AI customer service on WhatsApp: the small‑business playbook

Start with the questions you already get

Export a week of WhatsApp chats and count. In most small businesses, 70–80 % of messages are the same twenty questions: hours, prices, location, availability, “do you have X”, “how do I pay”, “can I change my booking”. Those are the AI’s job from day one. The remaining 20 % — complaints, unusual requests, negotiations, anything medical, legal or financial — stay with people, and the AI’s job there is to hand over quickly and politely.

Teach it like a new employee

• Facts: your website, price list, menu, brochure PDF, a Google Sheet of rooms or stock. In iChatUp you paste or upload; it builds a structured business memory you can edit.
• Policies: cancellations, deposits, delivery areas, what you don’t do.
• Tone: formal or warm, short or detailed, which languages, how to sign off.
• Boundaries: topics that must always go to a person.

Then test it the way you would test a new hire: message the number with the twenty questions and a few curveballs.

Accuracy: the only metric that matters

A fast wrong answer is worse than a slow right one. Demand three things from any AI on your number: it answers only from your facts; it says when it doesn’t know and flags the gap to you; and you can see why it answered and correct it once so the mistake never repeats. iChatUp fact‑checks every reply against your business memory before sending, runs a nightly accuracy exam against your own facts and shows the score on your dashboard, and turns an owner’s correction in the chat into a permanent fact.

Escalation rules that keep customers calm

1. Complaints → a named person, immediately, with the AI saying so in the chat.
2. Anything sensitive (health, legal, money disputes) → person.
3. “I want to talk to a human” → person, no arguing.
4. Three unanswered turns → person.
5. High‑value leads → person, with the conversation summary attached.

And the reverse: when the person is done, the AI resumes. One tap each way.

Languages

Customers write in whatever language they think in. The AI should answer in that language while keeping your facts and tone — English, Hindi, Malay, Arabic, Spanish, Tamil, Bahasa, dozens more — and it should never switch language mid‑conversation. Keep your business facts in one language; the model translates on the way out.

The 30‑day rollout

• Days 1–3: connect the number, load facts, test in practice mode with no customers involved.
• Days 4–10: go live after hours only. Review every conversation each morning; correct wrong answers in one tap.
• Days 11–20: go live all day. Watch the escalations and the “didn’t know” list; fill the gaps.
• Days 21–30: add actions — bookings, payment links, follow‑ups. Check the accuracy score and the time your team spends on WhatsApp.

What not to do

• Don’t let it pretend to be a person.
• Don’t give it actions without rules (holds, verification).
• Don’t broadcast from a phone‑linked number.
• Don’t run two tools on one number.
• Don’t set it and forget it — ten minutes a week keeps it accurate.