WhatsAppBusiness+AIautomation:apracticalguideforshopsinKurdistan

TL;DR: WhatsApp is where Kurdistan businesses live. AI can absorb 70–90% of routine WhatsApp message handling in EN/AR/KU while routing genuine sales conversations to a human. Setup is two weeks. This is the practical playbook.

Why WhatsApp first, before everything else

In Kurdistan, customer service doesn't run on email or chatbots on websites. It runs on WhatsApp. Hotels take bookings through WhatsApp. Restaurants take reservations. Clinics confirm appointments. Shops sell. If you're not on WhatsApp Business, you're not in the conversation. If you ARE on WhatsApp Business and a human is replying to every "what time do you open" message, you're burning the most expensive resource in your business — attention.

What an AI WhatsApp assistant should handle

• Hours of operation, location, parking, contact details
• FAQ-level product/service questions in EN/AR/KU
• Booking confirmation requests ("can I come at 7?")
• Status checks ("did my order ship?")
• Routine after-sales ("how do I reset the device?")
• Sending automated reminders before appointments

What it should NOT handle alone

• Pricing negotiations
• Complaints with anything more than mild frustration
• Anything legally binding
• VIP customers (route to a human immediately, every time)
• Anyone who has typed in three messages without resolution

The rule I bake into every system: the AI handles questions, the human handles emotions and money. If a customer's message contains a hint of either, it gets escalated, transcribed, and tagged for a human.

The two-week setup

Week 1

• Day 1–2: Apply for WhatsApp Business API via Meta Cloud (1–3 day approval).
• Day 3–4: Document your top 30 most-asked questions and your house's specific tone of reply.
• Day 5–7: Build the AI agent — system prompt, tool-use for booking lookups, escalation logic.

Week 2

• Day 8–10: Train on your real conversation history (anonymized). 40–60 example conversations is the sweet spot.
• Day 11–12: Internal pilot — your team uses it, breaks it, you fix the edges.
• Day 13: Go live for one customer segment (e.g. new bookings only).
• Day 14: Full rollout, with monitoring and easy human override.

Real cost numbers

Build: $1,800–$3,500 depending on integrations (your booking system, your CRM, payment gateway). Monthly run: WhatsApp Business API $40–$100 + LLM API $30–$150 + hosting $0–$25. Total monthly: $70–$275 at typical SMB volume.

Compared to: one $400/month WhatsApp-watching staff member. The AI handles 70–90% of messages and is awake at 2 AM. Even if the human stays (and they should — for the 10–30% that's actually valuable), the math swings hard in favor.

Kurdish in WhatsApp specifically

Modern LLMs handle Kurdish well enough on WhatsApp because messages tend to be short, transactional, and full of recognizable patterns. The hard part isn't language understanding — it's matching tone. A formal "تكاتە / تفضل" reply from an AI in a casual shop reads as sterile. I solve this by capturing 6–10 example conversations from your actual best staff and few-shotting them into the prompt. That single step does more for quality than any fine-tuning would.

Five rules I always enforce

1. Always offer to escalate to a human after one back-and-forth without resolution.
2. Never invent prices, opening hours, or product details — read from a database, not from the model's memory.
3. Log every conversation, audit weekly for the first month.
4. The owner gets a daily digest of what the AI handled and what it escalated.
5. One-tap "this AI got it wrong" feedback loop wired into the system from day one.

If you want to ship this

Send a brief on the contact page with: what business you run, your WhatsApp Business volume per week (estimate), and the single most-asked question your team gets. With those three, I can give you a fixed price within a day.