Most DTC brands don't have a creative strategy. They have a content calendar.

I run creative strategy across eight DTC accounts. Health tech, supplements, government services, consumer finance. Every month I see the same thing from the inside.

Teams produce 20, 30, 40 ads. They call it a "test." Most of those ads are the same concept in different clothes. New headline, same background. Different creator, same talking points. Color swap on a static that already ran last month.

Meta's algorithm treats those as one ad. Not twenty. One.

That gap between volume and actual diversity is where almost every brand under $500K/month in spend is bleeding money without knowing it.

The algorithm changed. Most creative workflows didn't.

Meta's Andromeda retrieval system rolled out over 2025 and changed how ads get matched to people. It scans creative at the visual, audio, and text level. When it picks up two ads that look, sound, or feel similar, it groups them under the same Entity ID and routes them to the same people.

So your 30 ads with minor variations hit the same 200,000 people. CPMs rise. Frequency spikes. Your "test" generates one data point, not thirty.

Motion's 2026 Creative Benchmarks report (550,000+ ads, $1.3B in spend) confirmed what people running accounts already felt: roughly 5% of ads become real winners, about 6% of creatives capture the majority of spend, and half of all ads never get meaningful distribution.

Performance advertising is a probability game. More genuinely different tests = more chances to find the 5% that carry the account.

Key word: genuinely different.

Whispers vs. screams

Two terms I use constantly with clients.

A whisper is a lazy iteration. Changing the headline color. Swapping one adjective. Tweaking button copy. To Andromeda, same creative signal. You're teaching the algorithm nothing.

A scream is a genuinely different psychological territory. Different awareness state (problem-aware vs. solution-aware). Different cognitive bias (social proof vs. loss aversion). Different emotional register (fear vs. aspiration vs. clinical).

Most brands never map this out. They produce ads by format ("we need five UGC videos and five statics this month") instead of by psychological territory.

That's a content calendar.

The Infinite Matrix

I built a system called the Infinite Matrix to fix this at the production level.

One axis is pain points. 8 to 12 psychological territories pulled from real customer language: Reddit threads, Amazon reviews, TikTok comments, support tickets. Not marketing copy. The words people use when they're frustrated, curious, or ready to buy.

Other axis is ad formats. 25 of them. 15 video styles and 10 static styles, from founder-to-camera and caption/b-roll through meme formats and comparison charts.

Every cell asks one question: "Can I make a [format] ad about [pain point]?" Yes? That's a brief. Check the box. Try it below.

25 formats x 12 pain points = 300 possible combinations. At 50% viability, that's 150 unique briefs from one matrix build. At 50 ads per month (the minimum for statistical significance), one matrix feeds 3 months of production before you rebuild anything.

Diversity isn't a mindset. It's a grid.

Why this works mechanically

Pull from different rows AND different columns each month and you're producing screams on autopilot. The grid enforces it. No separate "diversity check" needed.

When a pain point wins in one format, the matrix shows every other format you haven't tested with that angle. "Feeling fatigued all the time" worked as caption and b-roll? Interview style, before/after, product demo, founder-to-camera are all sitting there unchecked. Each is a new brief in 10 minutes. Each generates a new Entity ID.

Recombination math

You don't need 50 new concepts per month. You need to recombine what already works across new dimensions.

3 winning angles x 3 formats x 3 hooks = 27 unique combinations from 9 elements you've already validated. That's how 50 ads/month becomes real without 50 new ideas.

Where your columns come from matters more than anything

The matrix is only as good as its column headers. If your pain points read like marketing copy ("advanced formula for daily wellness"), every ad it produces will be generic. Andromeda can't categorize it. Nobody clicks.

Columns come from real customer language. I shorthand it as IVOC (In-Direct Voice of Customer). Fancy name for a simple thing: go read what real people say about their problem.

First is filler. Second is from an actual Amazon review. It has a scenario, a real emotion, a human cadence. Andromeda can match that specificity to the right person because it sounds like something that person has felt.

30 minutes per client per month on Reddit, Amazon, and TikTok comments. 10 raw phrases. Those become your column headers. AI can expand variations around them, but the seed language has to come from real people. Always.

How it compounds

Month 1 is the grind. Everything net new. Building from scratch.

Month 2, you have data. You know which pain points and formats the algorithm is spending on. You start decomposing winners into parts.

Month 3 onward is where it gets fun. 30%+ of your batch comes from recombining proven elements. Hit rate goes up. Effort per ad goes down. The system starts feeding itself.

The health tech wearable account I've run the longest scaled from near-zero US conversion spend to $170K+ in tracked revenue over the past year. 58% CPA reduction. 10x spend scale. The matrix wasn't the only reason, but it was the production backbone. Diverse output, every month, without burning out or running the same angles into the ground.

The uncomfortable part

Every batch needs 1-2 ads that make you cringe. Too simple. Too weird. Too ugly.

The matrix has room for these because "meme format" and "stupid simple text-heavy" are rows in the grid just like "founder-to-camera" and "product demo." They aren't afterthoughts.

The ad you personally hate might become the highest-spend winner. I've seen this across multiple accounts. The ugly stuff works because it looks like organic content. Andromeda doesn't care about your brand guidelines. It cares about whether someone stops scrolling.

If every ad in your batch feels safe, you're leaving the widest audience segments untouched.

If you're a DTC brand spending $10K-$500K/month on Meta and your account feels stuck, the problem is probably not your media buying. You're feeding the algorithm whispers and wondering why it can't find new people.

The brands that win the next 12 months won't produce the most ads. They'll produce the most different ads. And "different" has a very specific, mechanical definition now that the algorithm is doing the sorting.