The accounts still running pixel-seeded lookalikes as their primary cold prospecting tool are paying for something that degraded four years ago and has not recovered.
Facebook lookalike audiences work. The mechanism is sound: take a seed of your best customers, find people who resemble them across Meta's network. The problem is the signal quality of what goes into that seed. A pixel that stopped reliably seeing iOS conversions in 2022 makes a noisy seed. A noisy seed makes a noisy lookalike.
The finding: Lookalike audiences perform in direct proportion to seed quality. Pixel-seeded lookalikes have degraded since iOS14 signal loss. First-party data seeds - purchase lists, CRM segments, high-value customer exports - still produce competitive cold audiences. The accounts that abandoned lookalikes entirely have overcorrected. The accounts still relying on pixel events as their only seed have not caught up.
What we actually measured
This draws from account patterns we track at BAVS across DTC/ecommerce and SaaS/B2B accounts running Meta prospecting at varying spend levels. The observations are directional from first-hand operating experience across rolling 90-day windows - not a controlled trial.
We compared three seed types across cold prospecting campaigns: pixel event-based audiences (standard Purchase or Add to Cart events), first-party data uploads (customer purchase lists matched against Meta's identity graph), and Advantage+ Shopping campaigns with no explicit audience restriction. The pattern that emerged is consistent enough to warrant a framework for how to rank and use each option.
Pixel-seeded lookalikes carry signal noise they did not have three years ago
A pixel that tracked iOS conversions accurately in 2021 saw somewhere between 30-60% fewer directly attributed conversion events after Apple's App Tracking Transparency changes rolled through in 2022. The exact figure varies by account type, device split, and whether the advertiser implemented server-side tracking via the Conversions API.
What this means practically: the "Purchases" event powering your standard 1% lookalike now contains a meaningful share of modelled conversions rather than direct matches. Meta's modelling is sophisticated. It is not the same as a clean signal.
When the seed has noise, the lookalike audience built from it inherits that noise. A 1% lookalike of a clean 10,000-customer purchase list is structurally different from a 1% lookalike of a pixel Purchase event list that includes a 40% modelled component. Both are called the same thing in Ads Manager. Only one is built on clean signal.
First-party data seeds still perform where the pixel has weakened
First party data strategy ads starts with the seed, not the campaign type. A matched customer upload - 5,000 or more past purchasers matched against Meta's identity graph - gives the algorithm a starting point that does not depend on pixel event quality.
Match rates on a clean CRM export typically land between 50-70% for AU and US consumer audiences. That means 5,000 uploaded customers might produce 3,000-3,500 matched profiles. From that foundation, a 1-3% lookalike works against Meta's total audience to find profiles sharing modelled characteristics with your actual buyers - without the iOS signal noise built in.
The performance difference is not always dramatic at first glance. But we see it consistently in accounts with high iOS device penetration and narrower post-click attribution windows. The first-party seeded lookalike holds CPA more steadily through frequency cycling. The pixel-seeded version drifts faster and requires earlier creative rotation to maintain CPA targets.
Advantage+ has absorbed part of what lookalikes used to do
Lookalike audiences have not stopped working. The signal feeding them has.
The honest facebook lookalike audiences guide for 2026 includes acknowledging what Meta's own algorithm now does without explicit audience targeting. Advantage+ Shopping Campaigns run with no audience restriction. Meta's delivery system finds people most likely to convert based on the conversion goal and the account's accumulated signal. For accounts with 50 or more monthly purchase events, the algorithm has enough data to find its own "lookalike" without being told where to start.
We have seen accounts where removing lookalike targeting from cold prospecting and shifting budget to Advantage+ held or improved CAC. We have also seen accounts where that move collapsed performance - typically accounts with fewer than 30-40 monthly conversions, thin creative libraries, or product categories where broad algorithmic targeting cannot surface the right buyer profile.
The line is clearer than most guidance suggests: Advantage+ works when the algorithm has enough signal to self-navigate. Cold vs warm audience facebook ads structure in an Advantage+ world means supplying the algorithm with conversion events, creative variety, and sufficient budget to optimise - not restricting it with an audience definition that may be noisier than its own model.
| Seed Type | Signal Quality | Best Use Case |
|---|---|---|
| Pixel Purchase event (post-iOS) | Degraded - modelled component | Use only when no CRM data exists |
| Full CRM upload (all buyers) | Clean direct match | Standard cold prospecting seed |
| High-value segment (top 20% by LTV) | Strongest available signal | 1-3% LAL for scaling campaigns |
| Advantage+ (no targeting restriction) | Algorithm self-navigates | Accounts with 50+ monthly conversions |
Where this falls apart
The case for replacing lookalikes entirely breaks for accounts below $10,000 monthly spend with fewer than 200 purchases on record.
Advantage+ needs data to self-navigate efficiently. A 1-3% lookalike of a 500-customer CRM list still constrains delivery more than Advantage+ does, but the constraint comes with a benefit: when the algorithm lacks enough conversion signal to self-direct, an audience seed gives the delivery system a starting point that prevents broad, inefficient scatter spend while the account learns.
For cold vs warm audience facebook ads strategy at lower spend levels, a first-party seeded lookalike at 3-5% is often the more cost-effective cold channel than Advantage+. The audience restriction keeps spend focused while the algorithm accumulates the conversion data it eventually needs to operate without guardrails.
First party data strategy ads also pays forward at this scale: even a 1,000-person purchaser list is more signal than a noisy pixel event audience. A brand with 200 confirmed buyers and an engaged email list of non-buyers can combine both into a seed that gives the algorithm more to work from than pixel events alone.
What the Seed Quality Ladder means for Monday
The Seed Quality Ladder ranks your prospecting inputs from weakest to strongest signal. Start from the best you have available:
- Pixel Purchase event seed - still functional if no CRM data exists. Treat any performance benchmarks from pre-2022 as irrelevant baselines. Expect more variance.
- Full customer CRM list - 5,000 or more past purchasers uploaded and matched against Meta's graph. Replace the pixel event seed with this as your standard cold prospecting input wherever possible.
- High-value customer segment - your top 20% by LTV or purchase frequency. Run this as a 1-3% lookalike alongside the full-list version. If CPA holds at or below target for 14 days, this becomes the primary seed.
- Advantage+ with conversion-optimised creative - run parallel to your best lookalike campaign. At 30-day intervals, compare blended CAC against your target ceiling. Let the numbers determine which stays.
BAVai runs seed-versus-Advantage+ performance comparisons across every account we manage on a rolling 30-day basis. When any cold prospecting campaign drifts above 3x target CPA for three consecutive days, the flag surfaces before the spend decision becomes expensive. The comparison is always live, not a monthly retrospective.
The cold vs warm audiences guide covers how to structure the full funnel so cold and warm campaigns do not compete for delivery. The ad account health check covers whether your conversion signal is clean enough to seed meaningful lookalikes. For accounts where prospecting strategy needs a systematic rebuild, the BAVS Meta ads service page covers how we approach the cold channel from seed quality up.
If you compared the CPA of your current lookalike campaign against Advantage+ running the same creative pool today - which would you trust more, and do you actually have that comparison in front of you?