On this page
Meta's delivery system has become very good at finding buyers. With broad targeting and Advantage+ campaigns, the algorithm does much of the audience work that used to take hours. What it can't do is make your ad more interesting. That's why most accounts that stall today have a creative problem, not a targeting one.
The answer is to test creative on a schedule, not when performance drops. Here's a routine that works for small teams.
The idea: test concepts, then iterate
There are two kinds of creative test, and mixing them up wastes money.
Concept tests compare different stories. Problem and solution versus a customer reaction versus a before and after. These find out what message your buyers respond to.
Iteration tests keep the winning story and change one thing: the hook, the first shot, the call to action, the creator. These squeeze more performance out of an idea that already works.
Start with concepts. Iterating on a weak concept just makes a slightly better weak ad.
A weekly routine
| Day | What to do |
|---|---|
| Monday | Review last week's tests. Pick the winning concept and note why you think it won. |
| Tuesday | Make 3 new concepts for one product, plus 2 to 3 hook variations of last week's winner. |
| Wednesday | Launch them in a testing campaign with an equal budget per ad. |
| Thursday to Sunday | Leave them alone. Early numbers lie. |
That's five or six new ads a week. With traditional production that's unrealistic for a small team. With AI video it's an afternoon.
How to set up the test
- One ad set for testing, separate from your scaling campaigns, so tests don't steal budget from winners.
- Similar budget per ad. If Meta pushes all the spend to one ad in the first hours, the others never get a fair read.
- Enough spend to learn. A common rule of thumb is to let each ad spend at least a couple of times your target cost per purchase before judging it.
- Change one thing at a time in iteration tests, or you won't know what made the difference.
What to measure
Purchases are the goal, but they're slow and noisy at small budgets. Use leading signals to decide faster:
| Signal | What it tells you |
|---|---|
| Hook rate (3-second views ÷ impressions) | Did the opening stop the scroll? |
| Hold rate (e.g. 50% views ÷ 3-second views) | Did the story keep them? |
| Outbound click-through rate | Did the ad make them curious enough to leave the app? |
| Cost per purchase | Did it actually sell? |
A high hook rate with a low hold rate means the opening works and the middle doesn't. A low hook rate means the first two seconds need work, whatever happens later.
Where AI video fits
AI video is ideal for concept tests. You can turn one product photo into three genuinely different stories in minutes, at a cost that makes losing tests cheap. When a concept wins, you have two options:
- Iterate with AI. New hooks, new creators, new languages for other markets.
- Hand the winning brief to a human creator if you want a version with a real face. You now know the story works, so the creator fee buys scale, not a guess.
Keeping ads fresh
Meta shows the same ads to the same people until they stop responding. When frequency climbs and click-through falls, the ad is tired. The fix is not a new audience; it's a new version of the winning concept. Keep a steady supply of fresh versions and fatigue stops being a crisis.
Getting started
Pick one product. Write three concepts. Make them. Run them for a week. That first test usually teaches more than months of guessing.
AdVideoLab's 3 angles button writes three concepts from one product photo, and the free ad script generator does the same from a product page, no account needed. If you'd rather hand the whole thing off, our Meta Ads service runs the testing for you.
More in Strategy
AI UGC vs Hiring Creators: What Each One Really Costs
Creators, agencies and AI all make UGC ads. Here is how the costs actually add up, including the parts nobody puts on the invoice.
AI Video Ads for Dropshipping: A Practical Playbook
How to test products with video before you hold any stock, what to make for each product, and how to know when to kill a test.
Native-Language Video Ads: One Product Photo, Nine Languages
Selling abroad works better when the ad speaks the buyer's language. Here is how to localise video ads properly, and what translation alone gets wrong.