How much does AI automation cost for a small business?

When you ask for a quote on an automation, the price has three parts that are paid in different ways. Here is each one, with October 2026 prices, and the math worked through on a real example. What a small business can automate is covered in AI automation examples for small businesses.

The three costs at a glance

The build: the largest part

The build is paid once and depends on four things:

• How many systems the workflow connects. A form that writes to a spreadsheet is simple. A workflow that reads emails, looks the customer up in your CRM, writes to your invoicing software and sends a WhatsApp message has four connections, each with its own rules.
• How many exceptions it has. The expensive part isn't the normal path; it's the cases where information is missing, a customer writes something unexpected or a system doesn't respond.
• Whether it has an AI step. The model's instructions are written and tested on real examples until the results are right.
• Who builds it. A freelancer costs less than an agency, but what matters is who fixes it six months from now.

Two public examples show how wide the range is: a freelancer in Romania offers one well-defined n8n or Make workflow from $750, delivered in seven days, and a studio that specialises in n8n quotes $5,000 to $15,000 per project for multi-workflow systems.

The tools: from free to a few tens of dollars a month

The difference between an execution and a step matters. A ten-step workflow that runs a thousand times a month is 1,000 executions on n8n, but about 10,000 credits on Make and 10,000 tasks on Zapier. The full comparison is in n8n vs Make vs Zapier.

The AI model: paid by tokens

AI models are billed through the API by the number of tokens that go in (instructions, business information, the message) and come out (the reply). Anthropic's prices in October 2026, per million tokens:

The Haiku 5.5 price applies to requests of up to 100,000 tokens; what a small business sends is usually far shorter. The part that repeats with every request (instructions, the price list) can be cached, and cache reads cost much less.

A worked example: a thousand emails a month

Say a business gets a thousand emails a month, and the automation reads each one, labels it and drafts a reply. Each request sends the model about 4,000 tokens (instructions, business information, the email) and gets 400 back.

Add the tool: the workflow runs a thousand times, so it fits in n8n's Starter plan at €20 a month. The monthly total comes to roughly $25 to $50, depending on the model. For labelling and simple questions, a small model is usually enough; for long or sensitive emails, it's worth testing a larger one.

When it pays off

The math is simple: hours the work takes each month, times what an hour costs you. If someone spends 20 hours a month on those emails and their time costs the business $30 an hour, the work is worth $600 a month. Against a monthly cost of around $50, the build pays for itself within months.

The math fails in two cases: when the work takes an hour a month, and when the workflow changes so often it has to be rebuilt constantly. Then it's cheaper to keep doing it by hand.

What to ask for in a quote

1. The build price, with the workflows and exceptions included in writing.
2. The estimated monthly cost of tools and model at your volume.
3. Whose name the accounts are in (n8n, Make, the model's API key).
4. What happens when a step fails: who gets the alert and who fixes it.
5. What a change after launch costs.

Published prices from agencies in Romania, side by side: Best AI automation agencies in Romania.

Our AI automation page explains how we work: we choose the tool by volume, cost and how often the workflow changes, and the quote lists every cost, not just the build.