Creative and testing guides

Most creative tests fail before they start, and they fail in the same three ways.

They ask more than one question at once. Two concepts that differ in the hook, the format, the offer and the landing page have told you nothing when one wins, because you cannot say which difference did it.

They have no stopping rule. A test with no agreed end runs until somebody looks at it, which means it ends on the day the numbers happen to look interesting. That is a procedure for finding noise.

They are too small to answer the question they ask. A difference you cannot detect with the volume you have is not a finding either way, and running the test anyway converts a budget into a coin flip you will act on.

A guide here is an argument for the discipline: one variable, a size decided in advance, a stopping rule written down, and a definition of what would make you keep the loser.

Before a test has enough conversions to grade, it has cheaper signals. How many people stopped on the asset at all, how many stayed past the opening, how many clicked once they did — each one narrows where a weak result came from.

They are diagnostic, not conclusive. An asset that holds attention and converts nobody has a landing page problem or an offer problem, not a creative one. An asset nobody stops on has told you nothing about its message, because nobody got to it. Reading them in that order — did it get seen, did it hold, did it move anybody — is what turns a flat result into a next step instead of a shrug.

These guides do not publish a good number for any of them. A threshold stated as a fact about the world needs a real source, and the useful comparison is your own account's history anyway: the asset you are replacing is the benchmark that matters.

Worth stating plainly so the guides can be specific. The workspace holds a media library of the images and videos it has and has handed to the platforms, and the reading surface lists it. Campaign structure reads down to ads and their creatives, so an agent can answer which assets are live under which groups without a person opening a platform.

A change to a creative is a change request like any other, and one detail about it matters for testing: creative and targeting writes are complete replacements rather than patches, which is why the undo for one is the previous specification put back exactly rather than a field reverted. Uploading media is never a rule's to clear — it always waits for a person.

Tempio does not generate creative. A guide that implied otherwise would be selling you something you cannot use, so nothing below assumes it: every asset these guides discuss is one you had made.

The arrangement, stated once: your connections read your accounts, a change is written as a draft on the record with the state it would change from and the checks it passed, applying it is gated on approval — by you, or by a rule you wrote — and no platform write reaches a customer account until that platform has granted write access. Applying a cleared change to a platform is in private beta. Two creative operations are worth singling out against that: an upload and a replacement are both operations no rule may ever clear, so each one waits for a person however the rest of the workspace is configured.

The hardest part of the one-variable rule is agreeing what a variable is. Two assets that share a script and differ in their first three seconds are one variable. Two that share a first three seconds and differ in everything after are also one variable, and a different one. Two that were made by different people to the same brief are not a variable at all — they are two guesses, and whichever wins tells you nothing you can apply to the next pair.

The test that settles it is whether you can write down, in one clause, what you would do differently if this asset won. "Open on the product rather than the person" is a clause you can act on. "Use the second concept" is not, because there is no next asset it tells you how to make.

That is also the argument for keeping the losing asset running longer than feels comfortable. A variable proved on one pair is a hypothesis; the same variable proved twice is a rule about your audience, and the second proof costs a fraction of the first because you already know what to vary.

The least interesting section of any creative guide is the one that pays for itself first. An asset name is the only field that reaches every report downstream, and a naming convention invented per campaign is a report nobody can group.

The rule that holds: names encode the things you will want to filter by later and nothing else — the concept, the variable under test, the format, and the date it entered. Not the reviewer's initials, not the version number of the file, not the client's nickname for it. A name is carried onto the draft and onto the entity, so it is a decision made once and lived with for the life of the asset.

This subject is the one with the most variance in it and the least machinery around it.

• Measurement and metrics is the prerequisite: a test is only as good as the agreement about how its result will be read.
• Budget and bidding is the usual confound — a test run across a budget change has measured two things and can attribute neither.
• Media planning is where a test gets its own line and its own money, which is how it stops being graded against campaigns that are not experiments.
• Governance and approvals is why an upload and a creative replacement always wait for a person.
• Agent setup is how the library and the live structure become one question rather than a tour of several platforms.