Adding an LLM to a WhatsApp support line takes an afternoon. Making it safe to leave running takes longer. These are the six failures to design for.
It invents a discount
A customer asks “any offers?” and the model makes one up. Fix: answer only from retrieved content, and add a hard rule that prices and offers come from your product data.It guesses order status
“Your order will arrive tomorrow” with no lookup behind it. Fix: route order questions to a database lookup by phone number. Never let the model answer them from context.There’s no way to reach a human
The customer types “agent” three times and gets three AI answers. Fix: check for handoff keywords before calling the model at all.It answers off-topic questions
Someone asks it to write an essay or solve homework. Besides wasting tokens, Meta’s WhatsApp Business terms (since January 2026) don’t allow general-purpose AI assistants. Fix: scope the system prompt to your business and refuse anything else politely.Outdated product information
The bot quotes last month’s price. Fix: re-index content when products or pages change, not on a weekly schedule.Wrong language
The customer writes in Hindi or Tamil and gets an English reply. Fix: tell the model to reply in the customer’s language, and test it with real mixed-language messages (“size M available hai?”).
Bonus: log everything
Store the question, retrieved sources, answer and handoff reason for every message. You’ll find content gaps within a week.
Wabbiyo’s WhatsApp AI chatbot, a WordPress plugin on the official Cloud API, handles retrieval from your site, order lookups and human handoff: Click here to know more
