A helpful chatbot makes a simple question easier to answer. It should not become another obstacle between a customer and your team. For a Malaysian business, that means testing the way people actually ask about price, delivery and service, including mixed-language messages.
Start with answers you already trust
Build a small, maintained knowledge base from your approved service information, delivery areas, returns policy and common questions. Assign someone to update it when the business changes. Do not let the bot invent a price or promise stock that has not been checked.
Separate general questions from account-specific requests. Order details and personal information should require appropriate identity checks, rather than being exposed because someone knows an order number.
Test local phrasing, not just perfect prompts
Try “boleh hantar Sabah?”, “berapa lama delivery?” and their English equivalents. Add spelling mistakes and questions with missing context. Check whether the bot asks a useful follow-up rather than confidently guessing.
Test disagreement too. What happens when a visitor asks it to ignore the policy, disclose another customer’s details or give a discount it cannot approve? Use invented customer records while testing these boundaries.
Make the handover part of the design
Give customers an obvious way to contact a person. Explain your actual support hours and what happens outside them. When possible, pass a concise conversation summary to the team so the customer does not need to start again.
Review unanswered questions and incorrect answers after launch. A useful success measure is whether the customer got a correct answer or reached the right person, not simply whether a conversation ended.
Be clear that it is AI
Label the assistant honestly. AI Malaysia’s voluntary AI Code of Ethics emphasises responsible, human-centred implementation. Our practical recommendation is to give the bot a narrow job, clear boundaries and an accountable human owner.
