AI chatbot cost in Malaysia: what you are actually paying for

What changes the cost of an AI chatbot?

Scope changes the cost far more than the model does. A chatbot that answers from a short FAQ is a small project. One that searches your policies, reads live order data, works in English, Malay and Chinese and hands conversations to staff is a system with several connected parts.

Most business chatbots now use retrieval: the bot searches your own documents and passes the relevant passages to a language model, which writes the reply. This is often called RAG. The work sits in preparing those documents, keeping them current and testing that answers stay within them. A bot that confidently invents a refund policy is worse than no bot.

Channels add separate work. A website widget, WhatsApp, Facebook Messenger and an internal staff assistant each have their own setup, rules and testing. Integration with a CRM, booking system or order database is usually the largest single item, because the bot must read and sometimes write real records safely.

• Knowledge sourceStatic FAQ, document library, or live data from your systems.
• ChannelsWebsite, WhatsApp, Messenger, or an internal staff tool.
• LanguagesEnglish, Bahasa Malaysia and Chinese need testing in each language, not just translation.
• HandoverHow and when a human takes over, and what context they see.

What do Malaysian providers publicly charge?

Publicly listed ranges from Malaysian providers give a starting reference. Advisory Apps, for example, lists RM20,000 to RM50,000 for basic AI features such as chatbots built on existing language model APIs. Treat any published range as one supplier's positioning rather than a market rate for your brief.

The same provider lists RM50,000 to RM120,000 for intermediate work such as intelligent search or OCR with data extraction, and higher figures for custom-trained models. These ranges are useful for setting expectations with management. They do not tell you what a specific scope will cost, because data readiness and integration depth vary widely.

Macoin is flexible on how a project is structured. A common approach is a focused pilot on one channel and one set of questions, measured against real conversations, before committing to a wider rollout. That keeps the first spend proportional to what the business has learned. Send the brief through WhatsApp for a quotation.

What are the monthly running costs?

Running costs are usually model usage, hosting and channel fees. Model providers such as OpenAI, Anthropic and Google charge by tokens, which roughly track the length of questions, retrieved passages and answers. WhatsApp adds Meta's own per-message charges. Ask for an estimate based on your expected conversation volume.

Token pricing differs by model and changes over time, so check the providers' official pricing pages rather than a figure in a proposal from months ago. A smaller, cheaper model is often adequate for routine questions, with a larger model reserved for harder cases. That routing decision can matter more to the monthly bill than the headline price per token.

Hosting covers the application, the document index and logs. Keep the accounts in the business's name where possible, so billing and access stay with you. Agree who monitors usage, who receives alerts when spending rises unexpectedly, and how the documents the bot relies on are updated when a policy changes.

How do we keep the chatbot accurate and safe?

Limit the bot to approved sources, test it against real questions, and give it a clear way to say it does not know. Log conversations for review, protect personal data under Malaysia's PDPA, and define when a human takes over. Accuracy is an ongoing operating task, not a launch checkbox.

Before launch, collect a set of real customer questions, including awkward ones, and agree what a correct answer looks like. Re-run that set whenever documents or the model change. The OWASP Top 10 for LLM Applications is a practical reference for risks such as prompt injection and the bot revealing information it should not.

Macoin builds its own AI products, with three in development, and works with clients such as Brainbox, an AI platform for Malaysian SMEs. That experience shapes how we scope a chatbot: start narrow, measure real conversations and expand once the answers hold up. WhatsApp us your use case and expected volume for a quotation.