Home / Insights / Software Costs and Budgets
Software Costs and Budgets
AI Chatbot Development Cost: Setup, Integration and Ongoing Expenses
How to estimate the cost of a custom AI chatbot: define the task, size the one-time implementation, model recurring usage with explicit assumptions, compare three scopes and test sensitivity to low, base and high usage.
TopicsSoftware Costs and BudgetsGenerative AI and LLMsAI Agents and Automation
Key takeaways
- Chatbot cost has three parts: one-time setup, usage that scales with conversations, and ongoing maintenance and review.
- Build a usage model from conversations, messages, tokens and provider prices — and check provider pricing pages on the day you calculate.
- Scope drives cost: an FAQ bot, a permission-aware knowledge assistant and an action-taking assistant differ greatly in effort.
On this page
AI chatbots look simple from the outside — a chat window that answers questions — but cost depends heavily on what the bot must do. Answering public FAQs is very different from searching internal documents with permissions or taking actions in your systems. This guide breaks chatbot cost into setup, usage and maintenance, shows how to build a usage model and compares three scopes.
Usage prices change frequently and differ by provider and model. The calculations below use illustrative assumptions. Check current pricing on your provider’s official page before budgeting.
Define the chatbot task
| Scope | What it does | Main cost drivers |
|---|---|---|
| FAQ assistant | Answers questions from approved public content | Content preparation, guardrails, handover |
| Knowledge assistant | Searches internal documents, respecting permissions | Ingestion, retrieval, permissions, evaluation |
| Action assistant | Looks up records and performs actions (orders, bookings) | Integrations, authentication, approval flows, testing |
| Multilingual support | Any of the above in several languages | Content and evaluation per language |
Chatbot scopes.
One-time implementation work
| Work item | FAQ | Knowledge | Action |
|---|---|---|---|
| Discovery and conversation design | Low | Medium | Medium |
| Content or data preparation | Medium | High | Medium |
| Retrieval and indexing | Low | High | Medium |
| Integrations | Low (helpdesk handover) | Medium (document sources, SSO) | High (CRM, orders, payments) |
| User interface and channels | Low–medium | Medium | Medium |
| Evaluation set and testing | Medium | High | High |
| Security and permissions | Low | High | High |
Relative implementation effort by scope.
Evaluation is often underestimated. A reliable bot needs a set of real questions with expected answers, run before launch and after every change. Without it, you cannot tell whether a prompt or model change improved or harmed quality.
Recurring cost drivers
- Model usage: charges per token (input and output) or per request, depending on provider.
- Retrieval: search or vector database hosting and queries.
- Hosting: application servers, storage and logs.
- Monitoring and evaluation: dashboards, sampled reviews, regression tests.
- Human review and handover: staff time for conversations the bot escalates.
- Content updates: keeping source material current.
Three bounded scope scenarios
Illustrative effort at a blended $30 per hour:
| Item | A. Website FAQ assistant | B. Internal knowledge assistant | C. Order-status action assistant |
|---|---|---|---|
| Discovery and design (hours) | 40 | 80 | 80 |
| Content and data preparation (hours) | 60 | 120 | 60 |
| Build and integrations (hours) | 160 | 360 | 420 |
| Evaluation and testing (hours) | 60 | 140 | 140 |
| Project management (hours) | 30 | 70 | 70 |
| Total setup hours | 350 | 770 | 770 |
| Illustrative setup cost | $10,500 | $23,100 | $23,100 |
Illustrative setup estimates. Real effort depends on content, systems and quality targets.
Build a usage-based cost model
Monthly model cost = conversations × messages per conversation × (input tokens × input price + output tokens × output price)
Add retrieval, hosting and review costs:
Monthly running cost = model cost + retrieval and hosting + (escalated conversations × minutes per escalation × staff cost per minute) + maintenance hours × rate
Sensitivity example
Assumptions (illustrative): 6 messages per conversation; 1,500 input tokens and 300 output tokens per message (retrieved context makes input larger); illustrative model prices of $0.50 per million input tokens and $2.00 per million output tokens; hosting and retrieval $150 per month; 10% of conversations escalated at 4 minutes each and $0.20 per staff minute; 10 maintenance hours at $30.
| Usage level | Conversations per month | Model cost | Hosting and retrieval | Escalation cost | Maintenance | Monthly total |
|---|---|---|---|---|---|---|
| Low | 2,000 | $16.20 | $150 | $160 | $300 | $626 |
| Base | 10,000 | $81.00 | $150 | $800 | $300 | $1,331 |
| High | 50,000 | $405.00 | $150 | $4,000 | $300 | $4,855 |
Model cost per conversation = 6 × (1,500 × $0.50 + 300 × $2.00) / 1,000,000 = $0.0081. All prices are assumptions.
In this example, human escalation dominates running cost at every volume. Improving answer quality and handover design often saves more than choosing a cheaper model. Your numbers will differ, so recalculate with current provider prices and your own volumes.
Want this model filled in with your volumes and current provider prices?
Compare custom and off-the-shelf
Off-the-shelf support tools increasingly include AI agents with their own pricing. Intercom, for example, lists $0.99 per resolution for its Fin AI agent on its help centre. Whether a custom chatbot or an off-the-shelf tool costs less depends on volume, integrations and control requirements. We compare the two in detail in Custom AI chatbot vs off-the-shelf tools.
Frequently asked questions
What is the biggest hidden cost of a chatbot?
Often human review and escalation, and the effort to keep content and evaluation sets current.
Does the model provider affect cost much?
Model choice affects usage cost, but for many business bots, setup, integrations and review time matter more. Compare total cost.
Can we start with a small pilot?
Yes. A pilot on one channel and a limited set of questions gives real usage data for the cost model.
Will our data be used to train models?
It depends on the provider and configuration. Review data-retention terms and configure accordingly, or use private deployment.
Sources and methodology
Facts in this article were checked against the sources below on 5 November 2026. Directory data (ratings, review counts, rate bands and team sizes) changes over time and is shown as published on the date accessed. Illustrative examples are labelled as such and are not quotes.
- Intercom Help Center: Fin AI Copilot — accessed 5 November 2026. Fin listed at $0.99 per resolution
- PoliteTech: AI Chatbot Development — accessed 5 November 2026. Service scope
- PoliteTech Insights: RAG or fine-tuning? — accessed 5 November 2026. Retrieval background
Related reading
Continue your research
Considering a chatbot?
Tell us what it should answer or do, the channels and expected volume. We will outline scope, setup effort and a usage-based cost model.