Most creative tests we audit are five new ads with five different hooks, angles, formats and "audience experiments" piled on top. The team calls it variety. It's actually mush. Nothing in the result is attributable to anything you changed.
Ad creative frameworks fix that, not as a library of templates to mimic, but as a stack of layered choices you can isolate, test, and rank by impact. Pick one framework at one layer, hold the others constant, and a $200 test answers what $2,000 of mush couldn't.
This is the Lever Stack: four layers of creative choice, ordered from highest impact at the top to lowest at the bottom. Each layer has a short menu of named frameworks worth testing. Each layer has one rule: settle it before you touch the layer below.
The creative-first hierarchy tells you which layer to test. This post tells you which frameworks live inside each layer.
Takeaways:
- The Lever Stack is hooks, angles, formats, audiences, tested in that order.
- Most ad creative frameworks live in the top two layers: hooks and angles.
- Audiences are the last lever now, not the first.
- One framework per ad, or the test tells you nothing.
Layer 1: Hooks, the thumb-stop frameworks
The hook is the first three seconds. It decides whether anyone reads the rest of the ad. There are four hook frameworks worth running, and almost every winner we've shipped is built on one of them.
- Pattern interrupt - a visual or audio break from what the feed expects (a contradictory image, a smash cut, a slow zoom on something out of place).
- Direct call-out - "If you sell on Shopify and spend over $30k a month on Meta..." Name the viewer in the first second.
- Problem smack - a one-line statement of the pain, no setup. "Your AOV has been flat for six months."
- Promise tease - a result the viewer wants, withheld until the body delivers it. "Here's how we cut CPA 40% in two weeks."
Test one of these against another with the same body underneath. If both hooks pull the same hold rate, your problem isn't the hook, it's the angle below it.
Layer 2: Angles, the argument the ad is making
The angle is what the ad is actually selling. Not the product, the reason to care. Same product, five angles, five different ads.
The angles worth testing as ad angles for ecommerce sit in a short, usable menu:
- Status - "the brand people who know choose." Works for considered-purchase categories.
- Transformation - the visible before/after. Works for skin, fitness, apparel.
- Pain-promise - name the problem, promise the relief. Works for any utility purchase.
- Contrarian - the version of the category that breaks the rule everyone else follows ("we left out the ingredient everyone says you need"). Works when the category is crowded.
- Social proof - the customer's voice, the receipt on camera. Works when the brand is unknown.
Pick one angle per ad. Two angles in a single creative is a focus group, not an ad. The reason ad angles for ecommerce are worth a dedicated round of testing is that the algorithm routes the angle to the buyer, but only when the angle is legible enough for the algorithm to read.
Layer 3: Formats, the container
Hooks decide stop. Angles decide buy-in. Formats decide which surface the ad lives on. Same hook, same angle, three formats often produces three different click rates.
The four formats we run on most accounts:
- UGC - one creator, one camera, talking. Cheap to make, fast to iterate, native to feed.
- Founder-led - the operator on camera, voice-of-the-brand. Works for higher-trust purchases.
- Studio static - the clean shot, no people. Works when the product is the story.
- Motion graphic - text and animation. Works for B2B and SaaS where the value prop is the asset.
A winning hook on the wrong format is a near-miss. A losing hook on any format is just a loser. That's why format is the third lever, not the first.
A losing hook will not be saved by a better format. A winning hook will be magnified by the right one. Order matters because the math compounds.
Layer 4: Audiences, last and lightly
This is where the creative testing vs audience testing debate ends. Audience used to be the lever: five interest stacks, three lookalike percentages, the whole 2018 deck. It is not anymore. Meta's Advantage+ and Google's PMax are built to find the buyer. The creative is what tells them which buyer.
If you're still running creative testing vs audience testing as parallel programmes, you're double-paying for one signal. The angle is the audience. A peri-menopause angle finds peri-menopausal women more reliably than any interest layer ever did, and the platform reads it instantly.
What we still do at this layer: keep a broad opener and a retargeting bucket separate, and turn off interest stacks visibly under-delivering against broad. That's it. Two decisions, not twenty.
What good looks like
A clean account is mostly broad, mostly tested at the hook layer, with a small library of angle variants in rotation and a format change every two weeks. The reports get boring. The numbers get steadier. The graveyard of dead creatives gets longer than the active set, which is exactly what you want.
That discipline only holds if your testing process is structured and your account is watched daily. This is where BAVai earns its keep. Every account gets scanned at 7am, so the moment a hook test flatlines or a winning angle fatigues, the system flags it before the day's spend compounds the mistake. The human picks the next framework. The machine keeps the test honest overnight.
The checklist
- One framework per ad, never two stacked
- Hooks tested first, judged on hold rate not ROAS
- Angles tested on the winning hook, not on a blank canvas
- Formats tested third, only once the angle is settled
- Audiences kept to broad plus retargeting, full stop
- Every test logged against the framework it ran, no "I don't know what worked"
- Dead creatives archived, not left running at $5 a day
Are your creative tests teaching you which framework works, or just which ad got lucky this week?