In 2012 I used to download spreadsheets of every search keyword and manually change bids by plus or minus 10%. That was the job. Six years later most of that work had been swallowed by Smart Bidding. I haven’t manually adjusted a keyword bid since 2019.

That same compression is happening on the creative side right now, and it’s happening faster. The work that used to win Meta (audience selection, bid management, placement testing) has been quietly absorbed by the model. What’s left is creative. And most brands haven’t restructured their creative function to deal with that fact.

Why creative is now the primary input signal

When the platform decides almost everything else, the only thing you genuinely control is what the ad says, shows, and how it opens. That’s not a slogan. That’s the literal architecture.

Meta’s own delivery systems now treat creative as a first-class ranking signal alongside the obvious ones like conversion history. Andromeda’s retrieval model uses creative embeddings (vector representations of what the ad actually contains) to match ads to users. Two ads with identical settings but different opening visuals will be served to materially different users. The model is reading the ad before it ranks it.

What that means in practice: the only sustainable competitive edge left on Meta is being good at making ads. Targeting is shared infrastructure now. Creative is the moat.

Volume vs. quality: why the question is a trap

Brands love to argue this one. “Do we need more ads, or do we need better ads?” The honest answer is both, and the question is usually a smokescreen for not wanting to commit to either.

Here’s the maths I actually run with clients. A high-performing concept will, on most accounts, generate roughly 40% of a campaign’s revenue while live. It will last seven to fourteen days at scale before the system fatigues it. To keep that 40% slot filled, you need a steady supply of concepts that have a chance of being the next high performer.

In my experience about one in five concepts becomes a meaningful winner. One in twenty becomes a step-change winner. So if you want one step-change winner a month, you need to ship twenty concepts a month. Not twenty variants, twenty concepts. The volume question and the quality question aren’t separate. Volume is how you find quality.

The brands losing this argument in 2026 are the ones still trying to perfect the hero ad in pre-production. Good ads are downstream of shipped ads.

The role of hooks, formats, and iteration cycles

Direct answer in 50 words: in a post-Andromeda system, what matters most is the first three seconds, the format the user is scrolling in, and how quickly you can iterate on what works. The hook signals to the model which user to show the ad to. The format determines watch-through. The iteration cycle determines how fast your winners compound.

Hooks have become the highest-impact work in a creative team, and the most under-resourced. Most brands invest 80% of their budget into the production value of the ad and 20% into the first three seconds. That ratio should be inverted. I’d happily ship an ad with mediocre middle and a brilliant opening. The middle is what people see when the system has decided the ad fits them. The opening is what makes the system decide.

Formats matter because the model factors placement performance into delivery. A 4:5 static deck might convert beautifully on Facebook feed and tank on Reels. The brands shipping the same square crop to every placement are paying a “lazy format” tax of roughly 15–20% on CPMs without realising it.

Iteration cycles are where the real money is hiding. If your winning concept takes four weeks to spawn three variants, the variants land after the original has fatigued. Your iteration loop has to be shorter than the fatigue cycle. For most brands above £50k a month that means a new variant of a winning concept within 72 hours.

How the delivery system rewards variation and penalises stagnation

This is the bit that hadn’t sunk in for me until I watched it play out across a dozen accounts. Andromeda doesn’t just tolerate a varied creative pool. It actively rewards it with cheaper delivery. The model gets more value from your inventory when there’s more for it to choose between. CPMs on accounts with rich creative pools run 12–18% cheaper than equivalent accounts running tight rotation. Same audience, same bid, different cost.

The other side of that coin is brutal. Accounts running the same five ads for six weeks see CPMs creep up week-on-week, even as performance flattens. The system isn’t punishing you for being lazy. It just has less to work with, so it has to serve your tired ads to less ideal users to spend your budget. The result looks the same: declining performance with no obvious cause.

A useful diagnostic for any account: pull the last 90 days, count how many distinct ads delivered 5%+ of total spend, and divide by the number of months. If that number is under five, your creative pool is starving the model. Below three, the model is actively working against you.

What this changes about how you commission and brief creative

The implication isn’t “hire a bigger creative team.” It’s “redesign the brief, the shoot, and the approval process around volume of distinct concepts, not perfection of one.”

In practice that means modular shoots that yield ten end-products from one day on set. Standing UGC pipelines with creators on retainer. Shorter approval loops: a 48-hour cycle, not a fortnight. AI tools used to multiply variants of proven winners rather than to generate the winners themselves. Creative directors briefed on hooks and formats first, hero films second.

The brands who get this right will be unrecognisable from the brands who don’t. Not because their ads are cleverer. Because their machine is.

If you’re trying to work out whether your creative function is set up for the system you’re actually running on, I’m easy to reach.