How much does a chatbot cost in 2026?
A chatbot costs either a recurring platform subscription or a one-time custom build plus running costs. The price is driven by five things: the number of channels, the number of systems it connects to, whether it is scripted or AI-based, message and model usage, and maintenance. At Goldenroz, custom chatbot projects start at $25,000 at a fixed price per phase.

Two ways to pay for a chatbot
There are two pricing structures, and most confusion comes from comparing a number from one with a number from the other.
A platform subscription. You configure a bot inside a vendor’s product. You pay monthly. Vendors price by seats, by conversations, by resolved conversations, or by a mix. The vendor hosts it and decides what it can do.
A custom build. An engineering team builds the bot for your process and systems. You pay for the build, then pay running costs directly: model usage, channel fees and hosting. You own the code.
We do not quote market averages in this post. Published “average chatbot cost” figures mix a small website widget with an enterprise deployment, and we have no way to verify them. What can be stated reliably is what drives the price, and how to estimate your own.
The five cost drivers
1. Channels
Each channel is its own integration with its own rules.
- Website chat is the simplest: a widget and your own backend. See website chatbot development.
- WhatsApp adds Meta account setup, templates, the 24-hour window and per-message charges. The rules are in our WhatsApp Business API guide.
- Voice adds telephony, speech recognition and speech synthesis, and is a separate class of project.
- Email, SMS and in-app chat each add formatting and delivery handling.
The conversation logic can be shared across channels. The channel adapters, the testing and the channel fees cannot.
2. Integrations
A chatbot that only answers from documents needs one connection: its knowledge base. A chatbot that tells a customer where the order is needs the order system. One that changes a booking needs write access to the booking system, with checks.
Each integration needs authentication, error handling, rate limits, test data and a decision about what the bot may do. Integrations are the largest and most variable part of a build budget. Count them before asking for a quote.
3. Scripted or AI
| Scripted chatbot | AI chatbot | |
|---|---|---|
| How it decides | Decision tree and keywords | Language model with instructions and your data |
| Handles unexpected phrasing | Poorly | Well |
| Build effort for the conversation | Grows with every new path | Mostly in grounding, testing and guardrails |
| Cost per message | Near zero | Model usage per message |
| Risk | Dead ends and frustrated users | Wrong answers, which need evaluation and fallbacks |
| Fits | A few fixed tasks | Open questions, many topics, natural language |
Most production bots are hybrids: fixed steps where the process must not vary, such as identity checks and payments, and a model for understanding and answering.
4. Volume and usage
Usage costs are the part that scales with success. There are two kinds.
Model usage. Model providers bill by tokens, which are pieces of text, for both what you send and what the model writes. What you send includes the instructions, the conversation so far and any retrieved documents, on every turn. Long instructions and long conversations cost more. The levers are: a smaller model for simple turns, prompt caching for the repeated parts, and tight retrieval that sends only the relevant passages.
Channel fees. Website chat has none. WhatsApp bills per delivered message by category and country. As of November 2026, Meta’s pricing documentation states that free-form service replies are free only up to a monthly allowance per phone number and are charged above it, so support volume on WhatsApp is no longer a zero-cost line.
5. Maintenance
A chatbot is not finished at launch. Budget for:
- Keeping the knowledge base current.
- Reviewing conversations and adding failures to the test set.
- Model updates. Providers retire model versions, and each change needs a test run.
- Changes in the systems the bot connects to.
- Monitoring and on-call for a customer-facing service.
A reasonable planning assumption is that someone on your side owns the bot part-time from launch onward, even if a vendor does the engineering.
How to estimate your running cost
Use this formula. The numbers in the example are hypothetical and only show the arithmetic.
monthly model cost = conversations per month
x model calls per conversation
x cost per model call
Example: 10,000 conversations a month, 6 model calls per conversation, and a measured cost of $0.01 per call.
- 10,000 x 6 = 60,000 model calls
- 60,000 x $0.01 = $600 a month in model usage
The cost per call is the number you must measure, not assume. It depends on the model, the length of your instructions and how much retrieved text you send. Take the provider’s current published price per token, run 100 real conversations in a test environment, and read the cost from the usage report. Then add channel fees and hosting.
Do this before launch. A design that sends ten pages of context on every turn can cost several times more than one that sends two paragraphs, for the same answers.
Platform subscription vs custom build
| Platform subscription | Custom build | |
|---|---|---|
| Up-front cost | Low | Higher |
| Recurring cost | Monthly fee, often tied to seats, conversations or resolutions | Model usage, channel fees, hosting, maintenance |
| Time to first version | Short | Longer |
| Integrations | What the vendor supports, plus workarounds | Any system with an API or database |
| Control over behavior | Within the product’s settings | Full |
| Data location | The vendor’s infrastructure | Your choice |
| Ownership | You rent it | You own the code |
| Switching cost | High once flows are built in the vendor’s format | Lower, the code is yours |
A platform is the right first choice when your needs match what the product does out of the box, your help desk already has a bot add-on, and the volume is moderate.
A custom build makes sense when the bot must act inside your own systems, when per-conversation platform fees at your volume exceed what the bot costs to run yourself, when data must stay in your environment, or when the bot is part of your product.
Compare over three years, not one month. Multiply the platform’s per-conversation or per-resolution fee by your expected volume, and set it against the build price plus running costs.
What Goldenroz charges
Our published price for build work is the same for chatbots as for other projects: projects start at $25,000 at a fixed price per phase, quoted after a scoping call. Shai does the engineering. The client owns the code and the accounts, and pays model providers and channel fees directly from those accounts.
What moves a quote up from the starting point:
- More than one channel.
- Each system the bot reads from, and more for each it writes to.
- Per-user permissions on internal knowledge.
- A human handoff built into an existing help desk.
- Voice.
We do not publish hourly rates or tiers beyond this, because scope decides the price. Our AI chatbot development page lists what a project includes.
Where budgets go wrong
- Pricing the demo. A bot that answers ten prepared questions is a small part of the work. Error handling, handoff, evaluation and monitoring are the rest.
- Forgetting content. If the knowledge base is outdated, cleaning it is part of the project, whoever does it.
- No usage estimate. Model and channel costs are discovered on the first invoice.
- Building an agent when a chatbot would do. Actions cost more than answers. The difference is laid out in AI agent vs chatbot, and agent budgets in how much it costs to build an AI agent.
- No owner after launch. Quality drifts when nobody reads the conversations.
Budget checklist
- Channels listed, with fees for each
- Systems to read from and write to, counted
- Scripted, AI or hybrid decided per task
- Expected conversations per month
- Measured cost per model call from a test run
- Handoff design and the help desk it connects to
- Content cleanup scoped
- Maintenance owner named
- Three-year comparison against a platform subscription
What to do next
Fill in the checklist with your own numbers. If the bot is mainly for support, read our customer service chatbot page and the customer service chatbot guide. If you want a fixed quote, book the free one-hour consultation and bring the list of channels and systems. Projects start at $25,000 at a fixed price per phase.