The Running Costs of Automation Tools, Explained
By CodexierPublished 6 min read
An automation is cheap to build and easy to forget, until the monthly invoice from the automation platform doubles because a flow started running on every incoming email. The running cost is a pricing model multiplied by your volume, plus the APIs and AI models the flow calls. Here is how the three common models scale, what sits on top, and how to cap the total.
Three pricing models in automation tools
Zapier-style platforms count tasks: every action step that completes is one unit. Make counts operations in a similar way, including some steps you might not think of as work, such as routers and iterators over lists. n8n in the cloud counts executions, meaning one full run of a workflow no matter how many steps it takes, and self-hosted n8n costs whatever your server costs. Microsoft Power Automate is mostly licensed per user, with premium connectors on higher licences. We compare the platforms themselves in Zapier vs Make vs n8n; this article is about what they cost once they are running.
How costs scale with volume
The same business process can land in very different price tiers depending on the model. Take a flow that reads a new web lead, looks it up in the CRM, creates or updates the contact, posts a message to the sales channel and sends a confirmation email: five steps. At fifty leads a month nobody cares which model you use. At two thousand leads a month the per-task model counts ten thousand units, while a per-run model counts two thousand.
| Situation | Per task / operation | Per run / execution | Per seat |
|---|---|---|---|
| Few runs, few steps | Cheap, often a free tier | Cheap | You pay the seat regardless |
| Many runs, short flows | Grows linearly with runs | Grows linearly with runs | Flat, good value |
| Many runs, long flows with loops | Grows with runs times steps, the expensive case | Grows only with runs | Flat, but premium connectors may need a higher licence |
| Many users building flows | Shared pool, fine | Shared pool, fine | Grows with every user |
The practical consequence: design matters for cost. Filtering early (stop the flow before the expensive steps when the input is irrelevant), batching (process yesterday's orders once at night instead of one by one), and avoiding polling triggers that check for changes every few minutes all reduce the count without changing what the flow achieves.
API and AI usage on top
The platform is only the plumbing. Many flows call services that bill separately, and those bills scale with volume in their own way. AI steps are the fastest-growing item: a language model is priced per token, so the cost depends on how much text you send in and get back, and a flow that pastes a whole email thread plus a long instruction into every call pays for all of it every time.
- AI models: priced per input and output token; shorter prompts, trimmed context and a smaller model for simple classification cut cost directly.
- SMS and voice: priced per message or minute, and Swedish mobile numbers are not the cheapest destination.
- Paid data APIs: company lookups, address validation and credit checks often bill per lookup.
Where self-hosting saves and where it costs
Self-hosting an open-source tool such as n8n removes the per-run bill entirely, and for high-volume flows the saving is real. It also lets you keep data on a server in the EU that you control, which simplifies the GDPR picture. The cost moves somewhere else: someone must patch the server, update the software, watch the queue, back up the database and restore it when something goes wrong.
Self-hosting pays off when
Volumes are high and steady, flows are business-critical enough that you already have someone responsible for them, and data residency matters to your customers.
Cloud pays off when
Volumes are modest, nobody on the team wants to be on call for a server, and the flows can tolerate the platform's own outages.
The hidden line item
Maintenance hours. A server nobody updates is a security risk, so price at least a small monthly amount of someone's time into the comparison.
Setting a monthly cost ceiling
- List every active flow with its trigger, steps per run and expected runs per month.
- Multiply to get units per month in the platform's own currency: tasks, operations or executions.
- Add external usage: tokens per AI call times calls, SMS per month, paid lookups.
- Set hard spending limits where the provider supports them, both on the platform and on the AI provider account.
- Add an alert at a threshold below the limit, sent to a person who can act on it.
- Review the list quarterly and switch off flows nobody uses; dead automations still consume units if their trigger fires.
When you do not need help with this: if you run a handful of flows on a free or entry tier and the bill has been stable for months, leave it alone. It becomes worth an outside look when the bill grows faster than the business, when AI steps have been added without a limit, or when nobody knows which flows are still in use. Our workflow automation system is designed with volume and cost in mind from the start, and the pricing page shows the fixed build price. If you want a second opinion on your current bill, book a call and bring the run history.
Frequently asked questions
Which automation tool is cheapest?
None in general. For short flows with low volume the free tiers of most tools are enough. For long flows with loops on high volume, per-run pricing or self-hosting is usually cheapest. Count runs and steps first, then compare.
Why did our automation bill suddenly jump?
The usual causes are a trigger that started firing more often than intended, a loop over a longer list than before, a polling trigger checking very frequently, or an error that makes a flow retry repeatedly. Check the run history for the flow with the most units this month.
How do we keep AI costs predictable?
Set a monthly spending limit on the AI provider account, keep prompts short, send only the part of the text the model needs, and use a smaller model for simple tasks such as sorting or tagging. Log token use per flow so you can see which one drives the cost.
Want to know what your flows will cost to run?
Bring a list of the processes you want to automate and rough monthly volumes. In fifteen minutes we can tell you which pricing model fits and where the cost risk sits.
Book a free 15-minute call