You build a lookalike audience from your buyers. You test it. CPA is a disaster, ROAS doesn't add up, and you end up going back to interests because "lookalikes don't work in your niche." Before you accept that, there's a question almost nobody asks: how many of the purchases you used as a seed can Meta actually identify?
How Meta builds a lookalike audience
Lookalike audiences work like this: you give Meta a set of people — your seed audience — and ask it to find similar profiles across its entire user base. The quality of the lookalike depends 100% on how well Meta can identify those people.
To identify someone, Meta needs match parameters: email, phone, name, city, postal code. It matches that data against its own profiles and builds the similarity model.
The problem is that if your browser pixel isn't sending those parameters, Meta receives the purchase event with no identity attached. It knows someone bought. It doesn't know who.
How many buyers can Meta really identify?
Here's the number nobody wants to see: on average, the browser pixel lets Meta identify between 40% and 60% of buyers. The rest arrive as anonymous events.
Why? Because the browser pixel runs in an environment that works against you:
- Ad blockers: between 25% and 40% of users in LATAM use them, especially on desktop. Meta's script is first on the block list.
- iOS and ATT: if your buyer used Instagram's in-app browser on an iPhone and declined tracking, the identity signal gets cut.
- Safari ITP: Safari deletes third-party cookies after 7 days. The buyer who saw your ad today and bought 10 days later arrives unidentified.
- Accelerated checkouts: one-click payments skip the checkout steps where identity data is captured.
- Email not captured: if the pixel fires the Purchase event before the user enters their details, the email never gets attached.
The result: out of 1,000 purchases in your store, Meta can correctly identify 400-600. The other 400-600 are ghosts in your seed audience.
The pixel sees that they bought. But it doesn't know who they are. Meta isn't losing your sales: it's losing half the identity data on every purchase. Low EMQ, bad lookalikes, higher CPA. That's what's happening.
What's the connection with EMQ?
Event Match Quality (EMQ) is the 1-to-10 score Meta assigns to your events based on how many match parameters arrive correctly. An EMQ of 4 out of 10 doesn't just affect attribution: it directly affects the quality of the data you use to build your audiences.
A seed audience built from low-EMQ events has these problems:
- Meta can identify fewer people → the seed is smaller than you think
- The people it does identify may be skewed (for example, only those who bought on desktop without an ad blocker)
- The similarity model Meta builds is less accurate
- The lookalike targets a profile that doesn't represent your real customer
You can read more about what EMQ is and how to raise it. The short version: more identity parameters on every event → higher EMQ → better lookalike.
How many parameters should you be sending?
Meta accepts these match parameters on the Purchase event:
- Email (em): the most important. With the email alone, Meta can identify the user in most cases.
- Phone (ph): the second most valuable, especially in Argentina, where a lot of people have WhatsApp linked to Facebook.
- First and last name (fn / ln): strengthens identification, especially when the email isn't available.
- City and postal code (ct / zp): improves the geographic precision of the match.
- fbp and fbc: Meta's own cookies. If the ad blocker cut the cookie, they don't arrive.
A typical browser pixel sends 2 or 3 of these parameters, and only when the environment doesn't block it. With server-side tracking, you can send every available parameter from the server — with no ad blocker intercepting it — because your server has access to the full order data.
What happens when you raise your EMQ
Going from an EMQ of 4 to an EMQ of 7-8 has concrete effects:
- Your lookalike's seed audience goes from ~450 identifiable people to ~850 out of 1,000 events
- Meta's similarity model has twice as much information to work with
- The lookalike targets profiles that actually resemble your buyers
- CPA drops because the traffic is more qualified from the start
You didn't change the strategy or the budget. You just gave Meta more data to do its job better.
How to improve the data that reaches your seed audience
Three concrete steps:
1. Implement the Conversions API (CAPI)
Meta's Conversions API lets you send purchase events directly from your server. It doesn't depend on the browser, ad blockers don't block it, and it can include every identity parameter on the order. It's Meta's official solution for recovering lost signal.
2. Turn on event deduplication
When you use CAPI alongside the browser pixel (the most common setup), Meta can receive the same event twice. Without proper deduplication, that inflates the data and confuses the algorithm. The fix is to use a unique, consistent event_id on both channels so Meta can collapse the duplicates.
3. Make sure the buyer's email gets through
Email is the most valuable parameter. Check that your pixel fires the Purchase event after the user has completed the checkout form — not before. In many setups, the pixel fires too early and the email never gets attached.
What Escalafy does here
The Escalafy server-side pixel sends Meta every identity parameter available on the order — email, phone, name, city, postal code — directly from the server, with automatic deduplication against the browser pixel.
But there's something no other tracking tool does: since Escalafy already knows the real cost of each product, shipping, Mercado Pago or Tiendanube fees and installments, every event that reaches Meta includes not just the sale but the real profit behind it. You can optimize your campaigns for real profit, not revenue.
You can see how it works on the Escalafy server-side pixel page.
FAQ
Do I need to turn off the browser pixel if I use CAPI?
No. Meta recommends running both in parallel. The browser pixel captures pre-purchase behavior data (PageView, ViewContent, AddToCart) that the server can't capture as precisely. The key is having deduplication active for the events both of them report.
What's the minimum number of events for a useful lookalike?
Meta recommends at least 1,000 high-quality events (EMQ 7+) to build a reliable lookalike. With fewer, the model doesn't have enough signal. If you have 1,000 purchases but an EMQ of 4, in practice Meta could only identify ~400 people — the model works as if you had 400 buyers.
Do value-based lookalikes work better?
Yes, when you have reliable value data. Meta can build a lookalike weighted by purchase value. The problem is that if the pixel sends the list price but not the real margin, Meta optimizes for buyers who spend more, not for the most profitable ones. With real profit data, the lookalike targets buyers who leave more margin.
How long does it take for performance to improve after raising EMQ?
The first 7-14 days are a relearning phase. Meta is updating its models with better data. After that period, new lookalikes built from the improved seed start performing better. Don't abandon the test before 2 weeks.
Does this also apply to retargeting audiences?
Yes. If Meta couldn't identify 600 of your 1,000 buyers, those 600 people don't end up in your exclusion audiences or your abandoned-cart retargeting audiences either. You're paying to show ads to people who already bought, because Meta doesn't know they bought.
The problem isn't your lookalike strategy. The problem is that Meta has half the data it needs to build them. Raise your EMQ, improve the seed, and lookalike audiences start working the way they should.
Escalafy helps you recover that signal and see the real profitability of every campaign. See how the server-side pixel works.
Keep reading
What Escalafy does for your store
Connect your channels and see the truth behind your numbers. No Excel, in real time.