What determines the cost of a chatbot?
Three choices move the cost of a business chatbot most: the type of bot (rule-based, an LLM chatbot answering from your content, or an agent that takes actions), the channel it runs on and the number of systems it connects to. The budget combines one-off build work with running costs: model usage, WhatsApp message fees, hosting, monitoring and content upkeep.

Topic
AI
Published
17 November 2026
Updated
17 November 2026
What does the cost of a chatbot consist of?
The cost of a chatbot project has two parts: the work done once to build it, and the costs that continue for as long as it runs. Judging by the build fee alone is misleading. In some solutions the real spend sits in monthly usage and content upkeep.
This article gives no figures. Model prices and channel fees change often, and the project fee follows the scope. What it does explain is which choice pulls the budget in which direction.
Rule-based bot, LLM chatbot or agent: how do the costs differ?
The single decision that moves the budget most is the type of bot. Three types mean three different cost structures.
| Topic | Rule-based bot | LLM chatbot answering from your content | Agent that takes actions |
|---|---|---|---|
| How it works | Fixed menus and a scenario tree | A language model finds the answer in your content and writes it | A language model also queries and acts in your systems |
| Main build work | Writing the scenarios | Preparing content, a test set, hand-off rules | Integrations, permissions, approval steps, security testing |
| Main running cost | Updating the scenario tree | Model usage, content upkeep | Model usage (several model calls per task), integration maintenance |
| What makes it grow | Many branches and exceptions | Scattered content, several languages, high volume | Number of systems, irreversible actions, audit requirements |
| Where it fits | Narrow, predictable flows | Information questions: delivery, return terms, opening hours | Reversible actions with clear rules |
A chatbot that answers from your content usually works with RAG. An AI agent that takes actions connects to your CRM, order or booking system. Our post on chatbot vs agent assist explains which model suits which kind of customer request.
What are the one-off costs?
Discovery and scope. Someone writes down which questions the bot answers, which it hands to a person and which channel it runs on.
Content preparation. FAQs, return and delivery policies and product information need to be current and in one place. If the content is scattered, this is the largest build item.
Integration. A bot that only gives information needs little integration. If it should report order status, stock or account details, it has to connect to your ecommerce platform, CRM or ERP. Companies operating in Türkiye often run local ERPs such as Logo, Mikro or Netsis. Where a system has no API, the job grows.
Channel setup. A website chat needs a widget on the site. WhatsApp needs a WhatsApp Business Platform account, a number, approved message templates and an opt-in flow. Our post on a WhatsApp Business AI assistant walks through the steps.
Hand-off and safety rules. You define when the bot hands over to a person, what it must not say and which actions it may not take without approval. These limits are built as guardrails. For agents, risky actions get a human-in-the-loop approval step.
Evaluation. The bot is tested with real customer questions, separately in each language. Customers write with abbreviations and typos, and in Turkish often without the Turkish characters. The bot has to understand those messages too.
Data-protection and legal review. Chat logs contain personal data. Under GDPR, and under KVKK, Türkiye’s personal data protection law (Law No. 6698), the review covers the privacy notice, retention, security measures and transfers to a provider in another country, which Article 9 of KVKK makes subject to conditions. Article 50 of the EU AI Act requires that people are informed they are interacting with an AI system unless that is obvious from the context. This section is not legal advice; review it with your legal counsel.
What are the running costs?
Model usage. Language models are billed per token, and the Anthropic and OpenAI pricing pages price input and output separately. Every reply sends the system instructions, the conversation history and the retrieved content to the model. As a conversation gets longer, each new message carries a longer input. For current unit prices, check the provider’s pricing page.
Channel fees. WhatsApp carries Meta’s message fees, covered in the next section. If you connect through a solution provider, ask whether it adds a platform fee of its own.
Hosting and infrastructure. This covers the server or cloud service the bot runs on, a vector database for content search and storage for logs.
Monitoring. Wrong answers, handed-off conversations and customer feedback are reviewed regularly. That takes people’s time more than tooling.
Maintenance and content upkeep. When campaigns, prices and policies change, the bot’s content has to change with them. Without a content owner the bot gives outdated information. Updating WhatsApp templates and handling model version changes belong here too.
How do WhatsApp message fees affect the cost?
According to Meta’s pricing documentation, the WhatsApp Business Platform no longer charges per conversation. It charges per delivered message. The rate depends on two things: the category of the message and the country code of the recipient. Template messages fall into marketing, utility and authentication categories. Free-form replies that are not templates count as service messages and can only be sent inside the 24-hour customer service window that opens when the customer writes to you.
As of November 2026, the same page states that each business phone number receives a monthly allowance of free service messages, and that service messages beyond the allowance, and utility templates sent inside the window, have been charged since 1 October 2026. Volume tiers apply only to utility and authentication messages. The rules do change, so read the page yourself before you budget.
This has three effects on chatbot design:
- Message count matters. A bot that splits one answer into three bubbles uses more service messages than a bot that replies in one.
- Messages outside the window are templates. Reminders and campaigns are charged by category, so plan marketing messages separately from the bot budget.
- Country makes a difference. The rate for a message is set by the recipient’s country, which matters if you serve customers in several markets.
Which choices raise or lower the cost?
| Cost driver | Type | Makes it cheaper | Makes it more expensive |
|---|---|---|---|
| Bot type | One-off and running | A chatbot that gives information | An agent that takes actions |
| Content preparation | One-off | Current, written FAQs and policies | Scattered or conflicting information, or knowledge held only in people’s heads |
| Integration | One-off | No integration, or one system with an API | Several systems, an ERP without an API, custom adaptations |
| Channel | One-off and running | One channel | Web, WhatsApp and social media together |
| Language | One-off | One language | Separate content and testing for each added language |
| Data-protection and legal review | One-off | A general information bot that asks for no identity | Account, health or financial data, cross-border transfer |
| Model usage | Running | Smaller model, short history, caching | Large model, long conversations, long context on every message |
| WhatsApp message fees | Running | One message per answer, service replies inside the window | Frequent templates, marketing messages, many countries |
| Monitoring and maintenance | Running | A named content owner, regular sampling | Fast-changing campaigns, content nobody owns |
How do you cut cost without cutting quality?
Start with one channel. Pick the channel your customers use most. Add the second once the first has given you real volume and cost data.
Start with one group of topics. Open with narrow content that covers the most frequent questions. For anything outside it, the bot hands over to a person.
Use a smaller model where it is enough. On the pricing pages, smaller models have lower unit prices. For simple questions such as delivery times, a small language model may be enough. Decide by testing with your own questions, not by guessing.
Content and retrieval quality first. If the bot retrieves the wrong passage, a larger model will not solve it. Fixing the content is cheaper and longer lasting than moving to a bigger model.
Use caching. According to Anthropic’s documentation, prompt caching reuses repeated prefixes of a prompt and reduces processing time and cost. Long system instructions that stay the same in every conversation are a good fit.
Limit the conversation history. In long conversations, sending a summary of the history, not all of it, shortens the input.
Do not ask the model for fixed answers. Opening hours or an address can come from prepared text. A rule-based flow and an LLM can work together in the same bot.
Leave the agent for last. Grant the ability to act only after the information bot has been measured and is trusted, and only for reversible actions.
What should you prepare before asking for a quote?
- Content count and formats. How many FAQs, policies and product pages are there, and are they in Word, PDF, web pages or Excel?
- Systems to integrate. Ecommerce platform, CRM, ERP, booking system: which of them have an API?
- Channels. Website, WhatsApp or both, and do you already have a WhatsApp Business Platform account?
- Languages. Which languages do customers write in?
- Expected volume. How many conversations a month, how many messages per conversation, and when are the peaks?
- Who approves answers. Who owns the content, who receives handed-off conversations, and what happens outside working hours?
- Personal data. Will the bot see identity, order or account details?
- Sample conversations. Real customer questions from recent months describe the scope better than any brief.
Where should you start?
- Pull the most frequent questions from the last few months of support records.
- Check whether the answers to those questions exist in writing and are current.
- Choose one channel and one bot type. For most businesses that is an LLM chatbot that gives information.
- Set the hand-off rules and name the content owner.
- Ask for a quote that separates one-off and running items, and get the assumed volume for model and channel costs in writing.
At Corrplus we follow the same working model on every project. Discovery starts with a free first call, followed by our two-week AI discovery sprint. In design we settle scope, architecture, and the data and data-protection assessment, and prepare a written quote. Build runs in short cycles, each ending with a working release. In the run phase, monitoring and improvement continue, as a managed service if you prefer.
For details, see our AI chatbot development, AI customer service and LLM integration services. Our pricing page explains how we price our work.
Want help with this?
Book a call to apply what “What determines the cost of a chatbot?” describes to your company. We will listen and agree the first step together.
Sources
- Meta for Developers, WhatsApp Business Platform pricing
- Anthropic, Pricing (Claude API documentation)
- Anthropic, Prompt caching
- OpenAI, API pricing
- EU AI Act, Article 50: Transparency obligations for providers and deployers of certain AI systems
- KVKK (Turkish Personal Data Protection Authority), Personal Data Protection Law No. 6698
Written by the Corrplus Team. How we write and check our posts: editorial policy.
Frequently asked questions
Can anyone quote a chatbot price without knowing the scope?
A figure given without scope is misleading. A simple menu bot and an agent that connects to your order system and takes actions share the same label but have very different budgets. Write down the bot type, the channel and the integrations, and ask for a quote broken into line items.
Is a WhatsApp chatbot more expensive than a website chatbot?
It usually has extra items. On the WhatsApp Business Platform, Meta charges for delivered messages at rates that depend on the message category and the recipient's country, and template approval and opt-in handling add to the build. A website chat has no channel fee. Check Meta's pricing page for current rates.
How do you estimate monthly model usage cost?
Estimate the number of conversations per month, the average messages per conversation and how much text goes to the model with each reply. Then multiply that volume by the current unit price of the model you choose. Providers price input and output tokens separately, so take the figures from the provider's pricing page.
Does a rule-based bot still make sense?
Yes, when the flow is narrow and known. For tasks with a few steps, such as booking an appointment or checking an order, a menu bot is cheap and predictable. When customers ask a wide range of questions in free text, keeping the scenario tree current gets expensive and an LLM chatbot fits better.
Related services
Related articles
What drives the cost of a RAG project?One-off and running cost drivers of a RAG knowledge assistant, the choices that raise or lower them, and what to prepare before a quote.
How to keep an AI agent inventoryA method for tracking owner, data, permissions, model, cost and logging for every AI agent as their number grows.
MCP 2026-07-28: a stateless coreWhat MCP 2026-07-28 changes: no sessions, routing headers, deprecated features and a migration checklist for teams running MCP agents.Let’s talk about your transformation
Tell us which work repeats every week. We will tell you honestly whether it is worth automating.
How we start
- A free first callWe pin down the problem and the goal together.
- A 2-week discovery sprintWe look at the data, the process and the tools, and decide where to start.
- A written proposalScope, architecture and timeline, in writing.