Why Your Lookalike Audiences Stopped Working (And What To Do Instead in 2026)

Your lookalike audiences used to be a reliable scaling lever. You'd upload a customer list, Meta would find 100,000 similar profiles in seconds, and you'd see predictable ROAS. Then something broke—and it wasn't your ad creative.

Lookalike audiences stopped working because Meta's lookalike algorithm is now fighting a losing battle against first-party data loss, iOS privacy changes, and the sheer saturation of competitors all bidding on the same 1% lookalike tier. In this post, we'll show you exactly why this happened and the four audience strategies that are actually winning for eCommerce brands right now.

---

Why Did Lookalike Audiences Actually Stop Working?

Meta's lookalike algorithm does one job: find users who look like your best customers based on behavioral and demographic patterns. But that algorithm's inputs have degraded since iOS 14.5 rolled out in 2021, and the problem has only accelerated through 2025 and into 2026.

The iOS Data Loss Problem Is Still Crushing Lookalike Accuracy

Four years after iOS 14.5, we're still feeling the aftershocks. Here's what actually happened:

• Pre-iOS 14.5: Meta could see user behavior across 50+ websites and apps simultaneously. When you uploaded a customer list, Meta could say: "These 5,000 customers visited fashion blogs, clicked on YouTube ads for luxury goods, and engaged with Instagram reels about sustainable fashion. Find more like them."

• Post-iOS 14.5: Meta can only see on-site behavior for roughly 60% of your iOS traffic. The other 40%? It's a black box. Your lookalike source audience now contains incomplete behavior data, and Meta is being forced to guess about who these people actually are.

The result: Your 1% lookalike tier in 2026 includes 15-25% noise that used to be filtered out. That noise drives up CPC, tanks ROAS, and kills profitability on scaled campaigns.

---

The "Oversaturation Trap": Why 1% Lookalikes Are Bidding Like 5% Lookalikes

When lookalike audiences worked, they worked because they were exclusive. A 1% lookalike audience in 2019 might have had 200,000 people. Today, it has 400,000-500,000 people. But that's not the real problem.

The real problem: Everyone in your category is bidding on the same 1% lookalike tier.

A fashion eCommerce brand with €50K/month ad spend will go after:

• 1% lookalikes (their best performing)
• 3% lookalikes (secondary expansion)
• 5% lookalikes (testing scale)

But so will their competitor with €75K/month spend. And their competitor with €25K/month spend. And the five dropshipping brands trying to undercut everyone.

What happens: The 1% lookalike tier becomes a bidding war. CPM on lookalikes can swing from €2.50 to €8.50 in a single week because of auction density. Your historical ROAS of 3.5x becomes 2.1x. You cut the campaign. Your competitor scales. Everyone loses.

Smaller Source Audiences = Weaker Lookalikes

There's also a technical reason lookalikes are breaking down: most eCommerce brands are building lookalikes from source audiences that are too small.

Meta's algorithm needs density to find accurate patterns. If your "best customers" source audience is 300 people (maybe your VIP repeat buyers), Meta is extrapolating from a tiny sample. With iOS data loss, that extrapolation becomes speculation.

We've seen brands using:

• 500-person custom audiences as lookalike sources ❌
• 1,500-person audiences (the old "gold standard") ❌
• 3,000+ person audiences built from recent high-intent actions ✅

The ones winning are the ones with larger, cleaner source audiences.

---

What Strategy Should You Use Instead? (4 Replacements That Work)

Here's what's actually winning for our €5K-€100K/month clients right now:

1. Intent-Based Audience Stacks (Website Visitor Behavior Layers)

Stop thinking about who your past customers look like. Start thinking about what behavior signals predict conversion.

Build audiences around:

• Add-to-cart visitors (last 30 days) — signals serious intent
• High-value page visitors (product pages, pricing pages) — signals genuine interest
• Repeat visitors (visited 3+ times in last 60 days) — signals research behavior before purchase

Example structure for a €20K/month beauty brand:

• Primary audience: Add-to-cart visitors + High-value page visitors (combined: ~80K users)
• Secondary: 30+ day website visitors with engagement events (click, video play)
• Tertiary: Cart abandoners from last 90 days

Why this works: You're targeting proven behavior, not guessed similarity. iOS data loss doesn't affect on-site behavioral tracking nearly as much. And you're competing with fewer bidders (most brands still obsess over lookalikes).

Expected ROAS: 2.8x - 4.2x (better stability than lookalikes, less competition)

2. Value-Based Lookalikes (Not Conversion-Based)

If you're going to use lookalikes, build them the right way: use high-LTV (lifetime value) customers as your seed, not all converters.

Most brands build lookalikes from "All Purchases, Last 90 Days" — which is a bucket that includes your €10 one-time buyers and your €500 repeat customers in the same pile.

Instead:

1. Segment your customer list by LTV: Who are your top 20% spenders over the last year?
2. Create a custom audience from just those high-LTV customers (usually 500-2,000 people for €20K+/month brands)
3. Ensure minimum size: If you have fewer than 500 high-LTV customers, wait or expand the time window. Meta needs density.
4. Create a lookalike from only this segment

Example: A skincare brand we worked with had 40,000 total customers. Their top 1,000 customers (2.5%) accounted for 35% of annual revenue. They built a lookalike from just those 1,000 people.

Result: The value-based lookalike outperformed their old conversion-based lookalike by 31% ROAS over 4 weeks. But it's also smaller (~150K audience instead of 300K), so they can't scale it alone.

Use case: Pair this with intent-based audiences. Value-based lookalikes are your "explorer" budget (20-30%), intent audiences are your "optimizer" budget (70-80%).

3. Advantage+ Shopping Campaigns With Clean First-Party Feed Data

This isn't technically an audience strategy — it's a format shift that makes traditional audience targeting less important.

Advantage+ Shopping Campaigns (ASC) let Meta optimize across its entire platform (Feed, Reels, Marketplace, Audience Network) using only your product feed and conversion data as signals.

Why it's replacing lookalike reliance:

• Meta's algorithm learns faster from product data than from audience similarity
• You're competing on product relevance, not audience overlap
• iOS data loss affects ASC less because it relies on on-site signals and feed signals

To make it work:

• Feed quality is 80% of the battle. If your feed has poor images, wrong categories, or incomplete data, ASC will tank.
• Let it learn for 2+ weeks before pulling the plug (this is critical)
• Budget minimum €1,200/week per campaign (Meta needs volume to learn)

Real example: A €15K/month home goods brand shifted 60% of their lookalike budget to ASC. Month 1 was messy (ROAS 1.8x). By week 3-4, ROAS recovered to 2.9x. By month 2, it hit 3.4x. They kept it.

4. Predictive Audiences Based on Engagement Scoring

Meta's Predictive Audiences use machine learning to find users most likely to convert, based on how they've interacted with your ads historically.

This is underused, and we think that's a mistake.

Meta can now predict:

• Likely to purchase: Users who engage with your ads but haven't converted yet
• Likely to spend more: High-intent signals in ad engagement
• Likely to become repeat customers: Behavioral patterns post-purchase

How to set it up:

1. Go to Audiences → Custom Audience → Website Traffic
2. Create audiences for specific behaviors: video play (3+ sec), page visits, add-to-cart
3. In campaign setup, use "Predictive Audiences" to find similar high-intent users
4. Test at 10-15% of total budget to start

Result: You're letting Meta's algorithm do the heavy lifting, but you're feeding it first-party intent signals instead of vague similarity scores.

---

The Hybrid Model That's Actually Working (For Real)

Here's the framework our best-performing €20K-€100K/month clients are using right now:

| Audience Type | Budget Allocation | ROAS Range | Purpose | |---|---|---|---| | Intent-based (website visitors) | 50% | 2.8x - 4.2x | Optimizer | | Value-based lookalikes | 20% | 2.5x - 3.8x | Explorer | | ASC or predictive audiences | 20% | 2.4x - 3.5x | Scale | | Broad testing / new audiences | 10% | 1.5x - 2.5x | Discovery |

Implementation: Test this split for 2-3 weeks. Track ROAS and CAC weekly. If intent-based audiences outperform (they usually do), push more budget there. If value-based lookalikes surprise you, keep them. But never put more than 30% of budget into any single lookalike audience again.

---

Why You Should Still Keep Testing Lookalikes (But Differently)

We're not saying "never use lookalikes again." We're saying use them strategically, not as your default scaling lever.

Keep testing lookalikes if:

• You have a source audience of 2,000+ high-quality conversions
• You're testing at 15-20% of total budget (not 50%+)
• You're monitoring ROAS weekly and killing underperformers within 7-14 days
• You're combining them with intent-based audiences, not betting the farm on them

Kill lookalikes if:

• Source audience is under 1,000 conversions
• ROAS has trended down 25%+ over the last 4 weeks with no external reason
• You're running the same lookalike audience for 3+ months without testing new seeds
• You have zero first-party intent data to layer in

---