"How much does an AI chatbot cost" is one of the first questions we get from businesses looking into this, and it's a fair one because the answer varies wildly depending on what you actually mean by chatbot. A scripted widget that answers five FAQ questions costs almost nothing. A chatbot that pulls live order data, talks to your CRM, and handles a real support conversation is a different project entirely. Here's what actually drives the cost and what you should expect at each level, so you can ask a vendor better questions instead of anchoring on a single number that may not apply to your situation.

The three tiers of "AI chatbot"

At the low end is a rules-based bot with some AI polish on top, good for answering a fixed set of common questions with natural-sounding replies. In the middle is a chatbot connected to your knowledge base through retrieval, so it can answer open-ended questions about your actual product or policies instead of a script. At the high end is a chatbot with tool access, meaning it can look up an order, check account status, or escalate to a human with full context, not just talk about those things in the abstract. Most businesses asking about cost are picturing the middle or top tier without realizing the bottom tier exists and might be enough.

What actually drives the price

Setup cost is driven by how much your existing systems need to be connected. A chatbot that only needs to read from a document library is cheaper to build than one that needs live, secure access to your order management, billing, or CRM systems. Ongoing cost is driven mostly by usage-based model fees, which scale with conversation volume, plus any hosting for the retrieval or database layer. A chatbot handling a few hundred conversations a month costs meaningfully less to run than one handling tens of thousands, and that scaling should be part of the conversation before you build, not after.

The number and age of the systems you're connecting to matters too. A modern platform with a well-documented API is a straightforward integration. An older internal system with no API, or one that was never designed to be queried by anything outside its own interface, often needs custom middleware built just to give the chatbot something to talk to. That extra layer adds both setup cost and an ongoing maintenance responsibility that's easy to underestimate at the proposal stage.

  • Setup cost: driven by number and complexity of system integrations.
  • Running cost: driven by conversation volume and model usage.
  • Maintenance cost: driven by how often your knowledge base or policies change.
  • Integration cost: higher for older systems without a modern API.

What a chatbot can realistically do well

A well-built chatbot handles repetitive, well-defined questions reliably: order status, return policy, account basics, appointment scheduling, troubleshooting steps for common issues. It's good at working around the clock without wait times and at handling volume spikes that would otherwise overwhelm a support team. It's not good at handling genuinely novel situations, emotionally charged complaints that need a human's judgment, or edge cases your team hasn't documented anywhere. A chatbot is only as good as the information and rules it's given access to.

What "what to expect" actually means

Expect a real chatbot project to take weeks, not days, if it needs system integrations. Expect it to get things wrong occasionally in the first few weeks live, and expect to need a process for reviewing those misses and correcting the knowledge base or rules behind them. Expect that a chatbot without a clear escalation path to a human will frustrate customers more than it helps, since nothing is worse than being stuck in a loop with a bot that can't solve your problem and won't let you talk to a person.

Also expect that a chatbot's first few weeks live will teach you more about what customers actually ask than any planning document did beforehand. Real conversations surface question phrasing, edge cases, and gaps in your knowledge base that nobody anticipated, which is normal and part of why an early review process matters more than trying to perfect the bot's knowledge before launch.

  • Define escalation paths before launch, not after complaints start.
  • Budget time for reviewing real conversations in the first month.
  • Start with a narrower scope you can actually maintain well.

Scoping your first build

The businesses that get the most value out of a first chatbot pick one specific area, like order status and returns, get it working reliably, and expand from there once it's proven. Trying to launch a chatbot that answers everything about the business on day one usually means it answers most things poorly. Narrow scope, done well, builds the trust and internal buy-in to expand the chatbot's responsibilities later.

A useful exercise before you request quotes is pulling a month of real support tickets and sorting them by topic. The categories with the highest volume and the most repetitive, well-documented answers are almost always the right starting scope, and having that breakdown in hand also makes it much easier for a vendor to give you an accurate estimate instead of a rough guess based on a general description of your business.

If you're scoping your first AI chatbot and want a realistic cost estimate instead of a sales number, our AI automation team can help you figure out what's actually worth building first.