A fashion brand running 14 interest-stacked ad sets was averaging 2.1 ROAS. They had been adding interest layers for 18 months based on the conventional wisdom that more targeting precision equals better results. When we collapsed the structure to 2 broad campaigns — no interests, no stacking, just Advantage+ audience with creative doing the targeting work — ROAS hit 3.8 in 30 days on the same budget. The account structure they'd been carefully building for a year and a half had been actively working against them.
Why interest stacking made sense in 2020 (and why it doesn't now)
Interest targeting on Meta worked well when the algorithm was less sophisticated. You'd tell Meta "show this to people who like yoga and health and wellness and meditation," and it would do exactly that — limit delivery to that defined population. The value was in constraining delivery to the most qualified audience before the algorithm had enough data to find those people on its own.
Meta's algorithm in 2026 is a different tool. It has transaction-level signals from billions of purchases, app installs, and lead submissions across the platform. It can identify who is likely to buy your specific product based on behavioral patterns far more granular than the interest categories you can select in Ads Manager. When you stack 5 interests, you're not giving the algorithm precision — you're giving it constraints that limit the population it can search within.
The math compounds against you: a 14-interest stacked ad set might reach 200,000 people instead of 10 million. Meta has to find your buyers within that smaller pool. If your buyers don't perfectly match the interest profile you've constructed — and they usually don't — the algorithm starves. It can't find enough purchase signal to optimize properly, so it fills delivery with the least objectionable people inside your constraints, not the most likely buyers it could find without them.
What Meta's algorithm actually optimises for today
Modern Meta optimization is conversion-signal-driven. The algorithm is asking: among all the people I could show this ad to, who is most likely to complete the conversion event I've been told to optimize for?
To answer that question well, it needs signal: purchase events with rich data (value, content IDs, customer email for matching). When you give it a broad audience and sufficient daily spend (minimum $50/day per campaign to generate learning), it can find your buyers across the full population. It will identify patterns you'd never construct manually — a combination of behavioral signals, content consumption patterns, and device usage that predicts purchase intent better than "interested in fashion."
Interest targeting restricts the search space. In 2020, that was useful because the algorithm's broad search was imprecise. In 2026, it's a handcuff.
How the fashion brand restructured
Before: 14 ad sets. Average spend $70/day each. Audience sizes ranging from 180K to 1.4M. All interests stacked — "fashion + sustainable clothing + ethical brands + women's apparel." ROAS: 2.1.
After restructure:
- Campaign 1 — Prospecting (broad): Single ad set, no interests, Advantage+ audience enabled. $400/day. Top 3 performing creatives. ROAS target: 3.0+.
- Campaign 2 — Retargeting: Single ad set, 30-day site visitors and video viewers (50%). $200/day. Dedicated retargeting creative (not repurposed prospecting).
Total spend unchanged: $1,380/day → consolidated to $600/day in the two new campaigns (remaining budget had been burning in the weakest interest-stacked sets).
Day 30 result: 3.8 ROAS at account level. The prospecting campaign alone ran at 3.4. Retargeting pulled 5.6. The fashion brand's previous "targeting precision" had been costing them 1.7 ROAS points.
When to use interests vs when to go broad
Two situations where interest targeting still makes sense:
Small budgets (below $100/day total). At very low spend, broad targeting gives Meta's algorithm too little daily data to optimise effectively. You might generate only 1–2 purchases per day, which isn't enough signal for the algorithm to learn. A narrow interest audience at least constrains delivery to a more qualified group while data builds. Once you're above $150/day consistently, broad targeting almost always outperforms.
Product-specific audiences with no behavioral signal. If you're launching a brand-new product with zero purchase history and zero pixel data, interests serve as a proxy while you build signal. Run interests for the first 30 days to seed the algorithm with some signal, then broaden once you have 50+ purchase events.
For established Shopify brands with 90 days or more of Shopify Pixel data and at least $150/day in spend: run a 4-week test. Broad campaign with your top creatives vs. your best interest-stacked campaign at equivalent spend. Check ROAS by week 3. The result will tell you more than any case study.