Cost guide
How much does an AI chatbot cost?
Last reviewed , published by Gjethe
The short answer
An AI chatbot's cost depends on what it needs to know (your documents, website or database), which systems it connects to, which channels it runs on, how many languages it supports, the safeguards required and ongoing AI usage fees. A simple FAQ assistant on your own content costs far less than an agent that books appointments, checks orders and hands over to staff. Every project gets a fixed quote after a free consultation.
Modern AI chatbots are built on large language models from providers such as OpenAI, Anthropic or Google, connected to your own information. The model itself is rented per use; most of the build cost is in preparing knowledge, integrating systems, designing conversations and making the bot reliable. This guide explains each part.
Types of AI chatbot
| Type | What it does | Relative complexity |
|---|---|---|
| FAQ assistant | Answers questions from your website and documents | Low |
| Support assistant | Answers, collects details and hands over to staff | Medium |
| Transactional agent | Checks orders, books appointments, updates records | High |
| Internal assistant | Helps staff search policies, documents and data | Medium to high |
Which factors change the price most?
Knowledge sources
Clean website content is easy; scattered PDFs, spreadsheets and databases need preparation.
Integrations
Connecting bookings, orders, CRMs or helpdesks lets the bot act, and adds work.
Channels
Website, WhatsApp, Messenger, Slack or Teams each need set-up.
Languages
Testing answers in several languages adds effort.
Guardrails
Limits, escalation rules and review for sensitive topics.
Analytics
Dashboards showing questions, gaps and outcomes.
Usage costs: paying for the AI model
AI providers charge per use, typically based on the amount of text processed. Costs scale with the number of conversations, their length and the model chosen. Smaller, faster models are cheaper and often good enough for straightforward questions.
We design chatbots to keep usage efficient: retrieving only relevant information, limiting conversation length where sensible and choosing the right model for each task. Usage is billed to your own provider account so costs are transparent.
Reliability is where the work goes
Getting a chatbot to answer is easy; getting it to answer correctly, admit when it doesn't know and hand over gracefully is the real work. That means curating knowledge, writing instructions, testing many real questions and adding rules for sensitive topics.
After launch, reviewing conversations shows where answers are weak or knowledge is missing, and improving those areas is part of ongoing maintenance.
Ongoing costs
| Cost | Notes |
|---|---|
| AI model usage | Scales with conversations; paid to the AI provider |
| Hosting and vector database | Stores and searches your knowledge |
| Channel fees | Some messaging platforms charge per conversation |
| Maintenance | Updating knowledge, improving answers, monitoring |
Prepare before asking for quotes
- The top questions customers or staff ask
- Where the answers currently live
- Actions the bot should be able to take
- Channels and languages needed
- Topics the bot must never answer
- Who takes over when the bot can't help
Example: from FAQ bot to support agent
A first version might answer questions about opening hours, prices and policies using your website and help articles, and offer a contact form when it can't help. A second version adds order lookups through your store's API and hands complex cases to a live agent with the conversation history. A third version takes actions, such as changing bookings, with confirmation steps.
Starting with the first version shows quickly which questions customers ask and how well the bot answers, which guides investment in the next steps.
Questions that shape the estimate
| Question | Why it matters |
|---|---|
| How many knowledge sources are there? | Each source needs preparation and updating |
| Must the bot take actions? | Actions need integrations and safeguards |
| Which channels? | Each channel has its own set-up and rules |
| How sensitive is the data? | Affects provider choice, hosting and controls |
| Who reviews performance? | Ongoing improvement needs an owner |
Measuring whether it's worth it
Track how many conversations the bot resolves without handover, customer satisfaction ratings, common unanswered questions and the effect on support volume. These numbers show whether the chatbot pays for itself and where to improve it.
Reviewing a sample of conversations regularly is the most reliable way to catch wrong or unhelpful answers.
Choosing the right AI model
Larger models handle complex questions and reasoning better but cost more per conversation. Smaller models are faster and cheaper and often good enough for straightforward questions. Some chatbots route simple questions to a small model and difficult ones to a larger model.
Testing several models on your real questions is the best way to find the right balance of quality and cost.
Preparing your knowledge
Chatbots are only as good as the information they draw on. Outdated, contradictory or incomplete content leads to poor answers. Reviewing help articles, policies and FAQs before launch often improves both the bot and your website.
Assign someone to keep knowledge current after launch, especially prices and policies.
Launching in stages
Many teams launch a chatbot first to internal staff or a small share of website visitors. That reveals weak answers safely before everyone sees them.
Once quality is proven, the bot can be rolled out widely with confidence.
Frequently asked questions
Do I pay for the AI model separately?
Yes. AI providers charge per use. We set it up in your account so you see and control the costs.
Will the chatbot make things up?
Language models can produce wrong answers. We reduce this by grounding answers in your content, instructing the bot to say when it doesn't know and testing extensively.
Can it work on WhatsApp?
Yes, through the WhatsApp Business Platform, which has its own conversation-based pricing.
Is customer data safe?
We minimise data sent to AI providers, use business plans with appropriate data terms and follow GDPR principles. Review provider terms for your use case.
Can it hand over to a person?
Yes. Handover to live chat, email or a helpdesk ticket is a standard part of good chatbot design.
Can it speak Albanian and German?
Yes. Major models handle many languages; we test answers in each language you need.
Is a chatbot cheaper than a support agent?
It can handle routine questions at scale, but works best alongside people rather than replacing them for complex or sensitive issues.
Can I get a fixed price?
Yes, for the build. AI usage fees vary with volume and are paid to the provider.
Can the chatbot learn from conversations automatically?
Not by itself in most designs. Improvements come from reviewing conversations and updating knowledge and instructions, which keeps changes controlled.
Can the chatbot show sources for its answers?
Yes. Linking to the page an answer came from helps customers verify it and builds trust.
How long are conversations stored?
That's a design choice. We agree retention periods with you and configure storage accordingly, keeping only what's needed.
Related services
AI Chatbot Development
Chatbots that answer from your own content.
RAG Systems
AI that searches your documents and cites them.
AI Agent Development
AI that takes actions in your tools, with guardrails.
AI API Integration
Language, vision and speech models in your software.
AI Automation
Make, Zapier, n8n and Python workflows with AI steps.
Talk it through with us.
Tell us what you need. You get an honest assessment, a clear plan and a written quote. The first consultation is always free.
- What happens next
- 1You tell us what you need
- 2We aim to reply within 24 hours with questions
- 3You get a written quote for your project