Automation priced per task
Zapier charges you per task. Make charges you per credit. Both meters tick once for every step an automation takes, so the workflow that runs most often is the one that costs most. Nobody hides this. It sits on both pricing pages. But it is a strange incentive to sign up for, and most teams only notice the shape of it when a workflow they are proud of turns into the biggest line on the invoice.
What a task actually is
Zapier counts one task for each successful trigger or action step that touches another app. Its own built-in tools do not count: Formatter, Paths, Filters, Delay, Looping, Storage and Tables all run at zero tasks. Steps that call a model cost more than one. On the advanced model tier an AI step costs 3 tasks, on the premium tier 5, and a Zapier MCP tool call costs 2. The Free plan gives you 100 tasks a month with two-step Zaps and 15-minute polling. Paid plans start at $19.99 a month and the price moves with the task tier you pick, up to 2 million tasks.
Make uses the same idea with different words. One credit per module action, so adding a row to a Google Sheet spends a credit the same way fetching a Gmail message does. Free gives you 1,000 credits a month, 2 active scenarios and a 15-minute minimum gap between runs. Core is $12 a month for 10,000 credits, Pro $21, Teams $38. Unused credits expire at the end of the term, which is worth knowing if your volume is seasonal.
Read carefully and the model is fair. You pay for the work the platform does on your behalf, and at low volume it is a bargain. The problem is what the unit is measured against.
The bill grows with the thing that is working
Take an order workflow with five billable steps: read the order, look up the customer, create the invoice, notify fulfilment, write back to your internal system. At ten orders a day that is 1,500 tasks a month. At a hundred orders a day it is 15,000. The logic did not change. The steps did not get harder. Your business got better and the meter responded.
Zapier is explicit about what happens at the edge of your tier: hit the limit and you move to pay-as-you-go unless you turn that off or step up a tier. That is a reasonable default and it is also why the invoice tends to find out about your growth before finance does. On a queue you own, the same growth registers as a slightly busier consumer.
None of this makes per-task pricing wrong. It makes it a pricing model that assumes automation volume is a proxy for value delivered, and for a lot of back-office work that assumption breaks down early. Syncing a record is worth roughly the same whether you do it twice a day or two thousand times.
What the same runs cost on infrastructure you rent directly
Cloudflare Queues charges per operation, where an operation is 64 KB written, read or deleted. Delivering one message normally takes three: a write, a read and a delete. The Workers Paid plan costs a $5 monthly minimum and includes 1 million queue operations, then $0.40 per additional million. So a million messages produced and consumed in a month works out to 3 million operations, of which 2 million are billable, which is $0.80. The same plan includes 10 million Worker requests and 30 million CPU milliseconds before any request charge starts.
That number is not the whole comparison and pretending otherwise would be dishonest. Zapier ships roughly 9,000 app connectors, OAuth token refresh, retry behaviour, an execution log and a visual editor your operations lead can use without you. On your own queue you write all of that, or you pick libraries that do. What you buy back is a cost curve that flattens and logic you can put under test.
We saw the second half of that trade on Stay World Class. They had outgrown Webflow and Xano, and we moved them to Next.js, NestJS and Supabase with a four-stage ETL that mapped every legacy ID onto a PostgreSQL UUID so nothing lost its history. Lighthouse went from 55.91 to 91 out of 100 and LCP got 77% faster. The performance was the visible part. The quieter win was that business rules which used to live in a visual builder became code with tests around it.
When the meter is the better deal
If you run a few thousand tasks a month, stay where you are. Twenty dollars will not buy you an afternoon of engineering time, and no self-hosted queue gives you a working HubSpot connector before lunch. If your automations change every week, if the people who change them do not write code, or if the integration you need is one of the long tail Zapier already maintains, the per-task price is buying you real flexibility and you should keep paying it.
The number worth watching is not the invoice. It is monthly task volume plotted over the last twelve months. Flat or lumpy, stay put. Compounding, with the same handful of workflows responsible for most of it, and you are paying rent on logic that has stopped changing. That is usually the moment to move the busiest two or three workflows onto something you own and leave the rest exactly where they are.
