Corks Out: WhatsApp AI Handles 70-80% of Restaurant Replies

Corks Out is a wine bar and chargrill restaurant in Malaysia. Before chatavocado, guest enquiries arrived by phone and WhatsApp and got written into a notebook by hand. Now AI handles 70 to 80% of their replies. The floor staff stopped watching the phone and went back to serving.

Stephanie speaks in Mandarin. English captions are in the subtitles menu.

About Corks Out

Corks Out is a Malaysian wine bar and chargrill restaurant. Stephanie runs marketing there. In her words, the concept the business leads with is chargrill, which means the kitchen and the floor are the product. Nobody is hired to sit and answer messages.

"然后Corks Out主打的概念是Chargrill"
"The concept Corks Out leads with is chargrill."

That matters, because it explains why message handling never got a proper owner. In a restaurant, every pair of hands is already assigned to something that happens in front of a guest.

The old-fashioned way

Stephanie describes the before state without dressing it up.

"So before我们用AI的时候呢，我们是用一个比较老派的方式"
"Before we used AI, we did it in a fairly old-fashioned way."

Concretely: a guest calls, or texts the WhatsApp number. Somebody writes it down.

"就是比如customer他们call过来，或者是他们text我们Whatsapp，我们就用manually的，可能用书写啊，或者是我们的outlet manager会用电话还是记录这样子起来"
"For example, customers call, or they text us on WhatsApp, and we do it manually, maybe writing it down, or our outlet manager takes the call and records it that way."

This is the part most restaurant operators recognise instantly. The notebook is not a system. It is what you use when there is no system and service has already started.

What breaks when the phone is your ordering system

A paper notebook next to a ringing phone fails in three specific ways, and none of them show up in your P&L as a line item.

The enquiry arrives during service. The people qualified to answer are the people carrying plates.

Nobody knows what was missed. Stephanie names this one directly. Her staff had to keep watching the phone, checking whether a customer message had slipped past.

Read that again. The cost was not the replying. The cost was the constant checking, carried by someone who was meant to be looking at a table.

Handwriting does not survive a shift. A booking in a notebook exists in one place, in one person's handwriting, and it leaves when they do.

---

Your busiest hour is also your worst hour for replying to guests. See how chatavocado answers during service →

---

AI now handles 70-80% of replies. What that actually measures

Plenty of AI vendors will quote you a percentage and hope you do not ask what is in the numerator. So here is the exact claim Corks Out makes.

"我们目前AI能帮助我们reply的almost like有70-80%都是AI可以帮助到我们"
"Right now, AI can help us with almost 70 to 80% of our replies."

It is a share of replies. Not time saved, not money saved, not satisfaction. Those numbers are easy to inflate because nobody can check them. Share of replies handled without a person typing is the one figure a restaurant can verify from its own inbox on any given Friday.

The remaining 20 to 30% is where a human should still be. Large group bookings, complaints, wine questions that deserve a sommelier's answer, anything with money attached. A split like that is a design decision, not a shortfall.

What the floor staff do with the attention back

The payoff Stephanie describes is not a cost line. It is what her team is now free to be good at.

"至少我们的floor staff他们可以再more专注on他们on service，maybe food knowledge，wine knowledge，instead of他们每次要需要一直看着电话诶，有没有什么客户的message漏掉啊"
"At least our floor staff can focus more on service, maybe food knowledge, wine knowledge, instead of having to keep watching the phone every time, checking whether any customer message got missed."

For a wine bar, that trade is a direct revenue argument. A server who knows the list sells the second bottle. A server refreshing WhatsApp between courses does not.

She frames the whole thing as two problems solved at once.

"所以这个AI就可以帮忙解决掉我们staffing的问题，时间上的问题"
"So the AI helps solve our staffing problem and our time problem."

The staffing framing is the honest one. Corks Out did not hire a person to answer messages, and now they do not need to. This is the same shape as We Are Caring's 60% jump in bookings, where the constraint was never lead volume. It was the friction between a customer wanting something and a human being available to confirm it.

Why restaurants are a harder case than most

An office-hours service business has a fair shot at replying to everything. Somebody sits at a desk. Restaurants do not get that.

Your enquiries cluster around service, which is precisely when every staff member is committed to a physical task. A dedicated messages person rarely pencils out for a single outlet. And unlike a clinic or an agency, a slow reply is not a delayed sale, it is a lost table, tonight, to the restaurant that replied first.

This is why fast replies convert so hard in F&B. We have written separately on why response speed moves bookings, and the effect is hard to miss in a business that sells a table for one sitting.

Stephanie's recommendation is aimed squarely at the operators still in the notebook stage.

"如果遇到好像别的朋友啊，也同样是开餐厅，他们还是用着非常老式的方法去reply message，我会非常recommend他们去try这个chatavocado.ai，which可以帮助到他们省很多时间，然后同样可以帮助到他们员工可以有更好的效率去服务他们的客人"
"If I meet other friends who also run restaurants, and they are still replying to messages the very old-fashioned way, I would strongly recommend they try chatavocado.ai, which can help them save a lot of time, and equally help their staff serve their guests more efficiently."

She does say time saved there, and that is worth separating from the number. The 70-80% counts replies. The time saved is what she reckons it bought her, which is a different kind of claim and she is entitled to make it about her own restaurant.

How to set this up for your own restaurant

Four things, in order, and the first one is the only one that takes real thought.

1. Write down the twenty questions you answer every week. Opening hours, parking, corkage, minimum spend, whether you take walk-ins on a Saturday, halal status, private room capacity. That list is your AI's job description, and it is almost always shorter than operators expect.

2. Move your WhatsApp number onto the API. The consumer WhatsApp Business app cannot be automated properly and cannot be used by two people at once. Our migration guide covers keeping your existing number and chat history.

3. Decide your handover rule before you launch, not after. Groups above a certain size, anything involving a refund, anything where the guest sounds unhappy. Write the rule down and the AI will follow it.

4. Check the transcript for the first fortnight. Read what the AI got wrong and feed it back. The 70-80% at Corks Out is where they landed, not where they started.

On cost: you pay Meta's per-message rate, and we pass that through at zero markup, which we explain in our pricing stance. The full breakdown for Singapore and Malaysia operators sits in our WhatsApp Business API pricing guide. For a single outlet, it lands under what you currently pay in part-time hours for someone to do the same job worse.

Get your floor staff off the phone

Corks Out did not buy an AI product. They removed one job from their floor team and gave it to something that does not take breaks, does not quit, and does not lose a booking in a notebook.

If your team is still checking the phone between courses to see what got missed, that is the same problem, and it has the same fix. See how LinkTrade cut invoice work by 90% or how Zoom Academy uses WhatsApp broadcasts for the same pattern in other operations.

Try chatavocado for your restaurant →

Frequently asked questions

What does the 70-80% figure at Corks Out actually measure?

It is the share of guest replies the AI handles without a staff member typing them. It is not a claim about time saved or money saved. Stephanie's words were that AI helps with almost 70 to 80% of their replies.

Can WhatsApp AI take restaurant reservations and enquiries?

Yes. The AI answers opening hours, menu, location and availability questions the moment they arrive, then collects booking details in the chat. At Corks Out, those enquiries used to be written down by hand while someone was on the phone.

Does the AI reply in Mandarin, Malay and English?

Yes, and it matches how the guest writes. Malaysian and Singaporean guests often mix languages in one message, and the AI replies in the same mix rather than forcing everyone into English. We Are Caring uses this across their whole enquiry base.

What happens to enquiries that come in during service?

They get answered anyway. That is the point. Dinner rush is exactly when nobody is free to check the phone, and it is also when the most enquiries arrive.

Do I need to replace my POS or booking system to use this?

No. The AI sits on your WhatsApp number, which for most F&B businesses is already the busiest channel. Your POS, roster and reservation book stay where they are.

How much does WhatsApp AI cost for a restaurant in Singapore or Malaysia?

You pay Meta's per-message rate plus a monthly platform fee. For a single outlet that is usually well under the cost of the part-time hours currently spent replying to messages. See our WhatsApp Business API pricing breakdown for the full numbers.