Meta's automated campaign types took over most of the targeting decisions advertisers used to make by hand. The common reading is that this made advertising simpler. The more useful reading is that it moved the work.
If the system picks the audience, the placement and the budget split, then creative is the remaining variable you control — and the system's performance depends on having enough of it to choose between.
What the automation actually does
Roughly: you supply creative, a budget and a conversion goal; the system decides who sees what, where, and how much is spent on each combination.
Two consequences that matter more than the feature list:
Detailed interest targeting mostly stopped being your job. The system is better at finding the buyer than a hand-built interest stack. Time spent on audience construction is largely time wasted now.
The system needs options. It optimises by allocating spend across what you gave it. Feeding one video into a mechanism designed to choose between many is using a fraction of it.
Which means: more creative, not less
This is the counter-intuitive part, and the part most people get backwards.
Automation was sold as doing more with less. In practice the constraint moved from how well can you target to how many genuinely different creatives can you produce. Accounts that struggle under automated campaigns are usually creative-starved, not badly configured.
A reasonable floor is five distinct creatives per campaign, refreshed as they fatigue. "Distinct" is doing work in that sentence — see below.
What counts as different
The system can only choose between things that actually differ. Five edits of one video are one creative with cosmetic variation.
Differences that matter, in order of impact:
- The hook — the first three seconds. Most of the spread in outcomes lives here.
- The angle — which problem the ad claims to solve.
- The format — talking head, faceless demonstration, before-and-after, static with text.
- The presenter — who is on screen, or nobody.
- The setting.
Differences that mostly do not matter: music, colour grade, caption font, the order of the middle section.
A useful discipline: if you cannot describe the difference between two creatives in a sentence a customer would understand, they are not different.
How testing changes
Under manual targeting, you isolated variables: same creative, different audiences. Under automation you cannot — the system reallocates continuously, so a clean split test of audiences is not available and not the point.
What testing becomes:
- Give the system variety and let it allocate. Spend distribution is itself the result; the creative taking spend is the one winning.
- Read the creative-level breakdown, not just the campaign total. The campaign number tells you whether it works; the breakdown tells you why.
- Judge on spend share and cost per outcome together. A creative taking 60% of spend at an acceptable cost is the winner, regardless of what you expected.
- Kill slowly. The system needs a learning period. Pulling a creative after a day teaches you nothing and disrupts the optimisation.
What stops being useful: elaborate split tests of audience segments, narrow lookalike stacks, and most of the structural work that used to constitute campaign management.
The failure mode
The common pattern in struggling accounts: automated campaign, one or two creatives, performance declines, more time spent on settings.
The settings are rarely the problem. A campaign with two creatives has two options to allocate between, and when both fatigue there is nowhere for the system to go. The fix is upstream — more distinct creative, refreshed before the existing set exhausts itself.
This is uncomfortable because creative production is slower and more expensive than changing settings, which is why people keep changing settings.
Making the volume possible
The arithmetic: five distinct creatives per campaign, refreshed every few weeks, across however many products you run. At traditional production costs that is not feasible for a small seller, which is the actual reason accounts end up creative-starved.
This is the narrow case generated creative fits, and it is worth being precise about why. It does not make creative better. It makes the cost of an additional variant nearly flat — the fifth hook costs about what the first did. That turns the volume requirement from a budget problem into a scheduling one.
Faysell produces several variants from one product photo, with the hook as the thing you vary. The Playground covers one-off clips when a set needs one more angle rather than a new ad.
Once a creative is carrying serious spend and fatiguing, the economics flip and commissioning real production becomes the right purchase. The mistake in both directions is common: paying for production before knowing what to produce, or still running test-grade creative on something doing real volume.
A practical setup
- One automated campaign per product or clear product group.
- Five distinct creatives, differing at the hook or angle level.
- Correct aspect ratios for the main placements — 9:16 and 4:5 at minimum. Reframe rather than letting the system crop.
- Catalogue connected, if you sell physical products.
- Leave it alone through the learning period.
- Read the creative breakdown weekly, not the campaign total.
- Replace the bottom two with new angles as they fade.
The summary
Automation did not reduce the work; it moved it from targeting to creative. The system allocates across what you give it, so the binding constraint is how many genuinely different creatives you can produce and refresh.
Vary the hook hardest, judge by which creative the system chooses to spend on, and when performance declines, look at your creative supply before you look at the settings.