How much does AI automation cost for a small business in 2026.

What AI automation actually costs: the four layers

When someone asks "how much does AI automation cost," they're usually collapsing four separate expenses into one question. The first is the platform layer: the tool that actually runs the workflow, moving data from one app to another when a trigger fires. Zapier, Make, and n8n live here; our platform-by-platform comparison covers how they differ beyond price. The second is AI model usage: the per-request cost of having a language model read, summarize, or draft something inside that workflow, billed by the token (a token is roughly a chunk of a word). The third is the supporting software the workflow touches: your CRM, your scheduling tool, or your database, each with its own subscription. The fourth is labor: whoever designs, builds, and maintains the thing, whether that's you, an employee, or a consultant.

A workflow that just moves a form submission into a spreadsheet might cost nothing beyond a free platform tier. A workflow that reads incoming customer emails, drafts a reply with an AI model, and logs the interaction in a CRM touches all four layers at once. Pricing pages almost always quote only the first layer, which is why the number on a landing page rarely matches what a business actually spends.

DIY platform pricing: Zapier, Make, and n8n side by side

If you're building it yourself, the platform is your first real decision. The three main options price things differently enough that comparing sticker prices alone is misleading:

The pattern across all three: the free and entry tiers are genuinely usable for a single low-volume workflow. The cost climbs fast once you need multiple users, higher volume, or enterprise features like SSO, which is where teams get surprised. If you're deciding which one fits your situation, that trade-off between price, self-hosting, and feature depth is exactly what our Zapier vs Make vs n8n guide walks through.

The hidden line item: paying for AI model usage

Once a workflow needs an AI model to read or generate text, you're paying for tokens, and this is the layer people most often overestimate. OpenAI's pricing runs from $2.00 per 1 million input tokens and $12.00 per 1 million output tokens for its gpt-5.6-terra model down to $0.05 input and $0.40 output per 1 million tokens for its budget gpt-5-nano model, with gpt-5.6-luna ($0.20/$1.20), gpt-5 ($1.25/$10.00), and gpt-5-mini ($0.25/$2.00) filling in between. Anthropic's Claude models range similarly: Opus 5 at $5/$25 per 1 million tokens, Sonnet 5 at $2/$10, and Haiku 4.5 at $1/$5.

Anthropic's own worked example is the clearest anchor available: processing 10,000 customer support tickets, at roughly 3,700 tokens per conversation, on Claude Haiku 4.5 costs approximately $37 total, well under a penny per ticket. Batch processing (asynchronous jobs) cuts the rate in half again, and cached prompt reads run at a tenth of the base input price. For teams that don't want to build against an API directly, Claude's subscription plans run $20/month for Pro, or $200 billed annually, and Team seats at $25/seat/month billed monthly or $20/seat/month annual.

The upshot: for most small-business workloads, text volume, ticket volume, or document volume, the AI usage line is small relative to the platform and labor lines. It only grows large at genuinely high volume or when using the most expensive frontier models unnecessarily.

What it costs to hire the work out: consultants, freelancers, and agencies

There's no clean published rate card for automation consultants or agencies; retainer and project pricing varies too much by scope to trust generic numbers, and marketplace rate pages weren't independently verifiable at the time of writing. The defensible anchor is BLS wage data: the median hourly wage for software developers nationally is $65.38 (median annual $135,980, May 2025). A small automation build scoped at 20 to 40 hours implies roughly $1,300 to $2,600 in labor value at employee-equivalent rates, before any contractor markup, overhead, or project-management time is added.

That range is a floor, not a quote. It tells you what the underlying labor is worth if you were paying an in-house developer to do it. A freelancer or agency will typically price above that to account for business overhead, and complex integrations or ongoing maintenance push scope, and cost, well past a single build.

Automation vs. hiring a person: a real comparison

Phone handling is a useful case because there's real pricing on both sides. A full-time receptionist costs a median of $38,010 a year nationally, before payroll taxes and benefits, per BLS wage data. Smith.ai's human virtual-receptionist plans run from $300/month for 30 calls up to $2,100/month for 300 calls (with $8.50/call overage at that tier), which works out to $25,200 a year at the Pro plan; add-ons like appointment booking cost $1.50/call. An AI-based scheduling stack, Calendly Standard plus a mid-tier Zapier or Make plan, can run under $500 a year.

That doesn't mean the AI stack always wins. Smith.ai and a human receptionist can handle nuance and unscripted conversation that a scheduling automation can't. The right comparison isn't cost alone; it's whether the call volume and complexity justify a human's judgment or whether most of the traffic is routine scheduling that a workflow can handle without anyone noticing the difference. Our guide to AI receptionists goes deeper on where that line usually falls.

The stack most small businesses actually need, and what it runs per month

A realistic starting stack for a small business automating scheduling and lead handling: Calendly Standard at $10/seat/month (billed monthly; annual saves 17%), a mid-tier Zapier or Make plan in the $20 to $40 range depending on task volume, and Airtable's Team plan at $20/user/month billed annually if you need a shared database beyond a spreadsheet. HubSpot enters the picture if you need a full CRM. Its Starter tier is advertised from $7/mo/seat as a promotional rate against a $20/mo/seat list price, but the jump to Marketing Hub Professional at $800/month plus a mandatory one-time $3,000 onboarding fee is the real cliff to plan around if you outgrow the entry tier.

For most small operations, a working stack lands in the tens of dollars per seat per month at the entry tiers, with AI token usage adding only a few more dollars unless volume is high. The number that actually matters isn't the sticker price of any one tool; it's what happens to your bill when you cross a task, seat, or record threshold, since that's where most of these platforms' pricing steps up sharply.

When automation isn't worth it yet

Automation earns its cost when a task repeats often enough that the setup investment pays back, and when the workflow is stable enough that it won't need constant rebuilding. It doesn't earn its cost when volume is low, when the process changes every few weeks, or when a spreadsheet and a person checking it once a day already handles the load fine. A workflow that fires twice a month doesn't need a paid Zapier tier; a free tier or a manual checklist is the right tool.

It's also worth being honest about maintenance. Every integration is a dependency: when a connected app changes its interface or API, the workflow can break silently until someone notices a step didn't run. Factor in someone's time to monitor and fix that, or the cheap option on paper gets expensive in practice. If you're not sure which category your situation falls into, our guide on AI automation for small business walks through how to identify a good first workflow before you spend anything, and Async Automations starts every engagement with a free audit through AI Automation services to scope that out before committing to a build.