Skip to content
Back to all articles

Automation

What an AI automation really costs to run each month

The monthly costs behind an AI automation, from platform plans to per-token AI usage, with a worked example using illustrative numbers.

November 3, 2026

6 min read

A man smiling at his phone at a white desk

Building an automation is usually a one-time job. Running it is an ongoing cost. Once a workflow is live, it draws on several services every month, and each one bills you separately. None of these costs is hidden, but they are spread across different invoices, so it is easy to underestimate the total or to be surprised when volume goes up.

This article breaks the running cost into its parts and shows how to estimate it before anything is built. Every number below is an illustrative round figure, chosen to keep the arithmetic easy to follow. None of them is a current price for any provider. Pricing changes often, so check each provider's pricing page for the rates that apply to you.

The five cost components

A typical AI automation draws on some combination of the following.

The automation platform

Tools like n8n, Make and Zapier run the workflow itself. They watch for a trigger, move data between apps and call the AI model. They measure usage in different ways. Some count each step that runs, some count each full run of the workflow, and some can be self-hosted, in which case you pay for a server instead of a plan.

How the platform counts matters. A workflow with eight steps that runs 1,000 times a month could count as 1,000 units on one platform and 8,000 on another. That difference can move you up a plan tier.

The apps you connect

The automation connects to tools you probably already pay for, such as Gmail, Google Sheets or HubSpot. Sometimes it needs a feature that only exists on a higher plan, for example API access or more custom fields in your CRM. Check this before you build, because a required app upgrade can cost more than everything else on this list.

AI usage, billed per token

When the workflow calls a model from OpenAI, Anthropic or another provider through their API, you pay for what you use. Usage is measured in tokens, which are small chunks of text. A short email might be a few hundred tokens.

You are billed for the text you send in (your instructions, the email, any reference material) and for the text the model writes back. Output tokens usually cost more than input tokens, and larger models usually cost more per token than smaller ones. The main drivers of this line are:

  • How much text you send on each run, including long instructions and attached documents.
  • How many times the workflow runs.
  • Which model you choose for each step.

One point that catches people out: API usage is billed separately from chat subscriptions. A ChatGPT or Claude subscription for your team does not cover the API calls an automation makes.

Enrichment and other APIs

Some workflows call other paid services: looking up company details from an email domain, verifying an email address, reading text from scanned documents or sending text messages. These tend to charge per lookup or per request, sometimes with a monthly minimum.

Someone's time to review

This cost never appears on an invoice. Someone has to check flagged items, handle exceptions, look at the log now and then and fix things when a connected app changes. Even a well-built automation needs some attention. Put an honest number of minutes per week on it, and value that time at what it costs you.

How to estimate before you build

You need five inputs, and rough ones are fine:

  1. Runs per month. Count last month's emails, form submissions or invoices to see how often the trigger would fire.
  2. Steps per run. List them, including the trigger. This drives the platform cost.
  3. Text per AI call. Paste a typical example plus your instructions into a token counter (several AI providers offer one) and note the input size. Estimate the length of a typical answer too.
  4. Other paid calls per run. Enrichment, document reading or messaging.
  5. Review minutes per week. Be honest here, especially for the first month.

Multiply each by the unit price on the relevant provider's pricing page, add them up and leave a buffer for growth and retries.

A worked example with illustrative numbers

Take a hypothetical 6-person marketing agency that wants to qualify inbound leads. Each website form submission starts a workflow that looks up the company from the email domain, asks an AI model to score the lead against the agency's criteria and write a two-line summary, adds the result to the CRM and posts a message in the team chat for high-scoring leads.

Assumptions:

  • 400 form submissions per month.
  • 6 steps per run, so about 2,400 steps a month on the platform.
  • One AI call per run, around 2,000 tokens in (instructions, form data and company details) and 300 tokens out.
  • One enrichment lookup per run.
  • Someone spends about 30 minutes a week checking borderline leads.

Here is how the monthly estimate might look. The unit prices are invented round numbers in a generic currency, used only to show the arithmetic.

ComponentIllustrative basisIllustrative monthly cost
Automation platformA plan that covers 2,400 steps a month30
CRMAlready paid for, no upgrade needed0
AI usageAbout 920,000 tokens in total, at an assumed blended rate5
Enrichment API400 lookups at an assumed 0.05 each20
Review timeAbout 2 hours at an assumed 40 per hour80
Total135

Even with made-up numbers, a few things stand out:

  • In this example, AI usage is the smallest paid line. With short texts and moderate volume, per-token costs can end up small next to platform plans and paid lookups. With long documents or a large model, that can change quickly.
  • Review time is the biggest line. It does not arrive as a bill, which is why it gets ignored. Tightening the scoring rules so fewer leads land in the borderline pile may save more than switching to a cheaper model.
  • Volume changes the picture. At 4,000 submissions a month, the AI and enrichment lines grow roughly tenfold, and the platform may need a bigger plan. Review time depends on how many leads are borderline, not only on the total.

Ways to keep running costs down

  • Use the smallest model that handles the task well, and test it on real examples before deciding.
  • Keep instructions tight. They are sent on every run, so trimming them saves tokens every time.
  • Filter before the AI step. Simple rules can drop spam, duplicates and out-of-scope items before they reach a paid call.
  • Batch work where timing allows, for example one daily summary instead of one per event.
  • Set spending limits or usage alerts where providers offer them, so a loop or a spike in volume does not run up a bill unnoticed.
  • Compare the first full month of real costs with your estimate, and adjust.

When an automation is built inside your own accounts, each provider bills you directly. That keeps every line visible, and you can downgrade or cancel any piece yourself.

Where to start

Take the workflow you are considering and fill in the five inputs: runs per month, steps per run, text per AI call, other paid calls and review minutes. Then open the pricing page of each provider involved and work out your own monthly figure, with a buffer for growth. If the total is small next to the time the task takes today, the automation deserves a closer look.

If you want that one workflow designed and built inside your own accounts, that is what our Single Workflow Automation covers.

Single Workflow Automation, $297 one-time

One new workflow of up to 15 steps across up to 3 apps, with AI steps where they help and alerts if it fails.

See what is included

Get started with Ninelabs OS

Pick a deliverable, pay once and send your brief. If you are not sure which one fits, write to us first and we will tell you honestly.