One winning ad is no longer enough to win peak

Performance marketing teams run the same ritual heading into peak: find the ad that’s proven itself, then pour the budget behind it. A winning ad still does what a media team wants, since it validates the creative and earns the right to scale hard through peak’s biggest weeks. But leaning on that single winning ad to carry all of peak is no longer enough.

What’s changed is who decides which audience sees that ad next. Meta’s Advantage+, TikTok’s Smart+, and Pinterest’s Performance+ now handle audience selection by default, using whatever creative they’re fed as their main signal for who to target. Brands that enter peak leaning too heavily on one winning ad give the algorithm fewer creative signals to work with as it looks for new pockets of demand, a pattern peak season ad performance data across hundreds of ecommerce brands bears out.

One ad isn’t enough anymore

Under the old model, a marketer chose the audience, then judged the ad by how it performed with that audience, so one great ad could carry a campaign a long way. Under the new model, the platform chooses the audience, using the creative it’s given to find pockets of demand a marketer would never have identified manually. Creative is still a performance lever. Now it’s doing double duty as a targeting input too.

That’s what changes the maths on staying with one ad. A single ad, however well it’s converting, is still one hypothesis about who to reach and why. A wider set gives the same algorithm more hypotheses to test, and more chances to find demand outside what any one ad could reach.

The old logic for scaling a winner doesn’t work anymore

If one ad is winning, doesn’t scaling it make the safer bet? Isn’t a wider range just spreading the same budget thinner?

Scaling a proven ad used to mean putting more budget behind an audience and message that had already shown signs of working. That made sense back when a brand chose who saw it. More budget can expand reach, but the platform is still working from the same message, motivation and proposition. Staying with one message just adds spend behind a hypothesis the algorithm has already tested, instead of giving it anything new to learn from.

Give the platform more to work with

There’s no fixed rule for how much range is enough. Niket Shah, co-founder of Acceler8 Labs, offers one practical starting point. It’s 10 to 20 assets, spread across three things: persona, angle, and format.

In his own words, quoted in Fospha’s Peak Playbook 2026: “We used to build audiences and attach creative. Now we build creative and let it find the audience.”

The practical version of that shift starts even earlier, in the brief itself. Instead of opening with who the target audience is, the brief opens with what problem that customer is facing right now.

Persona variation means pitching the same product differently to, say, a gift-buyer versus someone shopping for themselves. For angle, that means testing a price-led hook against a quality-led one, or urgency against social proof. Format is the most visible of the three: a UGC-style clip in the same test pool as a static product shot and a founder-led video.

Five versions of the same message in five fonts test only one hypothesis. A real range tests different reasons to buy. What makes a range work is the number of different hypotheses inside it, rather than the raw count of assets. Vervaunt’s testing with Meta found the same pattern at scale: campaigns running more ad variants saw a 7.3% improvement in CPA, but only when those variants represented distinct ideas rather than simply adding volume. The risk is treating variation in format as enough, while keeping the persona and underlying reason to buy unchanged.

The range is the strategy now

None of this requires twenty new ideas from scratch. Two or three personas, a handful of angles for each, and two or three formats per angle can add up fast. The winning ad stays in the mix as one persona-angle-format combination inside a larger set, running alongside the rest, rather than getting dropped.

A winning ad is still a useful signal. The risk is treating it as the whole strategy instead of one input among several the platform can learn from.

Once targeting runs on automation, the stronger strategy is giving the platform more than one reason to find the next customer.