How Much Does AI Automation Cost for a Small Business?

AI automation cost for a small business has 5 parts: the labour to build it, a platform subscription, LLM API usage, hosting and ongoing maintenance. At current published prices, the running costs of a typical workflow come to tens of dollars a month, so the build labour and the maintenance hours decide the real bill. The rest of this guide prices each part with figures from vendor pricing pages, checked on 31 October 2026, and shows how to work out whether a project pays for itself.

I am Head of Automation at a US marketing agency, where I build automations with Python, n8n-style workflows and AI agents. The numbers below are public list prices, not quotes.

The 5 cost components of AI automation

Every AI automation project draws on the same 5 budget lines, for example a single email triage flow and a custom agent alike. Price each one separately.

1. Build and setup labour

Build labour is the hours spent mapping the process, building the workflow, writing prompts, connecting tools and testing on real data. Price it as hours multiplied by an hourly rate. For an in-house benchmark, the US Bureau of Labor Statistics (2025, Occupational Outlook Handbook) puts the median annual wage for software developers at $135,980 in May 2025, before benefits.

I found no reliable published survey of AI automation project prices, so treat any "average project cost" online with suspicion. Quotes vary with scope, the number of systems involved and how clean your data is.

2. Platform subscriptions

Workflow platforms charge a monthly fee that scales with usage. These are the published entry plans as of 31 October 2026:

Sources: Zapier pricing (Zapier, 2026), n8n pricing (n8n, 2026) and n8n hosting documentation (n8n, 2026), which states that "without a license key, n8n runs as the free Community edition." Make also sells credit-based plans, with 1,000 credits a month on its free tier, according to Make's pricing page (Make, 2026).

The billing unit matters more than the headline price. Zapier counts a task each time an action step completes, and its pricing page notes that some AI steps cost more than 1 task. n8n states that pricing "is based on monthly workflow executions, regardless of complexity." A 10-step workflow that runs 1,000 times a month uses 1,000 n8n executions but roughly 9,000 Zapier tasks, because the trigger is free and each of the 9 actions counts. Multi-step AI workflows therefore cost far more on per-task platforms.

3. LLM API usage

LLM providers charge per million tokens (MTok), with output tokens priced higher than input. These are the standard list prices on 31 October 2026:

Sources: Anthropic pricing documentation (Anthropic, 2026) and OpenAI API pricing (OpenAI, 2026). Both providers offer a Batch API at 50% off for work that does not need an instant answer, for example overnight report summaries.

Worked example 1, an AI-assisted enquiry workflow. Assumptions: 2,000 inbound enquiries a month, each classified and given a drafted reply, with 1,500 input tokens (instructions plus the message) and 300 output tokens per enquiry. That is 3 million input and 0.6 million output tokens a month.

• Mid-tier model (Claude Sonnet 5.5 or gpt-6.1-sol): 3 × $2 + 0.6 × $10 = $12 a month.
• Small model (Claude Haiku 5.5 or gpt-6-luna): 3 × $0.10 + 0.6 × $0.50 = $0.60 a month.

Worked example 2, a tool-using agent. Agents call the model several times per task and resend the growing context each time. Assumptions: 500 tasks a month, 6 model calls per task, 6,000 input and 500 output tokens per call. That is 18 million input and 1.5 million output tokens.

• Mid-tier model: 18 × $2 + 1.5 × $10 = $51 a month.
• Small model: 18 × $0.10 + 1.5 × $0.50 = $2.55 a month.

Two details change these numbers in practice. Anthropic notes that Claude 4.7 and later models use a tokenizer that "produces approximately 30% more tokens for the same text," so the same prompt costs more tokens on one provider than another; budget a 30% buffer. Prompt caching cuts the cost of a long, repeated system prompt to a fraction of the input price on both providers.

4. Hosting

Hosting costs nothing extra on Zapier, Make or n8n Cloud, and a few dollars a month for self-hosted tools or custom Python code. A DigitalOcean Basic Droplet (DigitalOcean, 2026) with 1 GiB of memory costs $6 a month, and the 2 GiB size costs $12. Add backups and an uptime monitor.

5. Maintenance

Maintenance is the recurring labour that keeps the automation working: fixing broken connections, updating prompts and handling edge cases. Budget fixed monthly hours for it from day one.

Price tiers by project type

Published market data for AI automation project prices does not exist in a form I can verify, so these tiers describe relative effort, not dollar ranges. Projects fall into 3 tiers:

• Simple workflow: moves data between 2 or 3 apps with fixed rules, for example a form submission added to a CRM and a Slack alert. Days of work, low running costs, little maintenance.
• AI-assisted workflow: adds 1 or 2 model calls for judgement, for example classifying enquiries or drafting replies. Extra time goes into prompt testing on real examples and a human review step.
• Custom agent with integrations: a model that chooses which tools to call, reads and writes to several systems and handles multi-step tasks. Weeks of work, the highest API usage per task, and real engineering for logging, permissions and failure handling.

Build vs buy: no-code tools or custom Python

Choose a no-code platform when the workflow fits its connectors, and custom Python when it does not. Zapier and Make win on speed to launch and need no server. n8n sits in between: visual building with the option to self-host and add code steps.

Custom Python wins in 3 situations:

• High volume: per-task fees grow with every run, while a server's cost stays flat.
• Unsupported systems: internal databases, legacy software or APIs with no ready-made connector.
• Complex logic: retries, branching and data validation that become unreadable as dozens of visual nodes.

Page generation at scale is a common custom-Python job; my guide on how to use Python for programmatic SEO walks through one build from dataset to WordPress drafts.

The trade-off is ownership. Custom code needs someone who can read it and fix it, so weigh the build saving against who maintains it in 2 years. The same question applies to hiring help in general, and my comparison of an independent consultant versus an agency covers how to judge it.

Hidden costs of AI automation

The costs that blow budgets rarely appear in the first quote. Plan for these 4:

• API changes: connected apps change fields, authentication and rate limits, and model providers retire older models, so prompts need retesting.
• Monitoring: a workflow that fails silently costs more than one that never ran, so it needs alerts and logs.
• Human review: AI output that reaches customers needs a person checking it, at least until error rates are measured.
• Data privacy: customer data sent to an LLM provider needs a reviewed data processing agreement and a decision on what data the model is allowed to see.

My security background shapes the last point. An agent with write access to your CRM and email is an attack surface: a malicious instruction hidden in an inbound email can steer it. Give each automation the narrowest permissions it needs, and keep a human approval step before anything is sent or deleted.

How to estimate the ROI of AI automation

Estimate ROI by valuing the hours the automation saves at the full cost of the staff who currently do the work. Use this formula:

• Monthly value = hours saved per month × loaded hourly cost
• Net monthly saving = monthly value − monthly running cost (platform + API + hosting + maintenance)
• Payback period in months = build cost ÷ net monthly saving

Loaded hourly cost includes benefits, not only wages. The US Bureau of Labor Statistics (2026, Employer Costs for Employee Compensation) reports that private industry employers spent an average of $46.89 per hour worked in June 2026, of which wages were $32.82 and benefits $14.07.

Illustrative example, not a client result: an automation saves 20 staff hours a month. Monthly value is 20 × $46.89 = $937.80. Running costs are $81 (Zapier Team at $69 plus $12 of API usage from worked example 1), so the net monthly saving is $856.80. Every $1,000 spent on the build then pays back in about 1.2 months. Subtract maintenance hours from the saving, and measure real hours saved after launch.

Where to start

Start with one high-volume, rule-based task that someone on your team does every day, for example sorting inbound leads, copying form data into a CRM or tagging support emails. Such tasks have 3 properties that make a first project succeed: the steps are already known, the volume makes the saving measurable, and a mistake is cheap to catch.

Record how long the task takes today and how often it runs before anyone builds anything; that baseline turns the ROI formula into a real number. Marketing teams often pick lead handling or reporting first; if AI visibility is also on your list, my guide to getting your business mentioned in ChatGPT covers the search side. You can read more about my background on the about page.

Frequently asked questions

How much does AI automation cost per month?

Running costs for a small-business workflow are typically tens of dollars a month at current list prices: $19.99 to $69 for an entry Zapier plan or €20 to €50 for n8n Cloud, plus a few dollars to about $50 of LLM API usage. Maintenance labour comes on top.

What does it cost to build an AI agent?

The cost to build an AI agent is mainly engineering labour, driven by how many systems it connects to and how much testing it needs. Its running cost is higher than a simple workflow because an agent makes several model calls per task, about $51 a month for 500 tasks in this article's mid-tier model example.

Is n8n cheaper than Zapier for AI workflows?

n8n is cheaper for multi-step workflows because it charges per execution regardless of the number of steps, while Zapier charges per action step. Self-hosted n8n Community edition has no licence fee, but you pay for a server and its upkeep.

Which LLM is cheapest for business automation?

Small models such as Claude Haiku 5.5 and gpt-6-luna cost $0.10 per million input tokens and $0.50 per million output tokens, 20 times less than the mid-tier models. Use them for classification and extraction, and reserve larger models for tasks where quality measurably improves.

How do I calculate ROI on AI automation?

Multiply the hours saved each month by the loaded hourly cost of the staff involved, then subtract the monthly running cost. Divide the build cost by that net saving to get the payback period in months.