Here is the whole thing before the caveats: every Monday you launch three hooks against one offer, each hook cut into two formats, six ads total. Wednesday you kill on a spend threshold. Friday you decide iterate or scale. Sunday you build next week's three hooks from what won. That's the ad creative testing framework we run for D2C accounts on Meta, and its only real virtue is that it changes one variable a week, so when performance moves you know why.
Most accounts we audit are doing the opposite. New offer, new landing page, new hook, new format, all shipped Thursday afternoon. CPA drops 18%. Nobody can say which change did it, so nobody can repeat it.
Why this ad creative testing framework holds the offer still
The offer is the loudest variable in any ad. Flat 20% off beats a founder story in almost every first-touch test, and it will beat it again next week, which tells you nothing you didn't already know. If you change the discount and the hook in the same week, the hook data is contaminated. So the offer is fixed for seven days: same discount, same bundle, same free-shipping threshold, same landing page.
Three hooks is the number because two doesn't give you a spread and five spreads your budget too thin to read anything. Pick three that come from genuinely different places:
- A problem hook. The first three seconds name the annoyance the product removes. Works well for personal care, home, anything with a visible before state.
- A proof hook. Review text on screen, a number, a certification. FSSAI licence numbers and lab reports carry real weight in supplements and packaged food.
- An outsider hook. Something oblique — a comparison, a mild contradiction, an unfinished sentence. Two thirds of these die. The third one occasionally halves your CPA.
Same product, same offer, same call to action across all three. Only the opening seconds and the framing change.
Two formats, and which two
One static and one motion, almost always. The static is where hooks are cheapest to test — you can produce six variants in an afternoon with a designer and a decent product shot. The motion version tells you whether the hook survives being spoken out loud by a person.
Which motion format depends on what you can actually shoot. UGC talking-head if you have creators on retainer. Screen-recorded demo if the product does something. Simple product-in-hand b-roll with captions if neither. What we avoid is testing two motion formats in the same week, because production time means one of them ships late and the comparison is already broken.
If your catalogue is wide and the ad drops people onto a collection page rather than a PDP, add a carousel as the second format instead of the static. Different job, different read.
What a decision-grade read actually costs
This is where most testing plans quietly fail. People run six creatives on ₹2,000 a day and wonder why week four looks like week one.
Work backwards. Say AOV is ₹1,800 and contribution after COGS, shipping and payment fees is ₹880 per delivered order. Sixty per cent of your orders are COD and roughly a fifth of those come back as RTO, so blended contribution lands nearer ₹700. That's your break-even CPA on a first order. Set the working target at ₹560 to leave margin.
Now the media side. At a ₹250 CPM, 1,000 impressions with a 1.2% CTR gives you 12 clicks, so ₹20.83 a click. If the landing page converts at 2.2%, those 12 clicks produce 0.264 orders, and ₹250 ÷ 0.264 is a CPA of ₹947. Well above break-even.
Lift CTR to 1.8% and nothing else changes: 18 clicks per 1,000 impressions, 0.396 orders, ₹250 ÷ 0.396 = ₹631. A hook that works and nothing more knocks a third off CPA. That is the entire argument for testing hooks separately from everything else.
For budget: at ₹947 CPA you need about ₹3,000 behind a creative to see three purchases. Six creatives at ₹1,000 a day for three days is ₹18,000 a week, ₹72,000 a month on testing alone. If your total Meta spend is under ₹1.5L a month, that ratio is wrong. Run three hooks in one format instead of two, ₹9,000 a week, and accept that you're learning at half speed.
The decision rule
Written down, applied the same way every week, no debating on Slack.
Wednesday, 72 hours in — kill. Any creative that has spent 1.5× target CPA (₹1,050 in the example above) with zero purchases is off. Any creative with CTR under 0.8%, or a 3-second view rate under 20% on video, is off regardless of what its CPA looks like, because a low hook rate means the audience isn't getting far enough for the rest of the ad to matter.
Friday, day five — diagnose before you judge. Split the failures. Good CTR with bad conversion is not a creative problem, it's an offer or landing page problem, and killing the creative teaches you nothing. Bad CTR with good conversion on the few who click means the hook is repelling the right people; try the same message with a softer opener.
Iterate anything sitting between 100% and 130% of target CPA with above-median CTR. Next week it becomes two variants: same body, new opening three seconds, and a swap of format. You are not making a new ad, you are making the same ad again with one thing moved.
Scale only on ten or more purchases, three consecutive days at or under target CPA. Move the winner into the always-on campaign as a new ad, don't rebuild it. Raise budget 20-30% every second day. Faster than that and you reset learning, which costs you three days to find out you were greedy.
Be honest about the statistics. Three purchases on ₹3,000 of spend is not significance. You're allocating money under uncertainty, not publishing a paper. The framework works because it repeats weekly, not because any single week is conclusive.
Naming, so the sheet reads itself
Six creatives a week is 300 a year. Without a convention you will lose the plot by March.
We use YYWW_offer_hook_format_version. So 2514_flat20_proof_ugc_v2. Week number first because that's how you sort, offer second because that's the thing you're holding constant, hook third. Every creative gets one row in a sheet with spend, impressions, hook rate, CTR, purchases, CPA and one free-text column for the decision made and why. That last column is the asset. Six months in, it tells you which hook family works for which product category, and you stop re-testing things you already lost money learning.
Your landing page is part of the test whether you planned it or not
A hook can only move CTR. The click-to-order rate belongs to the page. If your PDP takes 4 seconds to render on a mid-range Android over 4G, and a lot of India's traffic is exactly that, every creative in the test is being scored against a broken denominator, and the winner is just the ad that happened to attract the most patient users.
Fix the page before you start a testing programme, not during one. Check LCP on the actual URL your ads point to, on mobile, throttled. Under 2.5 seconds is the bar; under 1.5 is where paid traffic starts behaving. SwiftStore will scan the store, fix the easy wins and keep watching the score, which is enough for most themes. If the theme is heavy or you've collected twelve apps injecting scripts on every page, that's a deeper piece of work and worth doing first.
One more thing on pages: don't change the landing page mid-week. It's part of the constant.
Where the framework breaks
Single-product brands run out of hooks around week six. You cannot generate three genuinely distinct angles a week forever on one SKU. When that happens, shift the cadence to three hooks a fortnight and use the off week for audience or placement tests instead.
Accounts with heavy COD exposure get a distorted read, because Meta counts the order and your warehouse counts the return. A creative that pulls impulse COD orders at a great CPA can be your worst performer on delivered revenue. If COD is more than half your volume, wait for RTO data before scaling anything, or judge on prepaid purchases only and treat COD as upside.
Retargeting doesn't belong in this framework. Volumes are too small, the creative job is different, and mixing it in makes your prospecting numbers look better than they are. Keep it in a separate campaign and leave it alone.
And if you're spending under about ₹1L a month, the honest answer is that structured weekly creative testing is not your constraint. Two decent ads, a page that loads, and a working offer will take you further than a framework you can't afford to feed.
Build the library before October
Festive-season CPMs in India move enough that a creative which cleared target in August can be 40% off pace in the second half of October. You are not going to discover new winners at peak CPMs with peak urgency. So run the weekly cadence hard from June through September, and by the time the sale calendar starts you have eight to ten proven hooks to rotate, plus the sheet telling you which ones held up when costs rose last year.
During peak, stop testing hooks and test offers instead. That's the one week of the year when the offer is the interesting variable.
Start with one week. Pick the offer, brief three hooks, cut each into a static and a video, and write the kill threshold on the brief before anything goes live. If the landing page is the thing you're least sure about, a free audit will tell you whether your creative budget is being spent against a page that can convert it.