What a media buyer needs from your creative

You are paying somebody to run your ads. They asked for creative, you sent them the video you were proud of, and a week later they came back asking for four more versions, a different shape, and a copy with nothing written on it.

A buyer may need different compositions, openings and text treatments for the planned placements and tests. Ask which decision each file will support.

What follows is the list of what to send and, more usefully, why each item exists from the other side of the handover. Once you understand the why, you can brief it rather than being asked for it piece by piece.

Why does the buyer keep asking for more files?

Because the account learns which creative wins by comparison, and one asset gives it nothing to compare against. A single ad still teaches you about audience, placement and whether it converts at all; what it cannot tell you is whether a different hook or angle would have done better, because there is nothing beside it.

Several versions give a buyer something to compare. To interpret the difference, the test also needs appropriate allocation, measurement and enough data; a set of files alone does not establish the cause.

There is a second reason, which is that creative wears out. The same file in front of the same people stops working after a while, and the buyer needs somewhere to go when it does. A buyer with one asset can only turn it off.

One video is not one asset

Placements are shaped differently, and though the platform will stretch, crop or letterbox one file to fit them all, a composition built for the wrong rectangle arrives compromised. Purpose-built shapes are worth the extra exports.

A vertical full-screen placement wants the frame filled top to bottom, with the important content kept clear of the strips at the top and bottom where the interface sits. A feed placement is usually square or close to it, and it competes with things around it rather than owning the whole screen. Anything horizontal is a different composition again.

Plan the required shapes before filming. Some footage can serve several crops; other compositions need separate coverage. Checking early is usually easier than finding a missing view in the edit.

Send at least two shapes. Ask which two before anything is filmed.

What should change between variants?

It depends on the question. Change one element to investigate that element, or compare complete concepts to decide which package works better.

When several elements change together, the result compares the package. It cannot tell you which individual element caused the difference without further testing.

So the useful set is several openings on the same body of a video. One asks a question. One shows the result first. One says the objection out loud. Same footage after the first few seconds, same length, same everything else. That set produces a readable answer about openings, and openings are the most useful thing to have a readable answer about.

Film the alternatives while the camera is still set up. Afterwards they cost a shoot day. Before you break down, they cost minutes.

Keep the offer out of the picture

Because an offer burned into the video has an expiry date, and burning it in gives the whole asset the same expiry date.

A price, a deadline, a seasonal reference: any of these written into the frame means the file dies with the promotion. It also cannot run in a market where the offer differs, and it cannot be reused next year.

Keep a reusable version without the offer burned in. The published advert may still need a price, deadline or offer on screen; use a replaceable card or graphic and keep the text-free master.

The same applies to anything that dates: a member of staff who might leave, a packaging design about to change, an award with a year attached.

What is a text-free master for?

It is the version with no captions, no titles and no graphics baked into the picture, and everything else can be built from it later.

Captions can be added to it in any language. On-screen text can be rewritten when the message changes. A shorter cut-down can be made without a subtitle running off the end of a shot. And where a platform's own caption tool is preferred, it can be used without fighting text that is already in the picture.

Ask for it as a delivery item alongside the finished versions. It costs an export. Without it, the finished versions are finished in the most literal sense.

What does the buyer need in writing?

One short note per variant, saying what it is testing and what it assumes about the viewer.

"Version 1: opens on the problem, aimed at people who do not yet know this is fixable. Version 2: opens on the result, aimed at people already comparing suppliers." Two lines each. That note lets the buyer point each version at the right place rather than guessing, and it means the results come back as knowledge about your customers rather than as a ranking of files.

A short note gives the buyer the intended question and audience. If a concept performs well, that context helps plan a follow-up test without pretending the result explains every component.

The next time creative is commissioned, agree the shapes, the number of openings and the text-free master before anybody films anything, because every one of those is free to decide in advance and expensive to add afterwards.

Sources and evidence

Standards and platform facts. The reasoning that a single asset cannot be tested against anything follows the same logic set out in Microsoft Research's work on online experimentation: a result can only be attributed to a cause when there is something comparable to weigh it against. That paper concerns large-scale online testing generally, not creative production, so it supports the underlying principle rather than any of the specific production advice in this piece.

Our method. The list of what to send a media buyer, including purpose-built shapes, single-variable variants, a text-free master and a written note per version, is this studio's production practice rather than a documented platform requirement.

References.

• Microsoft Research. Online Experimentation at Microsoft. microsoft.com. Randomisation and proper design support causal claims. Platform experiments need power, guardrails and instrumentation.