AI agents have moved from demos to real work: answering support tickets, booking appointments, qualifying leads, reading invoices and updating CRMs. The question we hear most from business owners is simple: what does it cost to build one? Here is a straight answer, with cost tables, the factors that move the price, a worked example, and the running costs most quotes leave out.
AI agent development cost ranges
| Agent type | Build cost | Timeline | Typical use |
|---|---|---|---|
| Knowledge chatbot | $5,000 to $15,000 | 3 to 6 weeks | Answers questions from your FAQs, policies and product docs |
| Business process agent | $15,000 to $50,000 | 6 to 12 weeks | Books appointments, creates orders, updates the CRM with approvals |
| Voice or multi-agent system | $50,000 to $150,000+ | 3 to 6 months | Phone agents, several cooperating agents, multilingual at high volume |
A chatbot that only answers questions sits at the low end. Cost rises once the agent needs to do things, because every action needs permissions, validation, fallbacks and testing.
What drives the cost of an AI agent
1. How narrow the job is
The agents that work in production are usually narrow. "Handle all customer service" is expensive and hard to make reliable. "Answer order status questions and escalate everything else" is cheap and dependable. Scoping the job tightly is the biggest lever on both cost and quality.
2. Integrations and actions
Reading from a knowledge base is straightforward. Writing to your booking system, CRM, ERP or WhatsApp Business account means building secure connections, handling failures and making sure the agent cannot take actions it should not. Each integration typically adds a few days to a few weeks.
3. Data preparation
If your policies, product data and FAQs are scattered across PDFs, spreadsheets and people's heads, part of the project is organizing that knowledge so the agent can retrieve it accurately. Clean data makes the agent better and cheaper to run. Our guide to RAG in production explains how retrieval is built properly.
4. Guardrails and human handoff
Production agents need limits: what topics they refuse, when they hand over to a person, how they handle angry customers, and how every conversation is logged. This work separates a reliable agent from an embarrassing one. We cover it in our guide to implementing AI guardrails.
5. Voice and languages
Voice agents add speech-to-text, text-to-speech and real-time latency requirements. Supporting more than one language, especially right-to-left scripts like Arabic, adds testing and prompt work for each language.
A worked example: an appointment booking voice agent
A common request we build is a voice receptionist for clinics and service businesses. The agent answers calls in Arabic and English, checks availability in the booking system, books or reschedules appointments, sends a WhatsApp confirmation and hands complex calls to staff. Every call is logged in a simple CRM so the team can review conversations.
That sounds like one agent, but it combines telephony, two languages, a calendar integration, messaging, a CRM and an admin view. That is why voice agents with real actions sit in the upper band, while a website chatbot answering FAQs from the same clinic's documents would sit at the low end.
Monthly running costs after launch
Unlike a normal app, an AI agent has a usage cost every time it works.
| Cost item | Typical monthly range | Notes |
|---|---|---|
| Model usage (tokens) | $50 to $500+ | For a few thousand conversations; depends on model and conversation length |
| Voice minutes | $0.05 to $0.20 per minute | Telephony plus speech services, for phone agents |
| Hosting, database and logs | $30 to $300 | Servers, vector database if used, monitoring |
| Ongoing improvement | A few hours a month | Reviewing failed conversations and updating knowledge |
Model costs can be cut significantly with caching, routing simple questions to cheaper models and trimming context. See our article on reducing LLM token costs.
How long does it take?
A focused single-task agent can go live in 3 to 6 weeks. Agents that act inside several business systems usually take 2 to 3 months. We recommend launching to a small group first, measuring accuracy against real conversations, then rolling out wider.
Build vs buy: when an off-the-shelf tool is enough
If your need is generic, like a website FAQ bot, an existing SaaS tool may be cheaper. A custom agent makes sense when it has to work inside your own systems, follow your specific rules, protect sensitive data, or when per-seat SaaS pricing gets expensive at your volume.
Frequently asked questions
How much does a simple AI chatbot cost?
A chatbot that answers questions from your own documents and FAQs typically costs $5,000 to $15,000 to build, plus model and hosting fees that usually start under $100 a month.
What is the difference between a chatbot and an AI agent?
A chatbot answers questions. An AI agent also takes actions, such as booking an appointment, creating an order or updating a record in your CRM. Those actions are what add cost, because each needs integration, permissions and testing.
How much does it cost to run an AI agent per month?
Most small and mid-size deployments spend $100 to $1,000 a month across model usage, hosting and monitoring. Voice agents also pay per call minute.
Can an AI agent work in Arabic and English?
Yes. Modern models handle Arabic well, but each language needs its own testing, prompts and, for voice, suitable speech models. Budget extra time for every additional language.
How do I keep an AI agent from giving wrong answers?
Ground it in your own verified data, restrict what topics it handles, add clear handoff rules to a human, and review conversation logs regularly. Narrow scope is the most effective safeguard.
Get a quote for your AI agent
Softzee builds production AI agents and chatbots for support, sales and operations, including Arabic and English voice agents. See our AI agent development services and AI chatbot development, or book a free strategy call to scope your agent and get a clear cost range.