The audience-building skills that defined Facebook advertising are now mostly obsolete. Here is what replaced them.

What actually changed

Two things, both structural.

First, signal loss. Interest and behavior targeting depended on rich third-party observation of what people do across the internet. Privacy changes removed much of that, so the definitions became blurrier at exactly the moment advertisers were relying on them most.

Second, and more importantly, the delivery systems got better. Given a broad audience and a clear conversion signal, they now find converters more effectively than a human specifying who those converters should be. The human is guessing from a persona; the system is inferring from behavior at a scale no person can hold in their head.

What still matters in audience strategy

Broad is not the same as careless. Several audience decisions remain genuinely important.

• Exclusions. Keeping recent purchasers out of acquisition campaigns, and keeping retargeting audiences out of prospecting, remains basic hygiene and is frequently neglected.
• Geography. Where you can deliver, where you can deliver profitably, and where your return rates are acceptable are three different maps.
• Language, where it materially affects the offer.
• Customer lists for retention and for suppression — your own first-party data is more valuable now, not less, precisely because third-party signal degraded.
• Age and gender bounds where the product genuinely requires them, rather than where a persona document suggests them.

Notice that most of these are constraints imposed by your business rather than guesses about who your customer is. That distinction is the whole shift.

Are lookalikes still useful?

Less often than they were, and the seed matters more than the percentage.

A lookalike built from a large, high-quality seed — your best customers by lifetime value rather than everyone who ever purchased — can still outperform broad in some accounts, particularly where the product appeals to a genuinely narrow segment. A lookalike built from a small or generic seed usually just adds a constraint without adding information.

• Seed quality beats seed size beyond a few thousand people. High-value purchasers beat all purchasers.
• A 1% lookalike on a weak seed is a narrow audience derived from noise.
• Test lookalike against broad honestly rather than assuming either. The result differs by account and by category, and it changes over time.

Rebuilding an account built the old way

If your account has fifteen ad sets with layered interests and a modest total budget, the structure itself is your biggest problem — not because the audiences are wrong, but because none of those ad sets will ever gather enough conversions to stabilize.

1. Consolidate. Fewer ad sets with meaningful budget each, so learning can actually complete.
2. Move the testing effort from audiences to creative. Creative is now the variable that most changes who sees your ad, because it determines who responds.
3. Keep the exclusions. Delete the interest stacks.
4. Run the change as a deliberate transition with a before-and-after read on blended efficiency, not as a big-bang rebuild during a peak trading period.

Expect the first fortnight to look worse. Consolidation restarts learning, and the account has to re-stabilize before the comparison means anything.