Why This Niche Streetwear Brand's Ads Kept Failing — Even With a New Agency, New Creatives, and a New Budget

A niche anime streetwear brand had been through three to four agencies before onboarding with a new growth partner. Every agency had run campaigns. Every campaign had failed. The reason wasn't bad creatives — it was that years of broad, generic targeting had trained the Meta algorithm on entirely the wrong audience. The pixel wasn't just underperforming. It was actively misleading every campaign that ran on top of it. Fixing performance required rebuilding the data signal from scratch before any creative work could land.

The brand had a genuine product in an underserved niche. The target audience was specific, passionate, and real. The catalog was strong. Three to four previous agencies had run campaigns against it — and every single engagement had ended the same way: spend going out, wrong people coming in, ROAS going nowhere.

By the time a new agency inherited the account, the working assumption was that this was a creative problem. The ads weren't good enough, the hooks weren't right, the format needed work. So new creatives were briefed, new campaigns were launched, new budgets were committed.

And for two weeks, the results were almost identical to everything that had come before.

The 4-day ROAS hit 0.58. The account was spending more than it was making. The founder was watching the same pattern begin to repeat for the fifth time.

The problem wasn't the creatives. It never had been.

BRAND SNAPSHOT

• Industry: Niche anime & manga streetwear (D2C, India)

• Category: Premium anime-themed apparel — graphic tees, hoodies, joggers, jackets, jerseys

• Geography: India (pan-India, Shopify)

• Stage: ₹4L/month with ROAS 2.5x — generating revenue but not profitably, with audience data compromised by years of broad targeting

• Services: Meta Ads, Creative Strategy, Audience Rebuild, Interest-Based Targeting, Conversions API Integration, Microsoft Clarity

THE PROBLEM

The brand came in with a history most founders in its position would recognize: multiple agency engagements, each one burning budget on campaigns that brought traffic — but not buyers. Not the right buyers.

After onboarding with a new agency, ads went live within five days. The first two weeks of data told a familiar story. High add-to-cart counts. Low purchase conversions. ROAS fluctuating between 0.55 and 1.3 on any given day. Nothing holding.

The working diagnosis in week one was creative. More formats were tested — video, static, carousels, anime-inspired POV hooks. Performance didn't stabilize. The 7-day ROAS on March 1 sat at 0.79. The 4-day ROAS sat at 0.58. The account was in the red.

But the creative wasn't the root problem. The root problem had been accumulating for months before the new agency ever touched the account.

WHY IT WAS HAPPENING

Every time a previous agency ran a broad awareness campaign — targeting generic fashion interests, wide age brackets, mass-market lookalikes — they drove clicks from people who had no genuine connection to anime streetwear. Someone who clicked on a hoodie ad because it showed up in their feed is not the same as someone who wears Vinland Saga merchandise because they've watched the show three times.

But the Meta algorithm doesn't know that distinction. It only knows who clicked, who added to cart, who initiated checkout, who bought. Every interaction from a wrong-fit buyer was fed back into the pixel as a signal about who this brand's customer is. Over months and multiple agency engagements, the pixel had been trained on a composite portrait of the wrong person.

By the time the new agency inherited the account, the pixel wasn't neutral. It was pointing in the wrong direction. Any campaign launched on top of that data — regardless of how good the creative was — would start by reaching the audience the algorithm had already decided was most likely to convert. That audience was wrong. And every new campaign that ran on a polluted signal would continue polluting it further.

This is the hidden cost of agency churn that almost no brand accounts for. The wasted spend is visible. The corrupted data layer it leaves behind is not. But that data layer is what every future campaign learns from.

THE SOLUTION

Sentinel — Scientific Media Buying:

The strategic decision was to stop running on inherited assumptions. New conversion campaigns were built with interest-based targeting — specifically targeting anime-aligned interests, not broad fashion demographics. The goal wasn't to reach everyone who might theoretically buy a hoodie. It was to reach the specific subset of the Indian internet who had already demonstrated an affinity for the cultural references on the products.

This was a deliberate reset. Not optimization of existing campaigns — a rebuild. The new campaigns isolated the best-performing creative and the highest-intent interest signals found so far, stripped away everything else, and started building a fresh signal layer.

Mythos — Creative Advantage:

Alongside the targeting reset, a parallel creative hypothesis was tested: that this audience — already primed with fandom, already emotionally connected to the specific properties on these products — didn't need storytelling. They needed a clear product view and a reason to act today.

A static image was prepared. Clean product shot, offer copy, no production overhead. It was running in the new interest-based campaigns within days of the rebuild launch.

On March 5, the first directional signal emerged: "The static image is showing a positive sign."

On March 6, the data was unambiguous: the static image was running at over 6x ROAS while every video format hovered below 1.5x.

The 7-day ROAS, which had sat at 0.87 on March 5, climbed to 1.05 by March 7, 1.46 by March 8, and 1.67 by March 11. The account had gone from 0.58 ROAS at its worst to 1.88 on a 4-day window in ten days.

Vault — Brand Value Engine:

The interest-based rebuild did more than improve ROAS. It revealed information the polluted account had been hiding. With clean signals flowing in from an audience that actually wanted the product, two SKUs separated themselves clearly and quickly: one product consistently delivering 2.3–2.9x ROAS, another hitting 3.3x. These became the creative pillars — every subsequent brief, every new reel, every offer static was built around what the data had now confirmed.

The same data accuracy infrastructure that enabled the targeting rebuild — Conversions API reconnected through Shopify, Microsoft Clarity for on-site behavior analysis — also fixed the gap between ad-side reporting and real purchase events. The account was now operating on clean inputs and accurate outputs simultaneously.

Five weeks after the rebuild, the clearance sale launched. It opened at 15x ROAS in the first few hours.

THE RESULTS

• 4-day account ROAS: 0.58 at the low point — inherited from polluted data conditions

• 6x ROAS on the first static creative after the targeting rebuild — March 6, while all video ads remained below 1.5x

• 91% improvement in 7-day account ROAS in 6 days — from 0.87 on March 5 to 1.67 on March 11

• Hero products unlocked: two SKUs identified through clean-data conditions at 2.3–3.3x ROAS — invisible in the corrupted account

• Clearance sale ROAS: 15x on launch morning — the result of five weeks of clean signal accumulation

• ₹8L+ in one week — the brand's first scalable revenue event

LESSONS FOR SIMILAR BRANDS

• When a new agency inherits a bad account, the pixel comes with it. The new agency, new creatives, and new budget are all starting from a data foundation shaped by everything that ran before. If previous campaigns targeted the wrong audiences, the algorithm has learned wrong patterns. A targeting reset — not just new creatives on top of old campaigns — is often the first real intervention required.

• The length of your agency history is not the measure of your data quality. More campaigns do not mean better data. Three agencies running broad campaigns for eight months can leave an account with less usable signal than a single well-structured campaign running for six weeks against the right audience.

• For community-driven niche brands, interest-based targeting is not optional. A broad fashion audience and an anime streetwear audience are not different segments of the same market. They are different markets. An algorithm trained on one cannot find the other efficiently. The specificity of the niche is an asset in targeting — not a constraint.

• Good creatives on a polluted data signal will still underperform. The most common misdiagnosis when a new campaign doesn't work is creative. Before testing the twentieth creative variation, ask whether the audience the algorithm is reaching is actually the audience you're selling to.

CHALLENGES WE FACED

• Two weeks of below-benchmark performance before the root cause was addressed. The initial creative-testing phase was necessary — the interest-based rebuild wasn't launched immediately, and the data from those early campaigns, even at low ROAS, helped identify which creative formats and interest combinations were directionally right. But the founder was watching spend go out without results. Maintaining confidence through a deliberate rebuild phase — rather than making reactive changes — required clear communication about what was being done and why.

• High add-to-cart counts made it harder to diagnose the real problem. Throughout the early period, the campaigns were generating strong add-to-cart signals — which looks like intent. But those add-to-carts weren't converting, which pointed to either a funnel problem or an audience problem or both simultaneously. Separating the data layer issue from the conversion infrastructure issue required running both fixes in parallel: targeting rebuild and CAPI integration on the same timeline.

BELIEFS CHANGED

• "We've tried paid ads — they don't work for us." This brand had tried paid ads four times with four different agencies. The product was real, the audience existed, the category had demand. What didn't work was running generic campaigns against a generic audience in a category that requires a specific audience. The clearance sale at 15x ROAS on launch morning didn't happen because the product changed. It happened because the data finally reflected who was actually buying it.

• "If the new agency isn't performing, we need new creatives." Creative is the most visible lever and the easiest to change. But creative quality is bounded by audience accuracy. The 6x ROAS static image was not a better piece of creative than everything that came before it — it was the same level of creative work reaching a fundamentally different audience. The targeting rebuild was the intervention. The static was the first clean signal that it had worked.