The creative brief is a hypothesis document

Ask a creative team what a brief is for and most will say it tells production what to make. That is what briefs do in practice, and it is why so many creative programmes generate hundreds of assets a quarter and learn almost nothing from them. AppsFlyer's State of Gaming for Marketers 2026, published last month, described top advertisers producing well over two thousand creative variations per quarter. Volume at that scale without a learning structure is a factory, not a strategy.

The position here is that a creative brief should be written as a hypothesis first and a spec second. Its primary job is to state what the team believes about the audience, in a form that a test can prove wrong. The deliverables follow from the hypothesis. Practitioners who run brief-as-spec will object that this slows production. It does, by a few minutes per brief, and it speeds up everything downstream.

What a deliverables brief cannot do

A conventional brief specifies format, length, hook style, call to action, deadline. Production makes the assets, measurement tests them, and some perform; the team scales those and drops the rest.

Nothing in that loop generates a transferable belief. The team knows that asset 47 beat asset 46. It doesn't know why, because the brief never said what 47 had to demonstrate. Next quarter's brief asks for more like 47, and the team is now iterating on an artefact rather than on an understanding. Over a year this produces a portfolio of local optima and a creative director who can predict performance but can't explain it.

The mechanism behind this is that a spec gives production a target and gives measurement nothing. Measurement then defaults to whichever metric is easiest to read (usually CPI) and nobody goes back to check the brief's original intent, whatever it was.

The shape of a hypothesis brief

A hypothesis brief adds three lines above the deliverables and changes the meaning of everything below them.

The belief: one sentence about the audience or the game. "Players who respond to this title care about completing sets, not about speed."

The prediction: what should happen if the belief is true. "A hook that shows a nearly complete collection will out-convert a hook that shows fast play, and its cohort will retain at least as well."

The disproof: what result would make the team abandon the belief. "If the speed hook wins on both install rate and day-seven retention across two rounds, the belief is wrong."

Everything else in the brief, the formats, the lengths, the number of variants, is now in service of producing a clean comparison. Two hooks, matched on everything except the variable under test. Not fifteen variants that each differ in four ways.

The second-order effect: fewer assets, more usable ones

A hypothesis brief nearly always asks for fewer assets than a deliverables brief. This is the trade-off that makes teams hesitate. Testing one variable cleanly means holding others still, and holding others still means fewer combinations. A team used to shipping twenty cuts per brief will ship six.

What it gets in return is a result that generalises. A belief confirmed in one test applies to the next brief, the next title in the portfolio, and the next platform. The desk has argued elsewhere that concept-level thinking is where creative learning lives; the hypothesis brief is the document that makes concept-level thinking operational.

There is a less obvious effect on the production team. Editors and designers given a hypothesis make different choices from those given a spec. They know which element is under test and protect it. They stop adding the flourish that would confound the comparison. Craft improves because intent is visible.

A worked illustration

An illustrative mid-core studio has a strategy title whose UA has plateaued. The deliverables brief for the quarter reads: twelve videos, four playables, a mix of UGC-style and produced, a mix of hooks, end of month.

The hypothesis version of the same brief for the first fortnight reads: belief, the audience is under-served on base-defence fantasy and over-served on PvP conquest. Prediction, a defence-framed hook will lift install rate and hold retention against the current conquest-framed control. Disproof, if defence loses on installs or drops retention in two consecutive rounds, we return to conquest and test the next belief. Deliverables: two videos plus one playable per framing; same length, same music, same end card.

The first round runs and defence wins on installs while matching on retention; the second round holds. The studio now holds a belief rather than an asset, and the following brief tests whether defence framing transfers to the UGC-style layer. Six weeks in, the team has a map of the audience's motivations that it can hand to the product team, something a hundred unstructured variants would never have produced.

Running the brief as a log

The last piece of the method is to keep every hypothesis brief after the test, with the result written into it. Write in whether it held, fell or came back inconclusive, along with the round count and the metric that decided it. This log becomes the creative team's actual asset. It is what a new hire reads in the first week. It is what a creative director brings to a quarterly planning meeting in place of a highlight reel.

It also sets an honest bar for the word "insight". A team that has run thirty hypothesis briefs in a year has thirty statements about its audience with evidence attached, some of them wrong in instructive ways. A team that has shipped two thousand variants has two thousand assets and a spreadsheet. The second team will produce more ads; the first will know what its following brief should test.

When a test refutes a hypothesis, write the refutation in larger type than the confirmations. The beliefs a team was wrong about are the ones that were shaping its briefs without anyone noticing, and finding them is the only part of this process that pays for itself in the same quarter.

Related archive reading

These articles provide related context and remain subject to their stated review status.

• Creative when everyone has the same tools
• Creative Operations for a Scaling Game Portfolio