You do not choose the audience any more

The panel keeps getting smaller

Anyone who has run Meta ads for a few years remembers building an audience like a recipe. Interest in one thing, behaviour in another, exclude the people who already bought, a lookalike layered on top. The craft was in the stack.

That panel has been shrinking for years. Options get merged, retired, or quietly stop mattering because the delivery system spends outside them anyway — and often does better when it does.

This is not the platform being difficult. It is what happens when the signal that fed precise targeting thins out — app tracking permissions, browser changes, privacy regulation — and gets replaced with the one thing the platform still has in abundance: an enormous model watching who responds to what.

What replaced it

The system decides who sees the ad. You decide what the ad is.

That is the whole shift. It sounds like less control. In practice it moves control somewhere more useful, because the creative is now the targeting.

An ad showing a farmer standing next to the machine in his own field gets shown to people who behave like that farmer. An ad in Marathi finds Marathi speakers. An ad that opens with a price finds people who lead with price. You are no longer selecting an audience from a dropdown — you are describing one in the ad, and the system reads the description.

What this changes about the work

Fewer ad sets, more ads. Splitting a modest budget across eight narrow audiences starves each one of the data the system needs to learn anything. One broad ad set holding genuinely different creative usually learns faster and settles sooner.

Creative volume stops being a nice-to-have. If the creative is the targeting, three ads means three audiences. It is also why creative fatigue costs more than it used to — a worn-out ad is now a lost audience, not just a tired image.

Exclusions still earn their keep. The one place manual control clearly still pays is telling the system who not to bother with: existing customers, recent buyers, everyone already sitting in your pipeline.

Your conversion event becomes the steering wheel. The system optimises toward whatever you have declared to be success. Declare a form submission and it will find you people who submit forms — which is precisely how you end up with leads nobody can close.

The part nobody enjoys

Automation is very good at finding more of what you told it worked. It has no opinion about whether that thing was worth finding.

So the quality of the instruction matters more than it ever has. Give it a weak success event and it will pursue that weak event with real efficiency, and the report will look excellent while the sales team wonders what happened. Give it a genuine one — a qualified enquiry, a booked showroom visit, a sale the dealer has confirmed — and the same machinery starts working for you instead of against you.

The lever moved. It did not disappear.

Where the manual work went

It relocated, roughly like this:

Two of those three sit outside the ads manager entirely. That is the honest summary of where this has gone, and it is why our advertising work starts with what the ad says long before anyone opens a targeting panel.

• Before the campaign — the offer, the creative, the language, the landing page
• Inside the campaign — exclusions, budget structure, the conversion event
• After the campaign — reading which creative produced business, not which produced cheap clicks