Key takeaways
- There are three ways to get an AI chatbot: a SaaS platform ($39–$500/month), a custom build ($12,000–$70,000+ upfront), or a hybrid route using automation tools like n8n plus LLM APIs (a few hundred a month). They are completely different products.
- The quote you get is roughly a third of the real cost. Ongoing model usage fees, maintenance retainers, and seat minimums are where the money actually goes.
- At low volume, SaaS wins. Somewhere around 2,000–3,000 conversations a month, a custom build starts paying for itself.
- Always ask the vendor to define a "resolved conversation" in writing — per-resolution pricing games hide there.
- Most small businesses should start with a narrow pilot, measure the resolution rate for 60 days, then decide. Not the other way around.
You've probably been quoted everything from $29 a month to $60,000 for "an AI chatbot." Here's the thing nobody tells you up front: both quotes can be real. They are just pricing completely different things — a rented tool, a built asset, and everything in between.
We build AI chatbots and automation for businesses, so this isn't a buyer's guide written by someone selling you a platform. It's what we'd tell you if you sat down with us and asked what this actually costs in 2026.
The three routes, side by side
Every AI chatbot price you've ever seen falls into one of three buckets. Compare them wrong and you'll pick the expensive option by accident.
| Route | Upfront cost | Ongoing cost | Time to live | Best for |
|---|---|---|---|---|
| SaaS platform (Tidio, Chatbase, Intercom Fin, Crisp) | $0–$2,000 setup | $39–$500+/month at small-business volume | Days to 2 weeks | Standard use cases: FAQs, lead capture, basic support |
| Custom build (agency or developer builds you one) | $12,000–$70,000+ | $500–$3,000/month in model costs, hosting, and maintenance | 4–10 weeks | Deep integrations, sensitive data, high volume, full control |
| Hybrid (n8n or similar + LLM APIs, built for you) | $2,000–$8,000 setup | $100–$400/month in model and hosting costs | 1–3 weeks | One or two tight workflows; you want to own it without enterprise budgets |
The rest of this post prices each route honestly, including the costs nobody puts in the quote.
What SaaS chatbot platforms actually cost in 2026
The SaaS market has settled into a few pricing models, and understanding them matters more than any single price. Per a September 2026 pricing comparison of 14 platforms by Alhena AI, published AI rates in 2026 run from about $0.20 to $2.00 per AI-handled conversation, on top of any helpdesk seats you already pay for. At a realistic 70% resolution rate, that works out to roughly:
- A small store doing ~300 AI conversations a month: $80 to $350 a month
- A growing brand at ~1,000 a month: $300 to $1,400 a month
- A mid-market team at ~3,000 a month: $1,000 to $4,200 a month
The four pricing models — and which one favors you
| Model | How it bills you | Example platforms | Watch out for |
|---|---|---|---|
| Per resolution | You pay only when the AI closes a conversation | Intercom Fin ($0.99/resolution), HubSpot ($0.50), Zendesk ($1.50–$2.00) | Every vendor defines "resolved" differently — ask for it in writing |
| Per conversation / bundles | A plan buys a block of AI conversations | Tidio ($39/mo for 50), ChatBot.com, Crisp | Unused conversations usually expire; overage jumps are steep |
| Credit-based | You burn credits per message or action | Chatbase, Crisp AI credits | Hard to forecast until you've run it for a month |
| Per seat + AI add-on | Agent seats, AI priced on top | Freshdesk Freddy ($0.49/session + seats), Zendesk, Kustomer | Seat minimums (Kustomer's is 8) and $29–$50/month copilot fees per agent |
The headline price is the least informative number on a chatbot pricing page. The two numbers that matter: what you pay per conversation the AI actually closes, and what the bill does when your volume doubles in your busy season.
A concrete example of the per-resolution trap: one builder comparison found that at 500 monthly chats, Intercom's $0.99-per-resolution Fin agent lands around $174–$320/month once you add the base plan — roughly triple Tidio's flat $79 for comparable AI capability. Flat pricing looks boring. It's often cheaper.
What a custom build actually costs
A custom chatbot means someone builds the whole thing for you: it reads your documents, talks to your booking system or CRM, answers in your brand's voice, and lives wherever you want it to. No per-conversation tax, no vendor dashboard holding your configuration hostage.
Build costs in 2026, from developer-published breakdowns:
| Tier | Build cost | Timeline | What you get |
|---|---|---|---|
| Pilot / MVP | $5,000–$12,000 | 2–4 weeks | One tight use case, your documents, one channel, basic analytics |
| Standard custom bot | $12,000–$25,000 | 4–8 weeks | Multi-turn conversations, one or two system integrations, trained on your real content |
| Serious business bot | $25,000–$70,000 | 8–14 weeks | CRM/booking integrations, multi-channel, admin dashboard, proper testing |
| Enterprise / multi-agent | $90,000–$260,000+ | 3–6 months | Multiple specialized agents, core-system integrations, compliance and audit |
Those bands are corroborated from several directions: developers quote $8,000–$20,000 for an API-based chatbot build, one software consultancy's team-math breakdown puts a custom LLM chatbot at $12,000–$70,000, and a separate agency guide lands the median mid-complexity build at $75,000–$120,000. The spread isn't noise — it reflects how much of your business gets wired in.
What actually drives the build cost
The AI model is rarely the expensive part. Here's where your budget really goes:
- Your data preparation. If your knowledge base is scattered across PDFs, Google Docs, and old email threads, someone has to clean it up before the bot can learn it. Budget 5–15 hours of prep work, minimum.
- Integrations. Connecting the bot to your specific CRM, booking system, or inventory adds 20–40% to the build estimate. This is the line item lowball quotes skip entirely — then charge you for it as a "change request" later.
- Channels. Website chat is one thing. Adding WhatsApp, Instagram DMs, and email each adds configuration, testing, and sometimes per-channel fees.
- The tuning week. An honest builder will tell you the setup isn't done at launch. The version trained on your real content, tested against real customer questions, and connected to your systems typically needs two to four weeks of setup including about a week of just watching real conversations and fixing what's wrong.
If a quote doesn't mention data preparation, it's not a quote. It's a wish. That's the line item that separates builders who've shipped these from sales teams who haven't.
The monthly costs nobody puts in the quote
This is the section that saves people money. Whatever route you take, budget for these:
| Hidden cost | What it looks like in practice | Question to ask |
|---|---|---|
| LLM inference fees | A custom bot processing ~100,000 conversations a month spends roughly $800/month on model API fees — or around $100/month with optimized open-source models. SaaS vendors bake this into your per-conversation price. | "What's my model cost per 1,000 conversations at my volume?" |
| Maintenance retainer | AI systems drift. Prompts need tuning, edge cases need handling, models get updated. A realistic retainer is $500–$2,000/month — or about 17–30% of the build cost annually. | "What's included after launch day, and what costs extra?" |
| Knowledge upkeep | Your prices change, policies change, products change. Someone updates the bot's brain. That's your team's hours, every month. | "How many hours a month do your customers spend maintaining the bot?" |
| Double-billing | Some platforms count one conversation twice — e.g., an AI interaction that also consumes a helpdesk ticket. Seat minimums (like Kustomer's 8-seat floor) mean you pay for agents you don't have. | "Does an AI conversation consume anything else I pay for?" |
| Mid-contract changes | Vendors raise prices and start auto-billing overage mid-contract. One major platform began auto-billing overage in January 2026. | "Can my rates change before renewal? Is there a usage cap I can set in writing?" |
| Vendor risk | Platforms get acquired, pivot their product, or shut down. If your bot's brain lives on someone else's dashboard, that's your problem. | "Can I export my full configuration and conversation data?" |
The honest math: a custom bot that costs $15,000 to build will realistically cost you $6,000–$12,000 a year to run well. A $99/month SaaS bot at real volume often costs $200–$400/month once overage, seats, and add-ons pile up. Neither is a trick — but only one of them usually gets quoted upfront.
Which route should YOU take?
Here's the decision framework we'd actually use with a client:
| Your situation | Go with | Why |
|---|---|---|
| Under 300 conversations/month, simple FAQs | SaaS | Cheapest, fastest. A custom build can't pay for itself at this volume. |
| You need it live this week | SaaS | Custom takes 4+ weeks minimum. No way around it. |
| Sensitive customer data or regulated industry | Custom | You control where data lives and who sees it. Worth the premium. |
| Over ~2,000–3,000 conversations/month | Custom or hybrid | Per-conversation pricing stops being cute at volume; flat operational costs plateau. |
| You already pay for HubSpot/Zendesk/Intercom | Their AI add-on | The data's already there. Don't buy a second brain for your business. |
| One or two specific workflows, modest budget | Hybrid (n8n + LLM) | You own the automation, pay only model usage. Our favorite starting point for small businesses. |
| You're not sure the bot will work for your customers | Free tier pilot | Chatbase and Botpress have free tiers. Prove the use case before spending anything. |
Six red flags in chatbot quotes
We've reviewed enough of other people's quotes to know the patterns. Walk away — or at least push back hard — if you see these:
- No line item for data preparation. The bot is only as good as what it learns from. A quote that skips this is either going to surprise you later or deliver a bot that makes things up.
- "AI included" with no usage budget. Every AI answer costs fractions of a cent in model fees. A quote with no token/inference budget either doesn't understand the tech or plans to bill you for it later.
- You don't own the code. Some agencies build on their internal frameworks and charge a premium — then you discover you can't take your bot (or your data) anywhere else. Insist on code ownership.
- An oversized team for a v1. A senior AI engineer, a data engineer, and a solid full-stack developer can take most pilots from concept to production. You don't need twelve people for version one.
- No accuracy testing plan. If there's no phase for testing the bot against real customer questions before launch, the testing phase is your customers. That's not a plan, that's a gamble.
- Per-resolution billing without a written definition of "resolved." One vendor counts a customer who simply leaves without asking for more help as "resolved." That definition is doing a lot of work in their pricing.
Our honest recommendation
Start narrow. Pick the one job that eats the most of your team's time — usually answering the same 20 questions, or qualifying inbound leads — and pilot a bot for exactly that. Measure the resolution rate for 60 days. Then decide whether to expand, switch platforms, or build custom.
For most small businesses, the right order is: prove the use case on an affordable SaaS or hybrid setup first, then let the math tell you whether a custom build makes sense. The businesses that burn money on chatbots are the ones that do it backwards — signing a platform contract or commissioning a build before knowing whether their customers will even talk to the thing.
And if you're somewhere between "I need something real" and "I can't spend $30,000"? That's exactly the gap the hybrid route fills. An automation built on n8n with a good LLM behind it, wired into your actual booking system or CRM, owned by you, running for a few hundred a month. We build these regularly — it's most of our AI integration work these days.
Want a straight answer on what a chatbot would cost for your business, not a pricing page? Get a free site audit and we'll tell you whether a bot would even pay for itself — or book a call and we'll scope it honestly, including the option where we tell you not to build one.
