Key takeaways
- AI chatbot cost depends mostly on scope: the number of use cases, channels, knowledge sources and connected business systems.
- Estimated builds range from a few thousand dollars for a simple FAQ assistant to $250,000 or more for an AI agent that takes actions.
- Budget for ongoing costs such as model usage, hosting, monitoring, content upkeep and maintenance, not just the initial build.
- Off-the-shelf platforms suit standard support needs, while custom builds make sense for deep integrations, actions inside your systems or high volume.
- Starting with one use case, a proof of concept, clear guardrails and a human hand-over keeps chatbot spending predictable.
If you ask five vendors what an AI chatbot costs, you will likely get five very different answers. That is not evasiveness. A "chatbot" can mean a simple assistant that answers questions from your help center, or an AI agent that looks up orders, issues refunds and updates your CRM. The price follows the scope.
This guide explains what drives cost, gives broad market ranges for common tiers, separates one-time build costs from ongoing running costs, and shows how to get an estimate you can actually budget against.
The short answer: typical cost ranges in 2026
For US businesses, AI chatbot projects tend to fall into three broad bands. These are estimates based on typical market rates and vary widely by scope, vendor and region. They are not quotes.
- Simple FAQ assistant: a few thousand dollars to roughly $20,000 to set up, plus a monthly platform or usage fee.
- Integrated support assistant: roughly $20,000 to $80,000 to build, plus ongoing usage, hosting and maintenance.
- AI agent that takes actions: roughly $60,000 to $250,000 or more, depending on how many systems it touches and how much risk each action carries.
Ongoing costs can range from under a hundred dollars a month for a low-traffic assistant to several thousand dollars a month for a high-volume, integrated system. The sections below explain where those numbers come from.
What drives the cost of an AI chatbot
Six factors account for most of the difference between a small project and a large one.
Scope and use cases
The number of jobs the bot must do is the biggest driver. Answering shipping questions is one use case. Answering shipping questions, booking appointments, qualifying sales leads and handling returns is four, each with its own content, testing and edge cases. Every added use case adds design, build and testing time.
Channels
A widget on your website is the simplest channel. Adding SMS, WhatsApp, Microsoft Teams, Slack, a mobile app or voice means more integration work, different message formats and separate testing for each. Voice is usually the most expensive because it adds speech-to-text, latency concerns and call routing.
Knowledge sources and RAG
Most business chatbots answer from your own content: help articles, policies, product data, contracts or internal wikis. The common approach is retrieval-augmented generation (RAG), where the bot searches your documents and uses what it finds to write an answer. Cost rises with:
- The number and variety of sources (PDFs, SharePoint, a database, a ticketing system)
- How messy or outdated the content is (cleanup is often underestimated)
- Whether different users should see different content based on permissions
- How often content changes and must be re-indexed
Integrations and actions
Reading data is cheaper than changing it. A bot that looks up an order status needs a read-only connection. A bot that cancels orders, issues credits or updates customer records needs write access, confirmation steps, error handling, audit logs and much more testing. Each system you connect (CRM, ERP, helpdesk, scheduling, payments) adds work, especially older systems without clean APIs.
Security and compliance
If the bot handles personal, health, financial or payment data, expect extra work for access controls, data retention rules, logging, vendor reviews and possibly a security assessment. Industries such as healthcare, finance and insurance usually need more of this. Your legal and compliance team should confirm which requirements apply to you; this guide is not legal advice.
Languages
Modern language models handle many languages well, but each supported language still needs testing, translated source content where possible, and review by someone fluent. Supporting English and Spanish is meaningfully more work than English alone.
Three common tiers, from FAQ bot to AI agent
The table below compares the three tiers side by side. Cost figures are broad estimates for planning only.
| Simple FAQ assistant | Integrated support assistant | AI agent that takes actions | |
|---|---|---|---|
| What it does | Answers common questions from a fixed set of content | Answers from multiple knowledge sources and looks up customer or order data | Completes tasks such as refunds, bookings or record updates |
| Typical channels | Website widget | Website, plus one or two more (email, chat apps) | Multiple channels, sometimes voice |
| Integrations | None or very few | Read access to helpdesk, CRM or order system | Read and write access to several business systems |
| Human hand-over | Link or form | Live hand-over to an agent with conversation history | Approval steps and escalation rules for risky actions |
| Typical timeline | Days to a few weeks | 6 to 12 weeks | 3 to 6 months or more |
| Estimated build cost (USD) | A few thousand to ~$20,000 | ~$20,000 to ~$80,000 | ~$60,000 to ~$250,000+ |
The jump from tier two to tier three is mostly about risk. Once a chatbot can change data or spend money, you need guardrails, approval flows, detailed logging and far more testing.
One-time build costs vs ongoing running costs
Many budgets focus on the build and underestimate what it takes to keep the bot running well. Plan for both.
One-time costs
- Discovery and design: defining use cases, conversation flows and success measures
- Content preparation: cleaning, organizing and filling gaps in your knowledge base
- Development: the assistant itself, the retrieval setup, integrations and the chat interface
- Testing: accuracy checks, edge cases, security testing and user acceptance testing
- Launch and training for the team that will own it
Ongoing costs
- Model usage. Most large language model providers charge by usage, typically per token (roughly, per chunk of text processed). Cost depends on traffic, conversation length, how much document text is sent with each question and which model you choose. Check each provider's current pricing page, since rates and model options change often.
- Platform or hosting fees. Off-the-shelf platforms usually charge per seat, per conversation or per resolution. Custom builds pay for cloud hosting, databases and search indexes.
- Monitoring and quality review. Someone should regularly read conversations, track wrong or unhelpful answers and fix the causes.
- Content upkeep. When prices, policies or products change, the knowledge base must change too, or the bot will give outdated answers.
- Maintenance. Model updates, API changes in connected systems, security patches and small improvements. A common rule of thumb is to budget 15 to 25 percent of the original build cost per year, though actual needs vary.
Build vs buy: chatbot platforms or a custom solution
You do not always need a custom build. The right choice depends on how specific your needs are.
When an off-the-shelf platform makes sense
Many helpdesk, CRM and website platforms now include AI assistants, and dedicated chatbot platforms exist as well. They are usually the fastest and cheapest way to start when:
- Your use case is standard customer support or lead capture
- Your content already lives in the platform's help center
- You do not need deep integration with internal or older systems
- You are comfortable with the vendor's data handling terms
Pricing models vary: per agent seat per month, per conversation, per resolved conversation, or tiered plans. Read the fine print on usage limits and overage charges, and check the vendor's current pricing page.
When a custom solution makes sense
A custom or semi-custom build is usually worth it when:
- The bot must work with internal systems, proprietary data or complex permissions
- It needs to take actions inside your own software
- Per-conversation platform fees would become expensive at your volume
- You need control over where data is stored, which model is used and how answers are logged
Many companies land in the middle: a platform or cloud AI service for the core, with custom integrations and retrieval built around it. Low-code tools such as Microsoft Copilot Studio can also be a practical option for organizations already on Microsoft 365.
How to keep chatbot costs under control
The most expensive chatbot projects are usually the ones that tried to do everything at once. A few habits keep spending predictable.
- Start with one use case. Pick a high-volume, low-risk question type, such as order status or policy questions, and do it well before expanding.
- Run a proof of concept first. A short AI strategy and proof of concept phase tests answer quality on your real content before you commit to a full build.
- Set guardrails early. Limit topics the bot will discuss, require it to cite sources, and block actions it should never take. Guardrails reduce the risk of costly mistakes and rework.
- Design the human hand-over. A clean path to a person (with human in the loop review for sensitive actions) means the bot does not need to handle every rare case, which keeps scope smaller.
- Choose the model to fit the task. Smaller, cheaper models often handle routine questions well. Reserve larger models for harder requests.
- Measure from day one. Track resolution rate, hand-over rate and wrong answers so you can see whether added spend is paying off.
How to get an accurate estimate
Vague requests produce vague quotes. Before you talk to vendors, prepare a short brief covering:
- Use cases: the top three to five questions or tasks, ranked by volume
- Volume: rough monthly conversations or tickets today
- Channels: where customers or employees will use the bot
- Content: where your knowledge lives and how current it is
- Systems: which tools the bot must read from or write to, and whether they have APIs
- Data sensitivity: what personal or regulated data could appear in conversations
- Success measures: what "working" means (for example, fewer tickets or faster responses)
- Ownership: who will maintain content and review conversations after launch
Ask each vendor to separate one-time costs from monthly running costs, to state their assumptions about usage, and to explain what is excluded. Comparing proposals on those terms is far more useful than comparing headline prices.
Next steps
If you are early in planning, start by listing your top use cases and the systems involved. That alone will tell you which tier you are likely in.
When you want a second opinion, a short conversation with an experienced partner can help you check scope, compare build and buy options, and size a sensible first phase. You can review Invictus Hub's AI agents and chatbots work or contact us to talk it through. There is no obligation, and a clear brief will help with any vendor you choose.
