Audiences, and how little they matter now

For years the skill on this platform was audience building: stacked interests, lookalikes at every percentage, exclusions layered carefully. That skill has largely been automated away, and advice written before the change is still everywhere.

What changed

Delivery systems now find the people likely to take your action, given enough conversion signal to learn from. Narrow manual targeting mostly restricts that search without improving it. Broad targeting plus a clear conversion signal usually beats carefully assembled interest stacks.

The consequence for structure is large. The old approach was many ad sets, each with a different audience, each competing for the same people and splitting the data between them. The current approach is fewer ad sets, more budget each, and let the system learn.

Why splitting hurts, with numbers

Illustrative, to make the mechanism visible. Delivery needs a run of conversions before it stops guessing, and the figure usually quoted is around fifty a week per ad set.

Suppose your budget produces thirty conversions a week in total. In one ad set, that is thirty: short of the threshold, but close enough that delivery is learning something. Split across five ad sets, each gets six. Not one of the five ever leaves the guessing phase, so all five perform badly, and the conclusion drawn is that the channel does not work.

The same budget, the same creative, the same audience, and the only difference is how many boxes you put it in. Consolidation is not laziness; it is the difference between one ad set that can learn and five that cannot.

What targeting still earns its keep

• Your own data. People who visited, who bought, who are on your list. These are facts, not guesses, and they remain the most valuable targeting you have.
• Exclusions. Not showing a first-purchase offer to existing customers is still your job, and the system will not infer it.
• Hard constraints. Country, language, age where there is a legal or practical reason. These are filters, not optimisation.
• A genuinely narrow niche. If your market is a few thousand people with a specific professional title, telling the system that saves it time.

Notice what is absent: stacked interests assembled from guesswork about what your customer also likes. That was a proxy for information the system now has directly.

Building the data worth targeting

Since your own data is the valuable part, it is worth being deliberate about accumulating it rather than discovering later that you have none.

1. 01Install the tracking before you need it. An audience of past visitors only contains people who visited after it existed, so every week without it is a week of audience you cannot get back.
2. 02Build the audiences you will want later now: everyone who visited, everyone who reached the offer page, everyone who started and did not finish, everyone who bought.
3. 03Upload the lists you already have. Past customers and email subscribers are the highest quality input available, and most small businesses never upload them.
4. 04Keep the exclusion list current. Existing customers seeing a first-purchase offer is the most visible waste in this channel.

A structure that works

1. 01One campaign per objective, optimising for the thing you actually want rather than the step before it.
2. 02Few ad sets. Often one broad, plus one for your own data, plus exclusions. Add more only when you can explain what each one is for.
3. 03Several creatives per ad set, representing different angles rather than different crops of one idea.
4. 04Enough budget per ad set to produce conversions regularly. If it cannot, consolidate until it can.

If the automated campaign types are available for your objective, they are worth testing against your manual setup rather than assumed better or worse. They are the same bet as above taken further, and the only honest way to know is to run both.