In short
- The effect of tests shrinks from top to bottom: offer, buyer situation and message, format, hook, small tweaks. Do not test a lower level before the one above it is locked.
- Most tests lose. That is not a reason to test less. It is a reason to test things whose win covers all the losses.
- Every ad has one job and one hypothesis, written down before launch. Otherwise a losing test teaches you nothing.
Why the order decides
When you test two versions of a headline, the best possible win is usually a few percent. When you test who the ad speaks to and in which situation, the cost per purchase can halve. Same budget, same work, completely different stakes.
Most tests lose no matter how well they are done. That is why the thing you test has to be big enough that one hit pays for ten misses. Testing small things, you lose the same ten, but the hit does not cover them.
The order follows from this. The upper level is locked before you start tuning the lower one, because a lower-level win disappears if the level above changes later.
Five levels, top to bottom
| Level | What you test | Effect | When |
|---|---|---|---|
| 1. Offer | What the buyer gets, at what price and on what terms | Largest | First, then rarely |
| 2. Buyer situation and message | Who the ad speaks to, in which situation, and what it claims | Large | Continuously, one angle at a time |
| 3. Format and messenger | Who speaks and in what form: founder, customer, demo, image or video | Medium | Once the message is found |
| 4. Hook | The first three seconds or the first line | Medium | For every winning message |
| 5. Small tweaks | Headline, color, button, text length | Small | Last, if ever |
1. The offer
A weak offer cannot be rescued by a great video, but a strong offer carries a mediocre ad to decent results. Offer does not mean discount. It means what the buyer gets and on what terms: bundle, price, delivery time, guarantee, returns.
If the advertising produces almost no sales, look at the offer first. Things to test include a bundle instead of a single product, how the returns policy or guarantee is worded, and moving the shipping cost into the price.
2. Buyer situation and message
This is the level where we spend the most time. The same product is bought for different reasons in different situations, and an ad works when the buyer recognizes their own situation in it. That is why one ad speaks to one situation and makes one claim.
Ads have different jobs along the path to purchase, and each job is its own test:
- Recognizing the problem. The buyer is not yet looking for the product, but recognizes their everyday life in the ad.
- A new explanation. Why the product works when earlier solutions did not.
- Proof. A customer's words, a number or evidence that the promise holds.
- Removing an objection. The one doubt that stops the purchase: price, size, fit, ”will this work for me”.
- The offer at the moment of purchase. The buyer is already convinced and needs a reason to order today.
When you test the message, keep the format and the offer the same. If you switch from video to image at the same time, you will not know whether the message or the format won.
3. Format and messenger
In our webinar we were asked whether Meta always needs video. It does not. An image beats a video regularly when the message is simple and the product recognizable. Format is tested only once you know which message works.
Besides the format, test the messenger: founder, customer, an outside creator, or just the product and text. The same message told by the founder and by a customer can produce very different results, and the difference tells you what the buyer needs in order to believe.
4. The hook
The first three seconds of a video and the first line of an image ad decide whether anyone sees the rest. Meta shows this directly: three-second views as a share of impressions. If the hook stops 30 percent of viewers instead of eight, almost four times as many people see your message for the same money.
The hook is tested on a winning message, not a random one. The same video with three different openings is the cheapest test you can run, because the rest is already produced.
5. Small tweaks
Headline, button text, color, text length. These have their place once the four upper levels are locked and the ad is already producing. Before that they are a way to look busy while nothing changes.
How one test is built
- Write the hypothesis before launch. ”If this ad wins, it tells us the buyer doubts the price.” If you cannot write the sentence, the ad has no job.
- Change one thing. Same audience, same offer, same format, one variable. Otherwise you do not know why the winner won, and you cannot repeat it.
- Give the test enough data. Two purchases cannot decide anything. How to calculate the budget you need is in Set your Meta ads budget from purchases, not gut feeling.
- Decide the decision rule in advance. How many purchases or how many days, and which number decides. Deciding before the data prevents picking the winner by feel.
- Record the learning, including the loss. A losing test tells you what the buyer does not need. That is as valuable as a win, if it is written down somewhere other than memory.
Where new tests go
In the webinar we were also asked where to put new ads when you run one campaign, so that learning does not restart. Our way is to keep testing and scaling apart: new ads are tested in their own ad set, and only the winners move to the ad set that produces. Nothing new is added to the producing ad set weekly, because every significant change restarts its learning.
A system, not lucky hits
A single winning ad happens by luck to anyone. Repeated winners only come when every test leaves behind a learning that the next one builds on. That is the whole point of this order.
GreenGoKauppa's sales grew from €300,000 to €1.2 million in January–June, year over year. In the webinar recording (in Finnish) we walk through how the testing changed along the way: fewer new ads, more real tests. From the same page you also get the Growth Playbook, where the same thinking is laid out as a 60-day plan.
We do not know whether this fits your store. If you want a second opinion on what your account should test next, book 30 minutes below. We will tell you what we see.
Book a free 30-minute call →
