What Does AI Actually Cost a Small Business Per Month?
Real numbers, not "affordable." A worked example, and the hidden costs people don't budget for.
Ask most AI consultants what a project costs and you'll get "it depends" or "affordable" — technically true, practically useless. Here's what things actually run, so you can budget before you talk to anyone.
Two ways to pay, and it changes the math a lot
A flat subscription — ChatGPT Plus, Claude Pro, roughly $20/month per person as of this writing — makes sense when one or two people are doing varied, hands-on work with the tool every day. Pay-as-you-go API access — billed per unit of text processed, usually a fraction of a cent per interaction — makes sense when a workflow runs on a schedule or trigger rather than being driven by someone clicking around. At low-to-moderate volume, the API route is often dramatically cheaper than a subscription; it just takes a bit more setup.
Most single-person, hands-on workflows (drafting quotes, writing specials copy, one person doing it by hand each time) end up cheaper on a subscription. Most automated workflows (something that fires on a schedule or a trigger) end up cheaper on the API.
A worked example
Say a business handles about 200 customer emails a month and wants AI to draft a first-pass reply to each one for a person to review before sending. Using a mid-tier model at API pricing, a typical email exchange runs somewhere around 1,500–2,500 tokens of processing per email. At current published rates for a capable mid-tier model, that puts the raw model cost at roughly $4–6 a month for the full 200 emails — even accounting for some replies needing more back-and-forth than others.
The real monthly cost of that workflow almost never lives in the model usage. It's whatever tool actually gets the draft into the inbox — sometimes free (a shared inbox with built-in AI drafting), sometimes $10–30/month for a connector platform like Zapier or Make.
Typical ranges by workflow type
- Hand-drafted quotes, specials copy, review replies (one person, a few times a day): $0–20/month, one subscription seat.
- Automated email drafting at moderate volume (a couple hundred a month): roughly $3–6/month in raw model usage, plus whatever connects it to your inbox.
- A simple customer-facing chatbot, low volume (under 500 conversations/month): $15–50/month, API plus basic hosting.
- A multi-step automated system (several connected workflows): $50–200/month, scaling with volume.
Full table with more scenarios and specifics: what AI actually costs to run.
The costs people don't budget for
- Per-seat multiplication. $20/month sounds small until four employees each need their own subscription. Sometimes one shared workflow beats four individual seats.
- Review time. Every drafted-not-automated workflow has a real, ongoing labor cost to check the output. It should be less time than writing from scratch — if it isn't, the workflow needs fixing.
- Usage spikes. Pay-as-you-go costs scale with volume. Set a budget ceiling and an alert, don't get surprised by it.
- Tools nobody opens after week one. The most common wasted AI spend, by a wide margin. Start with one workflow tied to a task someone already does daily, not a subscription bought on optimism.
None of this replaces getting a number for your specific business — that's exactly what the AI Readiness Audit works out, with your actual volume and tools. But now you know roughly what range to expect before that conversation even starts.