Most WhatsApp bots fail for a boring reason: the knowledge behind them is thin. The model is fine; the FAQ is not. Here is the process we use at Oxedge when we set up WhatsApp AI agents for small businesses in Dubai.
1. Start from real chats, not from a brainstorm
Export the last 200 to 300 WhatsApp conversations (with customer consent and personal data removed). Group the first customer message of each chat into buckets. In most service businesses, five to eight questions cover over 70% of first messages: price, availability, location, timing, payment and "can you do X".
2. Write answers the way your best staff member talks
Short, specific and with one next step. "Our standard AC service is from AED 150 for a split unit. Shall I check a slot for tomorrow morning?" beats a paragraph of policy text.
3. Give the agent hard limits
- Never quote a price that is not in the price sheet.
- Never confirm a booking without checking the calendar.
- Hand over to a human when the customer is upset, asks for a refund, or asks the same thing twice.
4. Handle English and Arabic properly
Keep a separate, reviewed answer set for Arabic rather than relying on live translation for prices and policies. Mixed-language messages ("hi, kam el price?") are common in the UAE, so test them.
5. Measure qualification, not message count
The useful metrics are: share of chats that reach a qualified lead (name, need, budget or date), time to first reply, and handover rate. If handovers stay above 30% after two weeks, the FAQ needs more work.
6. Review weekly for the first month
Read every chat the agent could not resolve and add or fix one answer per gap. After a month the agent usually handles the routine questions on its own.
If you want to see how this looks for a real business, we wrote up our approach to WhatsApp AI agents in Dubai on the Oxedge site.
