For a small business in 2026, AI automation typically costs $2,000–$8,000 for a single scoped automation, $5,000–$15,000 for a customer-facing assistant done properly, and $15,000–$50,000+ for a multi-system build — plus monthly run costs of roughly $50–$500 in API usage and tooling. Those are the market’s real numbers, stated plainly; the rest of this page is what’s inside each band, how vendors quote, and when the right amount to spend is zero. Our quote for your workflow arrives in writing the same business day.
The price bands, with what’s in them
| Build type | 2026 market range | What it includes | Typical payback logic |
|---|---|---|---|
| Single workflow automation (intake routing, data sync, reporting) | $2,000–$8,000 | Scoping, build on your stack (n8n/Make/Zapier or custom), testing against your real historical cases, documentation, handover in your accounts | Hours saved weekly × loaded cost — clears in months or we tell you not to build |
| Grounded customer assistant (site/intake chatbot) | $5,000–$15,000 | Knowledge-base construction from your real answers, grounding + escalation design, measured-accuracy testing, launch + tuning period | Recovered missed inquiries + response-time gains |
| Scoped agent / multi-system build | $15,000–$50,000+ | Everything above across systems, human-approval gates, monitoring, staff training | Only built when the assessment math clearly clears — this tier gets refused most often |
Where the wider market sits, for calibration: providers who publish numbers (rare) show small-business builds from around $2,000 at the scrappy end to five figures for complex agent work, with dev-shop custom builds quoted anywhere from $5,000 to $180,000+ — a spread that says less about the technology than about who’s quoting. The way to compare quotes is the next two sections.
What actually moves the price
- How many systems the automation touches. One tool talking to one tool is cheap; five systems with edge cases multiply testing, and testing is most of the honest cost.
- How consequential a mistake is. A wrong internal summary costs a shrug; a wrong customer-facing answer costs trust. Higher stakes buy more verification, and verification is labor.
- The state of your knowledge. If pricing and policies are documented, assistant builds go fast. If we’re extracting them from your head, that extraction is real (billable, worthwhile) work.
- Volume. Mostly affects run cost, not build cost — which is why the split in the next section matters more than any single quote number.
Build cost vs monthly run cost — the split nobody shows you
Every AI automation has two prices, and proposals that show only one are hiding your real total:
- The build — one-time: scoping, construction, testing, documentation. Our ranges above.
- The run — forever: API usage (what the model charges per use), tool subscriptions (your n8n/Make/Zapier tier), and any maintenance. For typical small-business volumes this lands around $50–$500/month — usually tens of dollars, not hundreds, at the volumes we build for.
Because every account and API key is yours (our standing guarantee — your keys, your accounts, your automations), you see run costs raw on your own bills, at cost. We mark up nothing on usage. A vendor who keeps the API account in their name has given themselves a margin you can’t audit — and a hostage when you leave.
AI app development cost — when your idea is a product, not an automation
Sometimes the thing being priced isn’t an internal automation but a small application — a client portal with AI inside, a quoting tool your customers use. That’s real software development, and it prices like it: markedly above automation builds, because you’re paying for interface, accounts, security, and upkeep, not just a workflow. Our honest guidance: if the AI part can be proven as an internal automation first (the $2,000–$8,000 band), prove it there before funding an app around it. Most “we need an AI app” ideas are an automation plus a form — and that’s good news for the budget.
How AI automation agencies price — decoding the quotes you’ll get
Three pricing models dominate the market, and knowing which you’re reading changes how you compare: project pricing (ours — a fixed build price, run costs at cost, optional maintenance), retainer pricing (“AI partner” subscriptions from a few hundred to several thousand monthly — fine when there’s a real monthly workload, padding when there isn’t; ask what happens in a month where nothing changes), and usage-marked-up pricing (the platform holds the API accounts and margins the tokens — the model to be wariest of, since the vendor profits from your volume and controls your keys). Any quote can be normalized by asking three numbers: build cost, monthly run cost in whose accounts, and the price of leaving.
When NOT to spend on AI automation
The cheapest build is the one you skip. Honest disqualifiers, from real assessments:
- The volume isn’t there. Ten inquiries a week don’t justify an intake agent; a shared inbox and a template do. Payback math below ~2 saved hours/week rarely clears.
- The process changes weekly. Automation hardens a process; if yours is still finding its shape, harden it later.
- The knowledge isn’t written down. A grounded assistant needs ground truth. If pricing and policies live in your head, a $0 week of writing beats a $5,000 build — and makes the eventual build better.
- The bottleneck is demand, not operations. If the calendar is empty, automation polishes an empty pipeline. Spend on getting found first; automate when volume hurts.
Our free AI-readiness assessment runs exactly this math on your answers — one email, same-business-day reply — including telling you “not yet” when that’s the answer.
How much does a chatbot cost?
Done properly — grounded in your real knowledge, escalating honestly, tested to measured accuracy — a small-business chatbot runs $5,000–$15,000 to build across the market, with modest monthly run costs. Done as a $49/month template widget, it costs $49/month and answers from nothing, which is how support bots end up inventing policies. The difference you’re paying for is the verification layer, and it’s the difference between an asset and a liability with a chat window.
A useful decomposition when comparing chatbot quotes: roughly speaking, the knowledge base is a third of the honest work (extracting and structuring what the bot may say), the grounding and escalation design is another third (what it must never improvise, when it hands off), and testing against real past conversations is the rest. A quote that’s all “integration” and no knowledge work is describing a widget, not an assistant.
Frequently asked questions
How much does AI agent development cost?
Scoped small-business agents — systems that act within defined limits, with human gates on consequential steps — run $15,000–$50,000+ across the honest end of the market, while dev-shop quotes for custom agent platforms range from $5,000 all the way to $180,000+; the spread mostly reflects scope and who’s quoting. Before accepting any number, make the vendor split build cost from run cost and show you what “done” means in measured accuracy.
Is AI automation worth it for a small business?
Worth it when repetitive, rule-describable hours exist and the math clears: hours saved × loaded hourly cost vs build + run cost, with payback in months. A $3,000 automation saving five hours a week of $40/hour work pays for itself fast; the same build saving forty minutes doesn’t. Run your own numbers — or email us for the free readiness assessment; it’s the same arithmetic.
Why do AI automation quotes vary so wildly?
Because “AI automation” spans a no-code connector afternoon and a custom engineering quarter, and because many quotes bundle unaudited margins — marked-up API usage, platform accounts held in the vendor’s name, retainers for “monitoring” that’s an email filter. Split every quote into build, run, and ownership terms and the variance mostly evaporates.
What does AI automation cost per month after the build?
Two recurring lines: API usage — billed by the model provider to your own account, typically tens of dollars monthly at small-business volumes, scaling with use — and tooling, meaning your automation platform’s tier (from free for self-hosted n8n to platform subscriptions at higher volumes). Optional third line: a maintenance arrangement, priced when you actually need one. Because everything bills in your accounts, the monthly total is never our claim to trust — it’s your own invoice to read.
Do I need a monthly retainer for AI automation?
Not with us — builds are fixed-price projects, run costs bill at cost in your own accounts, and a maintenance plan is optional, not required to get a quote. Automations do need occasional attention when your tools or processes change; that’s real work worth paying for when it happens, not a mandatory subscription.
Want your number instead of a range? Tell us the workflow that’s eating your week — a real quote inside these market bands, same business day. The full service is at AI consulting & automation.