Budget $8,000–$15,000 and 28 days for a production-ready AI customer service chatbot. That is not a proof-of-concept that hallucinates answers or a basic decision tree that forwards every third question to a human. That is a live chatbot trained on your documentation, connected to your support tools, and capable of resolving the queries your team answers fifty times a day.
Western agencies quote $40,000–$80,000 for the same scope. The gap is real, and it has nothing to do with quality. It has everything to do with whether the team building your chatbot has replaced manual, repetitive work with AI tools.
What factors drive the cost of a customer service chatbot?
The honest answer: three things determine almost all of the price. How your chatbot retrieves answers, how many systems it connects to, and whether a human needs to step in when the bot cannot help.
Answer retrieval is where the biggest cost differences hide. A rule-based chatbot follows a fixed decision tree. Someone on your team builds every branch by hand, which is cheap to write and instantly obsolete whenever your product changes. An AI-native chatbot reads your documentation, FAQs, and past support tickets, then generates answers from that knowledge. Gartner found that AI-powered virtual agents resolve 65–80% of customer queries without human involvement, compared to roughly 30% for rule-based systems. The setup costs more, but the outcome is a bot that actually works.
Integrations are the second lever. A chatbot that only lives on your website and reads a static FAQ costs a fraction of one that pulls order status from your database, logs conversations to your CRM, and escalates to a live agent in your helpdesk. Every integration is a separate piece of work, and the complexity compounds when data has to move in both directions.
Live handoff is the third factor. Detecting when a customer is frustrated, transferring the conversation to a human with full context, and routing to the right team member is more complex than it sounds. Getting it wrong costs you the customer. Getting it right requires building logic that most basic chatbot platforms do not include out of the box.
How does an AI-native customer service chatbot process queries?
A traditional chatbot matches keywords. A customer types "refund" and the bot returns a canned response about your refund policy. It does not understand that the customer received the wrong item, has already emailed twice, and is on the verge of filing a chargeback.
An AI-native chatbot works differently. When a customer sends a message, the bot does not look for a keyword match. It reads the full message, searches your documentation for the most relevant information, and generates a response specific to that customer's situation. The mechanism is something called retrieval-augmented generation: your product documentation, FAQs, and support history become a knowledge base the AI searches before answering. Nothing is hard-coded. When your return policy changes, you update one document and every answer the chatbot gives updates automatically.
MIT's 2023 research found that AI-assisted customer service reduced average handle time by 34% and increased the share of issues resolved in a single interaction by 25%. The reason is specificity. A customer asking about a delayed shipment gets an answer that accounts for their order, not a generic "please allow 5–7 business days" that was written for a different situation entirely.
At Timespade, the build follows four stages. The first week locks the scope: which questions the bot handles, which systems it connects to, and what happens when it cannot help. Weeks two and three are the build, with AI generating the integration code and the knowledge retrieval layer while a senior engineer handles the logic unique to your product. Week four is testing, where real customer queries from your support history run through the bot to confirm it handles edge cases correctly before any customer sees it.
Should I build a custom chatbot or use a platform?
Platforms like Intercom, Zendesk AI, and Freshdesk have chatbot features built in. For some companies, that is the right answer. For others, it is an expensive trap.
The platform path costs $300–$2,000 per month in subscription fees, plus setup time, plus the cost of whatever you cannot do because the platform was not built for your specific workflow. That math works if your use case fits the platform's defaults. It breaks down when you need the bot to access your proprietary database, handle a conversation flow that differs from the platform's templates, or integrate with a tool the platform does not officially support.
A custom build costs more upfront and nothing ongoing beyond hosting. The calculus shifts once you hit roughly 18 months of platform fees. After that, the custom build is cheaper and does more.
| Approach | Upfront Cost | Monthly Cost | Customization | Best For |
|---|---|---|---|---|
| SaaS platform (Intercom, Zendesk AI) | $0–$2,000 setup | $300–$2,000/mo | Low | Simple FAQ handling, standard workflows |
| No-code builder (Tidio, Chatbase) | $500–$2,000 | $50–$300/mo | Medium | Small teams, limited integration needs |
| Custom build, Western agency | $40,000–$80,000 | $500–$2,000/mo support | High | Large enterprises with compliance requirements |
| Custom build, AI-native team (Timespade) | $8,000–$35,000 | $200–$800/mo hosting | High | Startups and growth-stage companies |
The no-code builders in the middle are worth knowing about. Chatbase, Tidio, and similar tools let you upload your documentation and get a working chatbot in hours. They are genuinely good for simple use cases, and they cost almost nothing. The ceiling is low: no custom integrations, no complex handoff logic, no ability to pull live data from your systems. When customers start noticing the limitations, you will rebuild anyway.
What ongoing costs come after the initial build?
The build is a one-time cost. The chatbot is not.
AI chatbots consume tokens every time they answer a question. OpenAI charges roughly $0.002–$0.015 per 1,000 tokens depending on the model. A customer service chatbot handling 10,000 conversations per month with an average of 400 words per exchange runs about $80–$120 in AI usage costs. At 100,000 conversations, that is $800–$1,200 per month in token costs alone. This is not a surprise expense if you plan for it, but many founders do not.
Hosting the chatbot itself costs $50–$200 per month depending on traffic. Maintenance, which covers keeping your knowledge base current, monitoring for hallucinations, and patching integrations when a connected system changes its API, runs another $300–$800 per month with an AI-native team. A Western agency charges $1,500–$3,000 per month for the same maintenance scope.
| Ongoing Cost | Monthly Range | What It Covers |
|---|---|---|
| AI usage (token costs) | $80–$1,200/mo | Cost per conversation based on volume |
| Hosting | $50–$200/mo | Keeping the chatbot live and fast |
| Maintenance and updates | $300–$800/mo | Knowledge base updates, integration patches, monitoring |
| Escalation tooling (if not already in place) | $100–$500/mo | Live chat software for human handoff |
One number worth tracking: the cost per resolved ticket. A human support agent costs roughly $8–$15 per resolved ticket when you factor in salary, benefits, and overhead (Forrester, 2023). A well-built AI chatbot resolves tickets for $0.50–$2.00 each. At 5,000 tickets per month, that is $25,000–$75,000 in human support costs versus $2,500–$10,000 for the bot. The ROI math is straightforward once volume is high enough.
How do I estimate ROI before committing to a budget?
Start with your current support volume and cost. How many tickets per month does your team handle? What does each one cost, including the time of the person who answers it?
A customer service chatbot typically handles 60–75% of incoming queries without human involvement, based on IBM's 2023 benchmarks across enterprise deployments. The remaining 25–40% get escalated with full conversation context, so the human agent does not start from scratch. The time savings on escalated tickets are real too, typically 40–50% faster resolution because the agent already has the customer's message, order history, and what the bot already tried.
A company handling 3,000 tickets per month at $10 per ticket spends $30,000 per month on support labor. A chatbot resolving 65% of those tickets at $1.50 each drops that to roughly $16,000 per month ($4,500 in bot costs plus $11,250 for the 35% handled by humans). That is $14,000 per month in savings, or $168,000 per year. An $8,000–$15,000 build pays for itself in five to seven weeks.
The ROI math does not always work. If your support volume is under 500 tickets per month, a no-code tool or a platform add-on is probably the right answer. The economics of a custom build only click when there is enough volume to justify the setup cost, or when your support workflow has enough proprietary complexity that a platform cannot handle it.
Timespade builds chatbots within the same 28-day framework used for every product, because the infrastructure is identical: a knowledge retrieval layer, an integration with your existing tools, and a handoff system for queries that need a human. The $8,000 entry point covers a chatbot handling one core use case (support FAQ, order status, or appointment booking). Broader coverage with multiple integrations runs $20,000–$35,000. Western agencies start conversations about similar scope at $40,000, before any ongoing fees.
If your team is answering the same questions more than twenty times a week, the cost of not building a chatbot is already measurable. Book a free discovery call
