Classic split tests of one button colour against another tell you little under Meta Andromeda, and launching ads into a big ad set can leave new ideas with no spend at all. This framework shows how to test creative fairly now, what to measure, and how to scale what wins.
Creative testing has always been the engine of good Meta advertising. But Andromeda changed how tests behave. Ads that look similar get grouped together, the system often spends most of the budget on one or two ads very quickly, and tiny changes rarely move results. Here's a framework that works with the system rather than against it.
What changed for testing
- Similar ads are grouped. Meta says it groups ads sharing key visual and thematic attributes. Testing two near-identical ads is often testing one ad.
- Spend concentrates fast. In a broad ad set, Meta quickly favours the ads it predicts will do best. New ads can get little spend if a strong ad already dominates.
- Concepts beat tweaks. Because the creative now does much of the targeting, a different concept can reach a different group of people entirely, while a colour change rarely does.
Step 1: Decide what you're testing
Test at the level that matters most first:
- Concept and angle: price vs quality vs convenience vs problem-solution.
- Person and proof: founder vs customer vs creator; review vs demo.
- Format: Reel vs static vs carousel.
- Hook: the first one to two seconds, or the headline on the image.
- Details: copy length, call to action, colours. Only once the big things are settled.
Step 2: Choose a test structure
Option A: test inside your main ad set
Add two or three new concepts to your main broad ad set. Simple, and the system judges them against real competition. The risk is that existing winners take most of the spend. This works well when your budget is modest and you add new concepts every couple of weeks.
Option B: a separate testing campaign
Keep a small campaign (often 10 to 20 percent of budget) with one broad ad set per batch of new concepts. Each batch gets guaranteed spend. Winners graduate to the main campaign. This suits brands that produce a lot of creative and want clearer reads.
Option C: Meta's A/B test tool
For a big strategic question (for example, founder-led videos vs product-only ads), use Ads Manager's A/B testing, which splits audiences so each version gets a fair share. It needs more budget and time, so save it for decisions that matter.
Step 3: Give each test enough to mean something
- Aim for at least several results per concept before judging, not just clicks.
- Run for at least five to seven days to cover weekday and weekend behaviour.
- Avoid editing the ad set during the test, which can reset the learning phase.
- Don't test during unusual weeks (a big sale, Diwali week) unless the test is about that event.
Step 4: Read the results properly
Look at these in order:
- Cost per result (purchase, qualified lead) and, for stores, ROAS on delivered orders.
- Spend share: did Meta give the concept a real chance? An ad with almost no spend hasn't lost, it hasn't been tested.
- [Hook rate](/glossary/hook-rate) and [CTR](/glossary/ctr): why it worked or didn't. Strong hook but poor sales suggests the offer or landing page; weak hook suggests the opening.
- Breakdowns by age, placement and region: different concepts often win with different people, which is exactly what diversification is for.
Step 5: Scale winners without breaking them
- Move winners into your main campaign using the same post (post ID) where you can, so likes and comments carry over.
- Don't pause a long-running winner just because a new ad looks good for three days.
- Make new versions of winning concepts with different hooks, people and formats. That's iteration, and it's valuable once the concept is proven.
- Raise budgets gradually, about 20 percent every few days. See the ad budget guide.
A simple monthly testing rhythm
- Week 1: launch two or three new concepts.
- Week 2: read early results; launch one or two more concepts.
- Week 3: move clear winners to the main campaign; iterate on winners with new hooks.
- Week 4: review the month: which angles and people win, what to make next month.
Every Varko plan includes 8 new creatives a month, 6 images and 2 videos, which fits this rhythm. See pricing.
Quick quiz · 5 questions
Creative testing quiz
1. You test a blue button vs a red button on the same image. What are you mainly testing?
2. A new ad got almost no spend in your main ad set. What does that tell you?
3. Which metric should decide a winner for an online store?
4. Strong hook rate, many clicks, few sales. What's the likely problem?
5. How should you scale a winning ad?
0 of 5 answered
Write a hypothesis for every test
A test without a reason teaches you little. Before launching, write one line: "We think [concept] will beat our current best ad because [reason], and we'll know by [metric] after [time]." For example: "We think a customer review video will beat the product demo for women over 35, because they told us trust matters most, and we'll know by cost per purchase after 7 days." When results come in, you learn something even if the test loses.
Example: a month of tests for a skincare brand
- Test 1, angle: "gentle for sensitive skin" vs "visible results in 4 weeks" vs "dermatologist-made", same format.
- Test 2, person: the winning angle shown by a college student, a new mother and a woman in her forties.
- Test 3, format: the winning angle and person as a Reel, a static and a carousel.
- Test 4, hook: three different opening lines on the winning Reel.
Each test builds on the last, moving from the biggest questions to the smallest. That's how diversification and iteration work together.
What to do with losing ads
- Check whether it was really tested (enough spend and results) before calling it a loser.
- Look at why: weak hook, weak offer, wrong audience, or poor landing page.
- Keep a log of losing angles so you don't repeat them without a new reason.
- Sometimes revisit: an angle that lost in July may win in Diwali week.
Keep a simple testing log
A shared sheet with these columns is enough: date, concept name, hypothesis, spend, results, cost per result, hook rate, CTR, verdict and next step. After a few months, it becomes the most valuable document in your marketing: a record of what your customers respond to.
Testing on a very small budget
With ₹500 to ₹1,000 a day, test fewer things more slowly. Add one or two new concepts every two weeks to your main ad set, and judge on cost per result over two weeks rather than one. Use cheaper signals (hook rate, CTR, cost per add to cart) as early hints, but confirm with purchases or leads before scaling. See the ad budget guide.
Next in the series: Andromeda for online stores and D2C brands. Back to the main Andromeda guide. For making the creatives themselves, see our ad creative guide.
Frequently asked questions
Two to four new concepts per batch suits most small budgets. More than that spreads spend too thin to learn anything.
It's useful for finding good combinations of headlines and images within a concept, but it reports results in combination, so it's less clear for comparing whole concepts.
It varies by budget and audience size. Watch frequency and cost per result over time; rising both together signals ad fatigue.
