What happened to us
In September we ran an ad to sell websites. In one week, 14 people messaged us wanting to buy a laptop, and Meta's dashboard marked that week as the best of the month: conversations had gone from 29 to 63.
It wasn't Meta's mistake. It was ours, and it's one of the most common ones. What follows works for any business that advertises to get messages on WhatsApp, Instagram or Messenger.
Why it happens: Meta optimizes for what you ask
When you set up a campaign you choose the result you want: messages, leads, visits, sales. Meta looks for the people most likely to do that, at the lowest possible cost. If the result is "conversations", a "price?" from someone who will never buy is worth the same as a message from your best customer.
It also learns from who responds. If the first people who message are the wrong ones, it looks for more people like them. That's why the problem tends to get worse over the weeks, and why the cost per message drops: people who message easily are cheap.
And targeting matters less than it seems today. With Meta's automatic audiences (Advantage+), what you set works as a suggestion, and the system is guided mostly by who reacts to the ad. What decides most who sees your ad is what the ad shows.
Signs it's happening to you
- Cost per message goes down and sales don't go up. A record week with no sales is a warning sign, not good news.
- Chats start with "price?", "info" or "is this still available?" and don't answer the first question you ask them.
- People message you from other areas, other industries, or about something you don't sell.
Meta's dashboard counts messages; what those messages say is in your chats. Read the last 20 the ad brought in and sort them into "fit" and "not a fit". That alone tells you which of these cases you're dealing with.
How to fix it
1. Look at your ad like someone who doesn't know you
Whoever scrolls past your ad decides in one second what you sell, and they decide based on the image. Check three things:
- What the image shows. That's what they think you sell. If you offer a service, show the result (the finished haircut, the painted house, the working website), not the tool.
- What the price looks like. If it matches what something very well known costs, people will message you about that other thing.
- What the first line says. It should say what you do, for whom and where. "Sculpted nails in Palermo, by appointment" filters; "Book your appointment" doesn't.
2. Filter before they message you
A lead that isn't a fit costs you money and time. It's worth losing some easy messages so the good ones come in.
- Say in the ad what it leaves out: the area you cover, your starting price (if you can say it), your minimum, who it isn't for.
- If the ad opens WhatsApp, use the greeting message or the FAQ to ask what you need to know: "What do you want to do and where are you?". Someone who won't answer that was unlikely to buy.
- If you use Meta's lead forms, choose the higher intent type: it adds a step where the person reviews their details before submitting. Fewer come in, but they're more serious. Add one or two questions of your own.
3. Measure leads that fit, not messages
The number that matters isn't what each message costs: it's what each lead that fits costs.
- Label every chat as "fit" or "not a fit". WhatsApp Business has labels for that.
- Once a week, divide what you spent by the number of "fit" chats. That's your real cost.
- Watch what share of leads fit. If that share drops, the ad is drifting, even if the dashboard says everything is fine.
In our case, with the corrected ad, the leads that came in with a specific request went from 38% to 53%, and the laptop requests dropped to zero.
4. If it already learned wrong, start from scratch
When an ad set has already learned to look for the wrong people, changing the photo isn't enough: what it learned is still there.
- Pause that ad set and create a new one, with the corrected ad.
- Give it time. Meta says an ad set needs around 50 results in a week to exit the learning phase, and every significant change puts it back into learning. Touching it every day keeps it learning forever.
5. Tell Meta who bought from you
It's what changes results the most and what people do the least. If Meta knows which of those conversations ended in a sale, it can look for people like your customers instead of people who just message.
- For ads that open WhatsApp or Messenger, there's the Conversions API for Business Messaging. Your system tells Meta when a chat ends in a purchase (or a qualified lead), and the campaign can optimize for purchases instead of conversations.
- It needs two things: WhatsApp connected through the API (not the regular app) and a system that records how each chat ended. A CRM connected to WhatsApp can do it on its own.
How we keep it under control
After the laptop episode we built our own system, which we use on our campaigns today:
- It reads the chats the ads bring in and flags which ones come with a specific request.
- It measures the cost per useful lead for each ad, every day.
- It alerts us when an ad drifts, as soon as the wrong leads start coming in, before it shows in sales.
If you want something like this for your business, it's part of the custom systems we build.
