Your Lookalike Audience Is Only as Good as the Data You Feed It — Here's How to Fix the Seed

Meta’s Lookalike Audiences are one of the most powerful targeting tools in paid social advertising. Give Meta a list of your customers, and it will find millions of people on Facebook and Instagram who share similar characteristics, behaviours, and demographics.

The problem is not the tool. It is the seed.

Most businesses create Lookalike audiences from their entire customer list — everyone who has ever purchased, regardless of how much they spent, how often they came back, or how profitable they were. Meta dutifully finds more people who look like all of those customers — including the one who bought once in a sale, left a negative review, and requested a refund.

The quality of a Lookalike audience is determined entirely by the quality of the seed you feed it.

What a Low-Quality Seed Produces

A seed list of all purchasers contains multiple customer types that pull the Lookalike model in different directions:

• One-time buyers: people who responded to a discount and never returned
• Bargain hunters: customers whose purchase was price-triggered, not brand-loyal
• Occasional buyers: moderate spend, moderate frequency, no clear LTV signal
• Refunders: people who completed a purchase but did not keep the product
• High-LTV customers: your best customers, who may represent only 15–20% of the list

When all of these are pooled together as a seed, Meta averages the characteristics. The Lookalike audience it creates reflects the average customer — not the best one. You are scaling mediocre acquisition, not finding more of your highest-value buyers.

How to Define a High-LTV Seed

The ideal Lookalike seed is the smallest, highest-quality list you can build. Meta recommends a minimum of 100 people for a Lookalike to generate, and performs best with 1,000–50,000 contacts. Above 50,000, the signal quality tends to dilute.

Define high-LTV using one or more of these filters in your CRM:

1. Purchase frequency: customers who have bought 3 or more times within 12 months — repeat buyers are far more predictive of future value than one-time purchasers

2. Total spend threshold: your top 10–20% of customers by cumulative spend — the actual revenue contributors, not the average buyer

3. Recency + frequency combination: purchased at least twice, with the most recent purchase within the last 6 months — this identifies active, loyal customers, not customers who bought heavily three years ago and have since lapsed

4. Referral activity: customers who referred at least one other person — brand advocates are a highly predictive seed because they represent both satisfaction and social similarity to other potential advocates

5. Engagement-based score: customers who opened 5+ CRM messages, attended an event, or replied to a campaign — high engagement predicts future LTV even before it shows up in purchase history

Engagement Data as a Lookalike Seed

One underused seed type is CRM engagement behaviour. Customers who consistently open your WhatsApp messages, click your email campaigns, or respond to your automations are signalling high brand affinity even before their purchase value fully materialises.

A segment of contacts who have opened 5+ messages in the last 60 days represents current, active engagement — a stronger LTV predictor than historical purchase data alone. This is particularly useful for businesses with longer purchase cycles (real estate, luxury, insurance) where high-value buyers may not have yet made a second purchase but are clearly highly engaged.

Testing Lookalike Performance

Once you have a high-LTV seed uploaded and a Lookalike created, run it against your existing default Lookalike (or broad interest targeting) as a split test with identical creative. Most businesses see the high-LTV Lookalike outperform within 2 weeks of launch.

Key metrics to watch:

• Cost per lead (not just CPM — lead quality matters more than click volume)
• Lead-to-customer conversion rate (tracked via CRM, not just Meta’s reported conversions)
• Revenue per lead (total revenue generated by leads from each audience over 90 days)

The best Lookalike audience you can build is a moving target — as your CRM grows and the high-LTV segment refines, refresh the seed every 30–60 days to keep the Lookalike current.

For how to structure your full CRM-to-Meta audience architecture across all segments, see How to Build Meta Custom Audiences That Actually Improve Over Time (Using Your Own CRM Data).