Andromeda changed retrieval.
Your brief needs to catch up.

Start with what Meta actually published.

Meta introduced Andromeda publicly in December 2024. It is a system for ad retrieval: selecting relevant candidates from a very large pool before later ranking stages choose what to show. Meta describes increased personalisation and an architecture designed to handle growth in eligible creative, including generated assets. Meta’s engineering explanation ↗

That description is not a rule requiring every advertiser to upload a fixed number of ads. It does not establish that a new background creates a new audience, that targeting is irrelevant, or that one campaign structure will work in every account.

For your next brief, start with three different reasons someone might buy. A commuter wants easy access. A careful buyer wants to check fit. A style-conscious buyer wants to see the bag with their everyday clothes. Each reason deserves its own argument and its own visual evidence.

A different file is not necessarily a different idea.

Think in two levels. A concept is the argument for buying: solve a daily frustration, demonstrate compatibility, make a design choice visible. An execution is the way you express that argument: the opening line, camera angle, crop or format.

Both levels matter. Rewriting a weak opening can rescue a useful concept. But twenty executions of one argument cannot tell you much about motivations you never explored.

Use the one-sentence test. Describe each ad without mentioning the colour, actor, font or aspect ratio. If every description is “our bag keeps you organised”, your batch has execution variety. That can be a deliberate test, but it is a narrower test than it first appears.

Count the reasons to buy.
Then count the assets.

The opportunity for a lean team is to reduce the cost of exploring those reasons. A better-organised competitor can carry forward what it learns while your next brief is still waiting for a slot. That is a process advantage you can act on without guessing how much the competitor spends.

Build a matrix around the buyer’s decision.

Start with the purchase tension. Add a promise the product can support. Choose evidence that answers the most likely doubt. Only then choose the format.

For the commuter, the useful scene is reaching an everyday item. For the careful buyer, it is checking dimensions. For the design-conscious buyer, the bag’s appearance in a real setting may do more work. Reusing the same headline across those scenes would miss the point of creating them.

Do not turn the matrix into automatic multiplication. Three tensions, three hooks and three formats produce 27 combinations on paper. Some will be redundant; others will not fit the evidence. Select the few combinations that make a coherent argument before spending money to learn whether they work.

Also separate exploration from refinement. When you compare different concepts, several elements may change together. You are testing a whole creative argument. Once one shows promise, keep the offer and visual stable while testing its opening if you need to learn about the opening specifically.

Your budget decides how much you can learn at once.

Imagine a team has $600 available for a creative experiment and a historical acquisition cost around $40. A rough planning calculation gives 15 expected purchases if performance remains similar. Split the same money across 30 ads and there may be very little purchase information per ad.

That arithmetic is not a forecast or a sample-size calculation. Delivery will be uneven, costs can change and conversion counts are noisy. It simply makes the constraint visible: inexpensive creative production does not create unlimited testing budget.

Choose a manageable batch, a primary business metric and a review window that respects your conversion lag. Decide what would justify a second test and what would make a concept uneconomic. If you need a causal comparison between two treatments, use an appropriate controlled experiment. Normal delivery reports are observational.

Low delivery and poor response are different observations.

An ad that spent almost nothing has not received the same test as an ad that spent heavily. “The platform did not allocate much delivery” and “buyers saw it and rejected the argument” are different conclusions. Keep that distinction in your notes.

For ads with meaningful delivery, look beyond the first attractive metric. A provocative opening can improve clicks while worsening purchase quality. A clear demonstration can attract fewer clicks and more relevant buyers. Compare similar reporting windows and check downstream outcomes.

When a result looks strong, record what you actually know. “The compatibility demonstration acquired customers efficiently in this campaign during this period” is useful. “Compatibility is our best angle forever” is a story the data has not earned.

Before declaring fatigue, look for alternative explanations: changes in spend, offer, stock, tracking, placement mix or the landing page. Refreshing creative is sensible when the creative is the problem. It is an expensive distraction when the checkout is broken.

A creative sprint you can run next week.

1. Audit recent ads by buying argument, not filename.
2. Find one important customer question your current creative does not answer.
3. Gather evidence that can answer it accurately.
4. Produce a small set of coherent concepts around that question.
5. Record the test conditions, delivery and business outcome.
6. Carry one specific learning into the next brief.

You do not need to chase every theory about the algorithm. You need a reliable way to turn what customers care about into creative, then discover which arguments justify more investment. The teams that shorten that cycle have more opportunities to improve it.