How much does AI automation cost for a small business in 2026
By Mariam Raouf. Published 2026-09-08. 7 min read.
Nobody will give you a straight number, so here is one. A small business automation setup costs somewhere between $20 and $400 a month to run, and the reason that range is so wide has almost nothing to do with how much automation you build.
It has to do with how your platform counts.
That is the part that catches people. You compare subscription prices, you pick the cheap one, and four months later the bill has quadrupled because the tool you chose charges you per step and you built something with eight steps. Below are the actual published numbers as of September 2026, where the money really goes, and how to work out your own figure before you commit to anything.
The three things you are actually paying for
Most articles on this treat AI automation as one line item. It is three, and they behave completely differently.
The platform. The thing that connects your apps and runs the sequence. Zapier, Make, n8n. This is a predictable monthly subscription plus usage.
The model. If any step involves an AI actually reading, writing or deciding something, you pay per token to whoever makes the model. This is pure usage and it is usually smaller than people fear.
The build. Somebody has to design and wire the thing. That is either your weekend or someone's invoice.
The first two are published and you can calculate them today. The third is the one where quotes vary wildly, and I will get to why.
Platform pricing, September 2026
| Platform | Entry paid plan | What you get | How it meters |
|---|---|---|---|
| Zapier | from $19.99 a month | Multi step Zaps, webhooks, premium apps | Per action |
| Make | $12 a month, about $10.20 on annual billing | 10,000 credits, unlimited active scenarios | Per credit, roughly per action |
| n8n Starter | €20 a month billed annually | 2,500 executions, unlimited steps, unlimited users | Per execution, meaning per whole run |
| n8n self hosted | No licence fee | Community Edition, you pay only your own server | Unlimited |
Sources: Zapier pricing, Make pricing, n8n pricing. Checked September 2026.
Look at that last column, because it is the whole article.
The metering trap, and the one calculation that saves you money
Zapier's own help documentation is clear about what it charges for: "A task is any successful action that runs in Zapier. Only successful actions count toward your task usage."
n8n's documentation is equally clear, and says something completely different: "An execution is a single run of your entire workflow. It doesn't matter how many steps are in the workflow or how much data it processes. It's still a single execution."
Now run the same automation through both.
Say you build something ordinary. A form comes in, the lead gets checked against your CRM, created if new, tagged, assigned to someone, a Slack message fires, a confirmation email goes out, and a row lands in a reporting sheet. That is eight actions. Say it runs 2,000 times a month, which is a busy but not enormous small business.
On Zapier that is 16,000 tasks. On n8n it is 2,000 executions, which sits inside the €20 Starter plan. Self hosted, it is your server bill and nothing else.
Same automation. Same volume. Wildly different invoice.
This is why "what does automation cost" has no single answer, and why the honest version of the question is "what shape is your automation, and does your platform charge for shape or for runs?" A few automations that each do a lot of steps punish you on Zapier. A lot of automations that each do very little are fine anywhere.
One more thing worth knowing about Make specifically: it now bills in credits, not the "operations" that every comparison table online still describes, and its own pricing page notes that "some advanced features leveraging Make's AI Provider may use more" than one credit per action. So an AI flavoured scenario burns quota faster than the same scenario built on plain connectors. If you read a Make cost comparison written before this year, the units in it are wrong.
What the AI part actually costs
This is the number people brace for and it is usually the smallest one on the page.
Current published API pricing, per million tokens:
| Model | Input | Output |
|---|---|---|
| Claude Haiku 4.5 | $1 | $5 |
| Claude Sonnet 5 | $2 | $10 |
| Claude Opus 5 | $5 | $25 |
| GPT-5.6 Luna | $0.20 | $1.20 |
| GPT-5.6 Terra | $2.00 | $12.00 |
| Gemini 3.5 Flash Lite | $0.30 | $2.50 |
Sources: Anthropic pricing, OpenAI pricing, Google Gemini pricing. Checked September 2026.
A million tokens is roughly 750,000 words. For most small business tasks, classifying an enquiry, drafting a reply, pulling structured data out of a document, you are spending fractions of a cent per run. Two thousand of those a month on a cheap model is a rounding error next to your platform bill.
One trap to note, because it is dated and specific. Google's Gemini 3.8 and 3.7 Flash prices are promotional through 31 December 2026 and double on 1 January 2027, going from $0.75 to $1.50 on input and $3.75 to $7.50 on output. If you are budgeting a year out on those models, budget the real number.
Agent platforms bill differently again and get expensive faster. Microsoft Copilot Studio charges $0.01 per Copilot Credit, where a generative answer is 2 credits and an agent action is 5. So one useful interaction is around 7 credits, or 7 cents. Ten thousand interactions a month is roughly $700, before any model cost at all. Zapier Agents works on activities rather than tasks, and its documentation notes that browsing the web and looking up attached knowledge each burn an activity, so a single useful outcome can consume several.
So what does the whole thing cost
Putting real numbers together for a business running a handful of automations at moderate volume, as of September 2026:
| Line item | Realistic monthly range |
|---|---|
| Automation platform | $12 to $70 |
| Model API usage | under $5 to about $40 |
| Agent platform, only if you need one | $33 to several hundred |
| Total to run | roughly $20 to $150 a month for most small businesses |
Build cost sits outside that and it is where quotes diverge by an order of magnitude. I am not going to invent an industry average, because no credible survey of what US small businesses pay for automation builds exists. What I can tell you is what drives the number: how many systems have to talk to each other, whether they have decent APIs, and how much of your process is written down before anyone starts. That last one moves the price more than the technology does.
What to do before you spend anything
Count the steps in the process you want to automate. Actually count them, on paper.
Multiply by how many times a month it runs. That gives you your task number, and you can price Zapier against n8n directly instead of guessing.
Then ask the harder question: does this process need an AI in it at all, or does it just need to be connected? Anthropic's own engineering guidance says to find "the simplest solution possible, and only increasing complexity when needed", and warns that "the autonomous nature of agents means higher costs, and the potential for compounding errors". A deterministic workflow that runs the same way every time is cheaper, faster and easier to debug than an agent, and for a large share of small business tasks it is also the correct answer.
Start with one process. The boring one that everybody in the business complains about. Price it, build it, watch it for a month, then decide whether the second one is worth it.
The honest summary
Running the automation is cheap. Almost always cheaper than people expect, and often under $100 a month for a small business doing real volume. The expensive parts are choosing a platform whose billing model fights your workload, and building something complicated before you have written down what the simple version was supposed to do.
If you want to work out your own number, the calculation in this article is genuinely all of it: steps, times runs, against your platform's unit. Ten minutes with a piece of paper will tell you more than any quote.
At Calpir this is most of what we do, connecting the systems a business already has so the boring sequences run themselves. If you want a second opinion on whether a process is worth automating at all, our AI automation and AI consulting pages explain how we work, and I am happy to tell you when the answer is no. Quite often it is.
Frequently asked questions
Is Zapier or n8n cheaper for a small business?
It depends entirely on the shape of your automations. Zapier charges per successful action, so an eight step workflow costs eight tasks each time it runs. n8n charges per execution, so the same workflow costs one. If you run a small number of long workflows, n8n is dramatically cheaper. If you run many very short ones, the gap narrows a lot.
Can I run AI automation for free?
Partly. n8n's Community Edition has no licence fee and you pay only for the server you run it on. Zapier and Make both have free tiers, at 100 tasks and 1,000 credits a month respectively, which is enough to test an idea but not to run a business on. Model API usage is never free at volume, though for light workloads it costs cents.
How much does an AI agent cost compared to normal automation?
Meaningfully more, in both directions. Agent platforms meter more aggressively, Copilot Studio charges 5 credits per agent action against 2 for a generative answer, and agents make more model calls per outcome because they loop. Anthropic's guidance is explicit that agentic systems trade cost and latency for flexibility, so only pay for one where the flexibility is the point.
Why do automation agencies quote such different prices?
Because most of the cost is discovery, not building. An agency quoting from a clear written process, with systems that have real APIs, is pricing a known job. One quoting from a vague description is pricing risk. If you want comparable quotes, write down the process first and send the same document to everyone.
Do these prices change often?
Yes, and quickly. Every figure here was checked in September 2026 against the vendor's own pricing page, and one of them, Google's Flash promotional rate, has a published expiry date of 31 December 2026. Treat any automation pricing article without a date stamp as fiction.