How much does an AI chatbot cost in 2026? A budgeting guide for founders and PMs

The first quote a founder usually gets for an AI chatbot is between two thousand dollars and three hundred thousand dollars. That spread is not a typo. It reflects how different vendors define the word “chatbot,” where in the world they’re based, and how little of the real cost is visible up front.

This guide is for the person about to commission one. Founders, product managers, operations leaders. The goal is to walk away with enough understanding of the cost structure that you can read a proposal critically, ask better questions, and avoid the budget surprises that show up three months into a project.

What you’re actually buying

The word “chatbot” covers four very different products. Knowing which one you need is the first cost-control decision.

Rule-based bot. Buttons and decision trees. Whatever you call it on the marketing page, this is glorified IVR. Cheap to build, cheap to operate, terrible user experience for anything off-script. Budget: $1,500 to $5,000 with a competent SE Asia agency.

Hosted vendor chatbot. A monthly subscription on a platform like Intercom Fin, Drift, or Tidio. Decent for FAQ deflection. You don’t own the data or the prompts. The platform decides when to raise prices. Budget: $200 to $2,000 per month, no build cost (vendor pricing is global).

Custom AI chatbot. Built on top of a large language model (GPT-4, Claude, or open-weight), trained on your documents, deployed to your channels. You own the code, the prompts, and the conversation data. This is what most people mean when they say “we want an AI chatbot.” Budget: $8,000 to $35,000 to build with a senior SE Asia team, plus ongoing costs. The same build from a US or EU agency typically runs $25,000 to $90,000.

Multi-tenant enterprise AI platform. Custom development that includes role-based access, audit logging, compliance work, and integration with your existing identity systems. Budget: $40,000 to $150,000+ from a SE Asia agency, $150,000 to $500,000+ from a Western shop.

Most projects we see should be the third one. The buyer assumed the second was good enough until it wasn’t, or the fourth was needed when it wasn’t.

Where the build budget actually goes

For a typical custom AI chatbot, here’s roughly how the money is spent:

Discovery and design (10-15% of budget). Mapping the actual conversations users will have, the knowledge sources the bot will draw from, the channels you need, and the handoff rules to human staff. Skipping this is the most common reason chatbots get rebuilt within a year.

Knowledge ingestion (15-20%). Getting your content (website, FAQs, PDFs, knowledge base) into a form the chatbot can search. This is rarely as simple as “upload our docs.” Real ingestion involves cleaning, chunking, and quality-checking the inputs.

Retrieval system (15-20%). The part that finds the right information to answer each question. Most vendors gloss over this; the quality of retrieval is the single biggest predictor of whether your chatbot will actually be useful.

Generation and grounding (10-15%). The prompts, the model selection, and the guardrails that make sure the bot only answers from your content and admits when it doesn’t know.

Channel integrations (10-20%). Web widget, Messenger, WhatsApp Business API, Slack, Microsoft Teams, Shopify. Each one is more work than it sounds. WhatsApp in particular requires Meta business verification, which can take weeks on its own.

Lead capture and handoff (5-10%). What happens when the bot detects high intent. Booking a call, creating a ticket, escalating to a human, dropping a record into your CRM.

Admin tools and analytics (10%). Where your team sees what the bot is doing, edits prompts, reviews conversations, and exports leads. This is often the first thing cut from budgets and the first thing missed once the bot is live.

The ongoing costs nobody quotes you on

Here is where surprises live.

LLM API charges. Every conversation costs real money. A single question-and-answer turn typically costs between half a cent and three cents depending on the model. For a chatbot handling a thousand conversations a day, that’s $150 to $900 per month in API fees alone, and it scales with usage. These costs are global — the model provider charges the same whether you’re in Manila, Singapore, or San Francisco.

Infrastructure. Hosting, database, monitoring, error tracking. Usually $80 to $300 per month for a chatbot at typical small-business scale on DigitalOcean, AWS, or Render.

Maintenance. Models get deprecated. The platform you integrated with changes its API. New compliance rules show up. Most production chatbots need a few hours of engineering attention every month just to stay running. If you’re not paying for this, it’s accumulating as technical debt.

Knowledge updates. When your products change, your pricing changes, or your policies change, the chatbot’s knowledge has to be updated. Some platforms make this self-service; others charge for it as a service.

Quality monitoring. Without an eval harness — a set of test questions run regularly to catch regressions — chatbot quality drifts. Bad answers go undetected until a customer complains. This is the second-most-common reason chatbots get retired within a year.

A reasonable rule of thumb: ongoing costs are 15-20% of the build cost annually, scaling with conversation volume. A $20,000 build costs roughly $3,000 to $4,000 per year to keep working well at modest traffic, and more once the LLM bill grows with usage.

The questions to ask before you sign

We’ve sat on both sides of these conversations. The questions below filter out vendors who will leave you with a regret six months in.

1. Who owns the chatbot when the project ends? If the answer is anything other than “you, on your infrastructure,” you’re renting, not buying. That’s not always wrong, but it’s a different deal.
2. How do we change the prompts after launch? If the answer involves the agency, you’re locked in. If the answer is “you edit a YAML file” or “an admin UI,” you’re not.
3. What happens when the LLM provider changes their model or pricing? A serious vendor has thought about model migration. A weak one will hand you a surprise invoice.
4. How do you measure whether the chatbot is working? “Customer satisfaction surveys” is the wrong answer. The right answer involves an eval set, a quality dashboard, and an alert when accuracy drops.
5. What’s your refusal strategy? When the bot doesn’t have a good answer, what does it say? “I don’t know, let me get a human” is the right answer. “Confidently make something up” is the wrong one.
6. Can it cite its sources? A chatbot that shows where each claim came from is dramatically more trustworthy than one that doesn’t.
7. What’s the actual monthly run-cost forecast at our expected volume? Get a number. If they can’t give one, they haven’t thought about it.

A realistic budget range

For a typical mid-market business commissioning a custom AI chatbot from a senior SE Asia team:

• Build cost: $15,000 to $40,000
• First-year operating cost: $4,000 to $15,000, depending on conversation volume
• Time to launch: 8 to 16 weeks from kickoff
• Ongoing maintenance: roughly 15-20% of build cost per year after year one

Anything under $8,000 is almost certainly a templated chatbot dressed up to look custom. Anything over $80,000 for a single chatbot (not a platform) from a SE Asia agency is paying for overhead, not engineering.

For context, US and Western European agencies typically quote 2.5 to 4 times these numbers for equivalent work — a $25,000 build here is the same scope a US shop quotes at $80,000. The talent gap is much smaller than the price gap suggests. The big enterprise quotes you’ll see — $200k, $500k, $1M — are real prices for real projects, but they’re for multi-channel, multi-language, compliance-heavy systems used by Fortune 500 companies. If you’re a mid-market business with a clear use case and a few integrations, that’s not your tier from any geography.

If you’re starting this conversation

We build custom AI chatbots end-to-end and have helped clients audit existing chatbot projects that went off the rails. If you want a written estimate based on your actual needs — not a generic deck — see our AI chatbot development page or email [email protected]. One scoping call. No deck. You leave with a number.