We test ideas, not single ads. Each idea gets its own space, a fair budget and a clear pass or fail. The winners move into one main campaign, where a small set of proven ads gets most of the money.
Testing takes a small, fixed share of the budget. Everything that proves itself moves across to Scale, and Scale stays lean so every ad in it gets enough spend to keep learning.
Every ad set below targets the same broad audience. Meta decides who sees which ad.
One ad set per idea. Each idea has 3 to 5 versions and its own set budget, so every idea gets a fair read.
Only proven ads. Meta moves budget to whatever's working best that week.
Creative matters from the first second, so we never test words on their own. Each idea is one message with creative made to say it: the video, the words on screen and the copy all say the same thing. Once an ad has proven itself, we refine it one change at a time.
One message plus the creative made to say it.
One ad. Each idea has 3 to 5.
The first 3 seconds: shot, words on screen, first line spoken.
Talking to camera, demo, review card, static.
Question: which idea sells best? One ad set per idea. This is how most of our testing runs.
Question: what gets even more out of an ad that already works? Only for specific cases.
Also for: the same image with a different headline on it.
Inside a normal ad set Meta picks a favourite early and gives it nearly all the budget, so the other versions never get a fair read. That's fine when we're judging the whole idea, but it can't tell us which hook is best. For that we use the A/B test tool.
Every version needs about 3× the target cost per sale on its own, so a three-way test costs about three times as much as one idea test. It's worth it on an ad that's already earning, not on an untested one.
A different buyer has a different reason to buy: a beanbag as a gift is a different message from a beanbag as a statement piece for yourself. So each persona gets its own ad set, with the hook and the copy made for them.
People see the same ads many times, in no set order, so one ad does the early work and the closing work depending on who's watching. Each ad carries copy for different stages as its text options, and Meta shows each person the one that fits.
The "why" comes from the test log: every idea is labelled by its message, its format and its proof, so patterns show up across rounds.
Every test answers one question. We go down these in order and stop at the first yes.
A different hook on the same video, or a different headline on the same image.
One version per hook or headline. Same copy on every version. Spend is split evenly so each one gets a fair read.
Different buyers, or different messages, for the same product.
Each ad set gets its own hook and copy made for that buyer. The product stays the same, so the buyer is what we're comparing.
A new idea, or a new take on one that's working.
3 to 5 versions: carousel, video, a re-cut, a static. The same copy on all of them, with 3 to 5 text options that vary the writing and the stage.
The creative is fine; we think the copy could work harder.
New writing structures and stage copy go in as extra text options. We read how each one does as a steer, not a verdict.
What each branch looks like when we build it.
We want to learn: does "build a cosy corner" sell, and which way of showing it works best?
We want to learn: which buyer should lead the beanbag ads in November? Run in October, so the winner is ready for the peak.
We want to learn: which opening gets the most out of a video that's already sold well?
Every rule is set against your target cost per sale, so it scales with the account.
About three times the target cost per sale before we call it.
Spent twice the target cost with no sale? We pull it.
Judged on 7-day click sales, not on who merely saw it.
Meta favours one version early. Versions with little spend are unread, not losers.
Every idea ends up in one of three places. Nothing sits in testing forever.
the idea against the other ideas running that week
Hit the target cost per sale. It's copied into Scale as the same post, so it keeps its likes and comments, and the weakest Scale ad comes out.
Close to target, or it didn't get enough spend to judge. Top it up, or change one thing and run it again.
Hit 2× target with no sale, or clearly behind. We write down what we learned so the next idea is smarter.
A steady rhythm stops us reacting to one good or bad day.
New ideas go live. Last week's winners move to Scale.
Tests run untouched so the read is clean.
We check the top spenders in Scale for signs they're wearing out.
What we tested, what happened, what we learned.
Before anything launches, we work down the list. The first question with an answer sets that week's tests.
If a top ad is wearing out, the first job is its replacement: ideas close to what made it work, or a fresh hook on the same ad.
Every log line ends with "so next we test…". A win gets refined, a loss moves to a different idea, an unclear one gets another week.
Work back from stock, launches and offers. Beanbags peak in November, so the gift vs statement piece test runs in October and the winner is ready in time.
Buyers with no idea built for them yet, and claims or formats we've never used. This is where new winners come from.
When there's more than one candidate, we score each on three things and run the highest first:
How much revenue rides on the answer.
Reviews, past ads, what's working for others.
Can we build it from what we already have?
Roughly one test that builds on a winner and one or two that try something new. The testing budget covers several ideas a week, so making the creative is usually the limit, not the money.
Every test starts with what we expect to happen. If we can't say which result would change our next move, it isn't worth testing yet.
One shared log, one line per test. Every idea is labelled by its message, format and proof, so after a few rounds we can see what's really winning, e.g. one message beating the rest whether it's a video or a static. Over time this becomes the playbook for what Hyde & Hare customers respond to.
| Test | Message · format · proof | What we expect | Result | What we learned |
|---|---|---|---|---|
| Persona A idea | Main reason to buy · talking to camera · a customer says it | Speaking to their main reason beats a general message | Filled in after 7 days | Filled in after 7 days |
| Hook A/B test on a winner | Same video · 3 different openings | Opening on the offer beats opening on the product | Filled in after 7 days | Filled in after 7 days |