Choose answerable questions
Group inbound requests into product information, service hours, request status, complaints, and transactions. FAQs with approved answers are a useful starting point. Refunds, pricing exceptions, and policy changes need separate rules. Keep anonymized examples and an owner for each source. If prices or availability change, define the source and refresh process. When information is missing, the bot should acknowledge that and offer staff assistance.
Design the workflow before selecting a model
A sample flow identifies the request, retrieves approved information, drafts a response, and determines whether staff are needed. Some services only require AI drafts for review. Direct FAQ responses can be considered after testing. Separate rule-based routing, information retrieval, and actions that change records. These choices affect integration and cost. Clear rules can handle many steps without broad model authority.
Make handoff usable
Pass the question summary, sources used, unresolved issue, and destination team. Tell the customer that staff will take over and how to continue. Define the queue, service hours, and owner. If staff are unavailable, provide an option the business can actually support. Ticket creation alone is not request resolution. Test missing information, explicit requests for a person, and downstream-system failures.
Check WhatsApp platform requirements
Confirm account and number ownership, Business Platform eligibility, helpdesk integration, and provider costs. The WhatsApp Business policy reviewed on October 3, 2026 covers opt-in for contacting people, templates for business-initiated conversations, and responses within the customer service window. Automated responses need clear escalation paths. Policies and prices can change, so check Meta’s current documentation during scoping. Chatbot development does not imply that every channel and platform cost is included.
Measure quality and customer experience
Test sanitized real questions, language variations, typos, ambiguity, and out-of-scope topics. Assess answer accuracy, appropriate escalation, staff wait time, and corrections. Bot message count is not resolved-case count. Compare resolution within equivalent categories and check customer experience after handoff. Monitor outdated sources and assign an update owner.
Prepare an implementation brief
Describe the business, question categories, channels, supporting systems, and staff workflow. State whether the first goal is faster retrieval, faster drafts, or direct answers to selected FAQs. KODE can help design AI chatbots, knowledge bases, evaluations, and scoped integrations. Begin with one question category with approved sources and a service owner.