How to Build Lookalike Audiences That Actually Convert in South Africa

Lookalike audiences are arguably the most powerful feature in Meta's advertising platform. They let you say to Meta: "Here are my best customers. Find me more people like them."

When done correctly, lookalikes consistently outperform interest-based targeting on cost per lead, cost per acquisition, and lead quality. When done incorrectly — which is how most SA businesses use them — they're just an expensive way to reach random people.

Here's how to build them properly.

What Is a Lookalike Audience?

A Lookalike audience is built from a "source audience" — a group of people you define. Meta's algorithm then analyses that source group, identifies patterns and characteristics, and finds other people on the platform who share those patterns.

The source can be:

• A customer email/phone list you upload
• Website visitors tracked by your Pixel
• People who completed a lead form
• People who engaged with your content
• People who made a purchase

The quality of your source determines the quality of your lookalike. This is the single most important concept in this entire article.

The Source Audience Hierarchy

Not all sources are created equal. Here they are ranked from most to least effective:

Tier 1: Actual Paying Customers

Upload a list of people who paid you money. Not leads. Not enquiries. People who converted. This is the gold standard because it tells Meta: "Find me more people who are likely to pay."

If you have 500+ paying customers with email addresses or SA mobile numbers, start here.

Tier 2: High-Quality Leads

People who enquired AND were qualified. Not all leads — just the good ones. If you track lead quality in a CRM, export your "hot" or "qualified" leads.

Tier 3: Website Converters

People who completed a specific action on your website — filled out a form, booked a call, reached a thank-you page. This requires a properly configured Pixel with conversion events set up.

Tier 4: All Website Visitors

Everyone who visited your site. This is a weaker signal because visitors include people who bounced after 2 seconds. But if you don't have the data for Tier 1–3, this is a reasonable starting point.

Tier 5: Engagement Audiences

People who liked, commented, shared, or watched your videos. The weakest source because engagement doesn't equal purchase intent. But it's data, and data is better than random targeting.

The rule: Always use the highest-tier source available to you. As your campaigns run and you collect more data, upgrade your source over time.

How We Reached All 9 Provinces in 60 Days

One of our clients was in agriculture. Their target audience barely existed online — farmers and agricultural businesses across South Africa. Traditional interest targeting was useless because Meta's "farming" interest includes everyone who's ever liked a farming meme.

Here's what we did:

Phase 1: Seed Campaign (Weeks 1–2)

We ran a lead generation campaign using broad geographic targeting (all of SA) with ad creative specifically designed to attract agricultural professionals. The copy referenced specific farming challenges — input costs, seasonal planning, equipment — not generic "farming" content.

The creative itself filtered the audience. In 2 weeks, we had 200+ leads from agricultural professionals.

Phase 2: Build the Lookalike (Week 3)

We uploaded those 200 qualified leads as a Custom Audience and created a 1% Lookalike targeting all of South Africa. Meta's algorithm took those 200 data points and found similar profiles across every province.

Phase 3: Scale (Weeks 4–8)

We ran the same campaign to the Lookalike audience. CPAs dropped 35% compared to the seed campaign. Lead quality stayed high. Within 60 days, we had leads from all 9 provinces — from an audience that "barely existed online."

The Lookalike didn't find farmers through interest targeting. It found them through behavioural patterns, device usage, content consumption, and dozens of other signals that Meta's algorithm detects but doesn't disclose.

Lookalike Audience Size: 1% vs 5% vs 10%

When creating a Lookalike, you choose a percentage that represents how closely the new audience should match your source:

• 1% Lookalike: The top 1% of people most similar to your source. Smallest audience, highest quality. In South Africa, this is roughly 200,000–300,000 people.
• 3% Lookalike: Wider net, still strong quality. Roughly 600,000–900,000 people.
• 5% Lookalike: Good for scaling. 1–1.5 million people.
• 10% Lookalike: Very broad. Quality drops significantly. 2–3 million people.

Our Recommendation

• Start with 1% — highest quality, best CPA
• Scale to 2–3% when your 1% starts to fatigue (frequency above 3)
• Use 5–10% only as broad prospecting audiences when budget allows

For most South African businesses with budgets under R15,000/month, a 1% Lookalike provides more than enough reach. Don't go broader just because you can.

Common Lookalike Mistakes

1. Using the Wrong Source

A Lookalike built from people who liked your Facebook page will find more people who like Facebook pages — not more buyers. Use conversion data, not vanity metrics.

2. Source Audience Too Small

Meta recommends a minimum of 100 people in your source audience, but we've found that 300+ produces noticeably better results. Below 100, the algorithm doesn't have enough data to identify meaningful patterns.

3. Not Refreshing the Source

Your customer base changes over time. A Lookalike built from customers you acquired 2 years ago may not reflect who's buying today. Refresh your source audience quarterly.

4. Stacking Lookalikes With Interest Targeting

Don't create a 1% Lookalike and then narrow it further with interest targeting. You're overriding Meta's algorithm with your assumptions. Let the Lookalike do its job.

5. Using Lookalikes in Special Ad Categories Without Adjusting

In restricted categories (insurance, finance, employment, housing), standard Lookalikes aren't available. You get "Special Ad Audiences" instead. These work on the same principle but with privacy restrictions. Build them the same way — best source data, start narrow, scale gradually.

Advanced: Value-Based Lookalikes

If you can assign a value to each customer — lifetime revenue, average order value, number of purchases — you can create a Value-Based Lookalike. This tells Meta not just "find similar people" but "find similar people who are likely to spend more."

How to Set It Up

1. Export your customer list with two columns: identifier (email or phone) and value (revenue amount)
2. Upload as a Custom Audience in Ads Manager, selecting the "value" column
3. Create a Lookalike from this audience

Value-Based Lookalikes are particularly powerful for businesses with varying customer values. If 20% of your customers account for 80% of your revenue, this approach focuses your targeting on finding more high-value customers.

The Compounding Effect

Here's why Lookalike audiences get better over time:

• Month 1: You have 100 customers in your source. The Lookalike is decent.
• Month 3: You have 300 customers. The Lookalike improves.
• Month 6: You have 800 customers. The Lookalike is significantly more accurate.
• Month 12: You have 2,000+ customers. The Lookalike is a machine.

Every customer you acquire makes your future acquisition cheaper. This is the compounding return that most businesses don't account for when calculating ROI — your current ad spend is an investment in future efficiency.

Getting Started

If you've never used Lookalike audiences:

1. Export your customer list (even if it's just 100 people)
2. Upload it as a Custom Audience in Meta Ads Manager
3. Create a 1% Lookalike audience for South Africa
4. Run a lead gen campaign targeting only that Lookalike
5. Compare performance against your interest-based campaigns

The difference is usually visible within the first two weeks.

---

We build and manage Lookalike audience strategies for South African businesses. From R4,999/month, no contracts. Get your game plan.