Case study

Nobody gets left on read.

A Lebanese cash-on-delivery shop. Twenty-one days of Meta ads into WhatsApp, and every single conversation read by us, one at a time. This is what the phone was hiding and what changed.

Client details withheld. Every number below comes from the live conversation log and the shop's own order sheet.

671Conversations reached the shop's number
$607Ad spend over the 21 days
0.4 minMedian first reply once the agent was on
10Sales closed start to finish with nobody typing
The situation

One phone. One owner. Every customer at once.

The shop sells small everyday products at $12 a pack, paid in cash at the door. Every order starts the same way: someone sees an ad on Instagram or Facebook, taps the button, and lands in a WhatsApp chat with the shop. The owner answered all of it herself, from the WhatsApp Business app on her phone, in between packing orders and running the rest of the business.

The ads were working. The chats were not. A quarter of first messages arrived outside nine in the morning and nine at night, and even in the day a customer could wait an hour for a price. By the time the answer came, most of them had moved on.

What we found

Most of the "conversations" were not conversations.

Meta reported 1,154 conversations started. Only 671 reached the number, and 397 of those never said a word beyond the button's canned opener.

That is 59 percent of everything the ads paid for, counted as a lead by the platform and treated as one by the shop. Nobody had been able to see this before, because it lived inside a phone where nothing gets counted.

  • The owner's median first reply was 49.7 minutes. One in four waited over 22 hours.
  • Only 26 percent of customers heard back within five minutes.
  • Almost every real message was one of the same twelve questions: price, delivery cost, how many in the pack, is it genuine.
671 people

Opened a chat

Tapped the ad's button. The canned opener went out. That is the whole event.

274 people

Said something real

A price, a delivery question, a yes. The other 397 stayed silent.

221 people

Got a real answer

Beyond the greeting. Before the agent, this number was whatever the phone allowed that day.

48 orders

Matched to an order

On the same number, from the shop's order sheet. Orders are entered in batches, so the true count is higher.

What changed

The agent joined the number. The owner kept her phone.

We connected alongside the WhatsApp Business app, so the number, the inbox and the chat history stayed exactly where they were. The agent was trained on the shop's own past conversations: the real prices, the real delivery times, the way the owner actually writes.

From then on every ad-started chat was greeted within seconds. Questions were answered in whatever the customer wrote in, Arabic, Arabic in English letters, English or French. When someone said yes, the agent asked for their location, confirmed the order, and handed the thread to the owner with everything attached. Voice notes, complaints and anything about an existing order went straight to her without a guess.

Anyone who asked and went quiet got one message before the day was out. Not a broadcast, a reply in the same thread about the thing they asked about.

The numbers

Speed first, because speed is the sale.

First real reply after the customer's first message. Greetings and follow-ups excluded, so this is the answer to the question they asked.

Median first reply
Agent0.4 min
From the phone49.7 min
Answered within five minutes
Agent85%
From the phone26%
Slowest quarter waited
Agent1.4 min
From the phone22 hours
123 Conversations handled with no human involved at all
Nine days 10 Sales closed by the agent alone, greeting to confirmed address
18 More closed by the owner after the agent got the location

The ones who went quiet

109 people asked something and then stopped replying. Each got one message inside the day, in their own language, about what they had asked. 20 of them answered, and 2 of those ordered. Small numbers, but they are people the shop had already paid to reach and would otherwise never have heard from again. The follow-up now reads the whole thread first and never nudges anyone who has already bought.

What it cost

Published pricing, no surprises.

This account runs on the Agent + Tracking tier. The ad spend is the shop's own and goes to Meta directly. Meta's per-message fees for service replies start on 1 October at $0.0091 a message in Lebanon, and we pass them through at cost, in writing.

Ad spend, 21 days$607
Cost per conversation Meta counted$0.53
Agent + Tracking$450/ month
Setup, trained on the shop's own chats$1,200once
What is next

Teaching the ads who actually bought.

Every one of these conversations arrived with the click that caused it. So far 348 click ids have been captured and 326 customers tied back to the exact ad they came from. When a chat reaches a confirmed order, that is posted back to Meta as a purchase, against that ad, and Meta goes looking for more people like the one who bought instead of more people who tap a button and say nothing.

This part is running in dry run today. 43 lead events and 28 purchase events are queued and checked by hand before anything is sent, because a wrong event costs the shop real money. We do not claim a cost-per-purchase improvement until it is measured on live events.

What we are not saying

No revenue total, because orders are entered from the courier sheet in batches and lag the chats by days. No attribution lift, because it has not happened yet. The numbers above are the ones we can stand behind today, and every one of them can be walked through on a call.

Your phone is hiding the same numbers.

A short application, so we only take on accounts we can move. Setup $1,200, then from $250 a month.

Purchase sent to MetaNose Strips Aug 31 · $16
now
Agent closed 3 chatsNo human needed
now
Order written to Shopify#1227 · Rania · Ghazir · $16
now