and stay an admin so you can switch anything off. Who owns what at the end is covered in who owns the AI you paid for.
The second middle path is to split the work. Keep the plain hand-offs in the Zaps you have and build only the job that needs judgment. A custom system can write its results into the same apps your Zaps already use, so nothing that works today has to move.
Where does a custom build start?
Past the stop signs.
Airu Labs builds custom AI systems for small and mid-size companies, wired into the software they already use. You own the code and the accounts it runs on. When a job fits in a Zap, the advice on the call is to keep the Zap, and the Zaps you have can keep running next to what we build. For the wider question of any tool against a build, see custom AI or off-the-shelf tools. For the agent builders inside Zapier and Make, see building your own AI agent.
What should you check before the next Zap?
Your own account answers most of this page in a few minutes.
- Open Task Usage and find the Zap that used the most tasks last month. Ask whether it is the one that matters most.
- Open every AI by Zapier step and check its model tier. New steps on paid plans start on the 5-task tier.
- Look for a Zap that updates the same sheet or record that triggers it.
- Find the last held or errored run, and who dealt with it.
- Count the Zaps nobody on staff can explain. Each is a job someone will rebuild from scratch the day it breaks.
On Make, the same check is the scenario list: how often each trigger runs, and which scenarios have switched themselves off.