A seller I was helping had her phone face-down on the counter because the buzzing was making her anxious. Sixty-one unread messages on WhatsApp. She opened the first one: "Do you have this in a 38?" The second one, same. The fourth, the seventh, the eleventh: same question, different people.
She typed the same answer eleven times. By the time she reached message forty, a woman who had asked about a jacket an hour earlier had already bought one somewhere else and stopped replying.
That afternoon is the whole problem in miniature. WhatsApp isn't hard. The trouble is that a small handful of questions arrive over and over, all day, and answering them by hand is the thing standing between you and the orders sitting further down the list.
So let's talk about what WhatsApp FAQ automation actually is, and what it isn't.
Your inbox isn't 500 different questions. It's about 20.
Every seller I've watched believes their chats are chaotic and unpredictable. Then we count.
Price. Is it in stock. What sizes. Shipping cost. How long does delivery take. Do you have other colors. Is this the original. Can I see it on. Do you have a store I can visit. How do I pay. Can I return it. Is there a discount if I take two.
That's twelve, and I wrote them from memory. Add eight more that are specific to your category (fabric and washing for clothes, warranty for electronics, ingredients and expiry for cosmetics) and you have the list that covers roughly four out of five messages you receive. The rest are the real conversations: the hesitant buyer, the complaint, the person who wants something custom.
FAQ automation means the first eighty percent stops touching your hands, so you have energy left for the twenty percent that actually needs a person. That's it. It's not a robot pretending to be you.
Finding your real twenty takes one evening
Don't guess this list. Guessing produces the questions you wish people asked. Pull the real ones:
- Open WhatsApp and scroll back through the last 100 to 150 incoming first messages. Not whole conversations, just the opener each person sent.
- Write each one down in a note, one line per message, in the customer's own words. Don't tidy the wording yet.
- Group the duplicates and put a number next to each group.
- Sort by count. Stop at the point where a question appears fewer than three times.
You'll usually end up with 15 to 25 questions, and the top three will be so far ahead of the rest that it's almost funny. Keep the raw customer phrasing next to each one. "How much is it?", "price?", "قیمت", and a bare photo of your product with a question mark are all the same question, and your automation needs to recognise all four.
Three answers decide whether there's a sale
Not all twenty questions are equal. Three of them carry the money, and they're almost always: price, availability, and how fast it arrives.
When one of those goes unanswered for an hour, the customer doesn't wait politely. They go looking. I've written before about the three unanswered DM questions that quietly kill sales, and the pattern on WhatsApp is identical, just with fewer emojis.
So when you set up your answer bank, treat those three as non-negotiable. Every product, every variant, needs a current price, a true stock status, and a delivery window you can actually keep. If you don't know your own delivery window, your automation will confidently lie for you, at scale, which is worse than silence.
Keyword rules break the second a human types like a human
Most "auto reply" tools work on triggers. If the message contains the word "price", send message A. If it contains "shipping", send message B.
This works beautifully in a demo and falls apart in a real inbox. Customers write "how much is the green one", "still available?", "does this come in 42 and how much", and "hi 🙂". The last one is the most common opener in the world and it triggers nothing.
Worse are the compound messages: "Hi, is the beige bag still there, what's the price with shipping to Shiraz, and can I pay on delivery?" A keyword rule fires one canned reply and ignores two-thirds of the question. The customer now has to ask again, which is exactly the friction you were trying to remove.
Real FAQ automation reads the whole message, understands that three things were asked, and answers all three from your product data in one reply. That's the difference between a rules engine and an assistant. My longer take on where simple auto-replies help and where they cost you money is in this breakdown of WhatsApp auto reply for business.
Build one answer bank, not forty canned replies
Here's the mistake that makes automation collapse after three weeks: people write forty separate reply templates. Then prices change, and thirty-one of those templates are quietly wrong.
Build a single source of truth instead. One place that holds:
- Products, with price, sizes, colors, and current stock
- Shipping: cost by city or zone, cutoff time, realistic delivery days, what happens on holidays
- Payment: which methods you accept, whether cash on delivery exists and where
- Policies: returns, exchanges, warranty, what counts as a valid claim
- Store facts: address if you have one, working hours, whether people can visit
Then the assistant composes answers from that, in your tone, instead of reciting frozen paragraphs. Change a price in one place, and every future answer is right. Do this and your maintenance job drops from "rewrite forty templates" to "keep one list current", which is a job you'll actually do.
Tone matters more than sellers expect. If you normally answer with two short lines and a "🙏", an assistant that replies in formal paragraphs will feel like a stranger took over your account. Give it three or four of your own real replies as examples and it will sound like the shop, not like a help desk.
The questions you should never hand to a bot
Automation earns trust by knowing where it stops. Route these straight to a human, every time:
- Anyone who is angry, or whose order has already gone wrong
- Refunds, damaged items, and anything involving money going backwards
- Custom or bulk orders where the price is negotiated
- Health, medical, or safety claims about what you sell
- Any message where the assistant is not confident about the facts
That last one is the important one. An assistant that says "let me check with a colleague and get back to you in a few minutes" costs you nothing. An assistant that invents a delivery date costs you a customer and a review. If your tool can't be configured to escalate on low confidence, it isn't ready for your inbox.
Also: keep the handoff invisible to the customer. Same chat, same thread, no "an agent will join shortly, ticket #4471". Nobody wants a ticket number from a shop that sells scarves.
How to tell if it's actually working
Four numbers, checked weekly. That's enough.
First reply time. This is the one that moves sales, and the gap is brutal: a person answering by hand averages tens of minutes, an assistant answers in under ten seconds. The whole argument for automation lives in that gap.
Deflection rate. What share of conversations ended without you touching them. Somewhere between 50% and 75% is healthy for a product shop. If it's over 90%, check whether it's answering things it should be escalating.
Handoff quality. When it does pass a chat to you, is the reason sensible? Read ten of these a week. This is where you find your next fix.
Questions with no good answer. Every "I'm not sure about that" is a hole in your answer bank. Fill one or two a week and within two months the holes are gone.
One more thing worth watching: if you're growing past a few hundred conversations a day, the app on your phone becomes the bottleneck, not your answers. That's the moment to read about when to move to the WhatsApp Business API.
Start with three questions, not twenty
You don't need the full system on day one. Take your top three questions, get those answers exactly right, and let automation handle only those. Watch a week of real replies. Fix what sounds off. Then add five more.
Sellers who try to automate everything on the first weekend usually turn it all off by Wednesday. Sellers who start with three questions are still running six months later, with fifteen automated and a quieter phone.
If you'd rather not wire this together yourself, Vardast connects to your WhatsApp, learns your products, prices and policies once, and answers the repetitive questions in your own voice around the clock, passing the real conversations to you with the full history attached. Your top twenty questions stop being your job, and the customer who asked at 2am gets an answer at 2am.
The buzzing phone on the counter isn't a sign of a busy business. It's a sign that twelve people are all asking what size 38 costs. Answer that once, properly, and let the machine repeat it.