AI chatbot development services will quote you one of two prices: a custom build from a dev shop, which runs anywhere from a few thousand dollars to well over $150,000, or a platform "AI employee" you rent for $50 to $500 a month with per-conversation fees on top. Both numbers are real. Both are also mostly markup, because the thing you think you're buying, the intelligence, is a public commodity that costs a fraction of a cent per conversation.
Here's what you're actually paying for, and why the real asset isn't the AI at all. It's the wiring, and who holds the deed to it.
What do AI chatbot development services actually charge?
The range is enormous, and it's worth seeing laid out, because the spread tells the story. From the published cost guides (all written, worth noting, by the dev shops doing the quoting):
| What you're buying | Typical price |
|---|---|
| Basic rule-based bot | $2,000 to $40,000 |
| AI bot with natural-language understanding | $75,000 to $150,000+ |
| Generative / agentic AI system | $80,000 to $250,000+ |
| Ongoing maintenance retainer | $5,000+ per month |
| Platform "AI employee" subscription | $50 to $500 per month |
| Per-conversation platform fee | $0.50 to $0.70 each |
Reputable shops publish these numbers openly, like Cleveroad's chatbot cost breakdown and MasterOfCode's pricing guide, where the $0.50-to-$0.70-per-conversation platform figure comes from. The numbers aren't a scam. What they leave out is what makes them possible.
The model is the cheap part
Here's the fact that reframes everything: the AI model itself is a commodity with a public, per-token price. A capable model like GPT-4o-mini runs about $0.15 per million input tokens and $0.60 per million output, listed right on OpenRouter's pricing page alongside the exact per-token rates and dozens of models that cost even less.
Do the division. A typical support conversation is roughly five thousand words of input and a thousand of output. At those rates, that conversation costs about one-seventh of a cent. A platform charging you $0.50 to $0.70 for the same conversation is marking up the underlying API call by three hundred to five hundred times.
The AI didn't get more expensive. The billing wrapper did. And that $75,000 custom build? You're not paying for intelligence, because the intelligence is a public API any developer can call. You're paying for roughly fourteen hundred hours of glue code, plus a permanent maintenance retainer to the one shop that kept the keys.
So what are you really paying for?
The wiring. And that's fine, because the wiring is real work: connecting the model to your CRM, your calendar, your knowledge base, your phone system, giving it a persona, guardrails, and somewhere to put the leads it captures. That's the part with genuine value.
The problem isn't that the wiring costs money. The problem is who owns it when the invoice is paid. With a dev shop, they keep the deed. You re-hire them for every change, every new integration, every prompt tweak, forever. With a platform, you never touch the API key or the config at all. You rent a seat on their stack, metered per conversation, while your actual token cost drifts down and your bill drifts up.
Either way, you bought a car and someone else kept the title.
Build vs. buy vs. own
The industry frames this as two choices, build custom or rent a platform, and carefully leaves out the third:
- Build custom ($75K+ and a retainer): you get ownership on paper, but the shop holds the working knowledge and you pay them for every change.
- Rent a platform ($50 to $500/mo plus per-chat fees): fast to start, expensive to stay, and you own nothing. Same 300x markup, monthly, forever.
- Own the wiring (the option nobody sells you): the chatbot lives inside software you hold title to, wired straight to the AI providers at their actual per-token rate. You pay the wholesaler, not a reseller's markup, and the config is yours to change.
That third option is the Allodra model. The AI chat and voice ship inside a CRM you own, wired direct to the providers, so the leads it captures land in your pipeline and the only per-conversation cost is the fraction of a cent the model actually charges. It's the same move we broke down for phone answering in the white-label voice agent markup story: same AI, no middleman, the markup goes in your pocket instead of a platform's.
The bottom line
AI chatbot development services aren't ripping you off, exactly. They're selling you wiring at a premium and letting you assume you're buying intelligence. The intelligence is a commodity you can call directly for pennies. The wiring is the only real asset, and the whole game is whether you own it or rent it.
Before you sign a $75,000 build quote or a per-conversation platform contract, understand what the model actually costs and decide whether you want to own that wiring outright. Run the math on your own volume in the cost analyzer. The AI is the cheap part. Make sure you're not renting it at five hundred times cost.