Most advice about Meta ads right now says the same thing: turn on broad targeting, give Meta no hints, and let the system find your buyers. I have spent almost $3M on Meta ads over seven years, across online stores, online courses and lead generation, and I still start a new account with interests. Broad comes last in my plan, and it gets a test budget, not the main one.
Below is the order I build audiences in, how I split the budget between them, and why broad targeting did not work for me on anything more expensive than an impulse buy.
Why I don't start with broad
With broad targeting you set only the country, age range and gender, and Meta decides who sees the ad. For a cheap product that people buy on impulse, that can work: almost anyone scrolling past could be a buyer.
For anything pricier, the buyer is a narrower group of people who already care about the topic. I tested broad on products like that, and it didn't work for me. I wasn't surprised: the ads went to everyone, with no filter on who saw them. On those products Meta needs a hint about who is interested, and interests and lookalike audiences give it that hint.
How I split the budget
I keep 80% of the budget on cold audiences, people who haven't met the brand yet, and 20% on warm audiences, people who already visited the site.
- 40% on interests. This is where every new account starts.
- 30% on lookalikes, built from people who already bought.
- 10% on broad, as a test.
- 20% on remarketing: site visitors, people who added to cart and people who started checkout.
Step 1: interests
An interest is a topic Meta links to a person based on what they follow, like and click: interior design, running, home cooking. You pick interests that match what your buyer cares about, and Meta shows the ad inside that group.
I start with the most obvious interests for the business, the ones a buyer of this product would clearly have, and I don't add many. Even a campaign with a single interest can sell well. You can test different interests in separate ad sets to see which one brings sales, or put up to five related interests together in one.
Step 2: lookalikes
A lookalike audience is a group of people Meta finds because they resemble a list you give it. The quality of the lookalike depends on the quality of that list, so I build mine only from people who already paid or came close to paying:
- the store's customer list, uploaded from the store;
- top customers and customers with the highest order value;
- people who bought in the last few months, from the Meta pixel;
- people who started checkout.
Most of my lookalikes are 1%, the group closest to the original list. A 1% lookalike built from people who already bought is usually one of the best audiences in the account. Meta's own description of how lookalikes work is on its About lookalike audiences page.
Step 3: broad, as a test
Broad gets 10% of the budget and runs next to interests and lookalikes as a test. It didn't work for me on products people don't buy on impulse, so in my accounts it stays a test.
The 20% on remarketing
Remarketing shows ads to people who already know the store: they visited the site, added a product to the cart, or started checkout and left. For stores with a product catalog, I also run dynamic remarketing, where Meta shows each person the exact products they looked at.
These are the people closest to buying, so they get their own 20% instead of competing with cold audiences for the budget.
Two campaigns I keep apart
Next to the audience split, I keep two types of campaigns separate.
The first is a catalog campaign. Meta builds the ads from the product feed, with the product photos and prices from the website. It needs no new creatives.
The second is a conversion campaign with my own creatives: images and short videos I make in Higgsfield, with a clear message about one product or one reason to buy. Both campaigns work well for me, and I keep them apart. I wrote about how I make the first frame of an AI creative in this article.
Where those ads run matters too. In my placements article I show which placements brought back the most money and how I keep the weakest ones from taking the budget.
If your Meta budget goes to broad and the sales don't come, book a consultation and I'll look at your audiences with you.
My tip: A lookalike built from an uploaded file, like a list of leads or customers, doesn't update itself. When the list grows, upload a fresh file and build a new lookalike on it, so Meta looks for people like your latest buyers.
