Now open to non-executive and advisory board roles. to board roles.Boards & advisory
  1. Home
  2. Insights
  3. Meta ads
Meta ads

Why Lookalike Audiences Break in a Small Market

In a small market the one percent lookalike and the broad audience converge, so the percentage you pick barely matters. What decides the result is the quality of the seed and the exclusion list. Stacking more lookalikes mostly re-buys the same people at rising frequency, which reads as fatigue.

By Stephen EllulPublished 26 Sept 20267 min read
Why Lookalike Audiences Break in a Small Market cover

Almost everything written about lookalike audiences assumes you are advertising into a country of sixty million people. At that scale the percentage is a real dial: one percent is a genuinely tighter slice than five, and expanding gives you somewhere new to go. I run Growth Bully, a Malta performance marketing agency, and in a market of a few hundred thousand adults that logic quietly stops holding.

The reason is arithmetic rather than opinion. A one percent lookalike is one percent of the population you target it at. Narrow the country and you narrow the audience with it, until the tight lookalike and the broad audience are describing almost the same group of people. Once that happens, the dial you have been turning was never connected to anything.

Why do lookalike audiences stop working in a small country?

Because the percentage is a share of the targetable population, not a fixed number of people. In a small market one percent can land in the tens of thousands, which is already close to the whole reachable audience for a specific product. The tight audience and the broad one converge, so the setting stops changing who sees the ad.

That convergence explains a result I see constantly. Someone builds a careful audience structure, three lookalike tiers plus interest sets, and every one of them performs about the same. The natural conclusion is that targeting does not matter. The real conclusion is narrower: at this population size, that particular lever does not matter, and the levers that do were never being tested.

What matters more than the lookalike percentage?

The seed. A lookalike is only a model of the list you fed it, so a seed of five hundred people who actually bought produces something useful, while a seed of everyone who ever visited the site produces a slightly warmer version of broad. Seed quality decides everything the audience does afterwards.

I rank seeds by how close they sit to money. Purchasers or closed clients first, then qualified leads, then people who spent real time on a pricing or booking page, then all form fills, then site visitors, then video viewers. The last two are barely worth modelling in a small market, because the audience they generate is indistinguishable from what the platform would have found on its own.

The seeds worth building from, in order:

  • Customers or closed clients, ideally with purchase value attached so the model learns who is worth more.
  • Qualified leads, meaning the ones your sales process accepted rather than every enquiry that arrived.
  • Deep intent visitors: pricing, booking, checkout, anyone who took a step that costs effort.
  • All leads combined, which is usually the most you can reach while keeping a useful signal.

Does a bigger seed always make a better lookalike?

No. Size and purity trade against each other, and in a small market purity wins. A seed of a few hundred real buyers beats one of ten thousand visitors, because the model has something specific to look for. Below roughly a hundred people it is simply guessing.

The failure I see most often is a seed built for volume. Someone worries that a few hundred customers is not enough, so they add every lead, every visitor and every engagement into one list. The audience gets larger and the signal gets diluted, and the result is an expensive way to reach a slightly filtered version of the whole country.

Why does stacking more audiences make performance worse?

Because in a small market they overlap heavily, so you are not adding reach, you are adding routes to the same people. The platform then competes with itself, frequency climbs faster than you expect, and the symptoms show up as rising cost per result and falling click through, which everyone reads as creative fatigue.

It is worth being precise about that misdiagnosis, because the fix people choose is expensive. They rebuild the creative, which was fine, and leave the structure that was actually the problem. If the same person is being served three audiences worth of impressions, the ad is not tired. It is simply being shown too often to too few.

What should the exclusion list do?

More work than the targeting, in most small market accounts. Excluding existing customers, open opportunities and recent purchasers stops you paying prospecting prices for people who are already in your pipeline, and it keeps the prospecting number honest rather than quietly inflated by warm traffic.

My default exclusions on any prospecting campaign here:

  1. Existing customers, unless the campaign is deliberately a repeat purchase or upsell push.
  2. Anyone who converted in the last thirty to ninety days, matched to the length of your sales cycle.
  3. Open opportunities already in the CRM, because sales is handling them and ads should not be bidding for them again.
  4. Your own retargeting pools, so prospecting and remarketing are measured separately instead of borrowing each other’s results.
  5. Employees and suppliers, which sounds trivial until you notice how much of your engagement data they generate.

Should I just use broad targeting instead?

Usually yes, and it is a relief rather than a compromise. If the tight audience and the whole country are nearly the same set, pick the larger one and let the delivery system do the work it is genuinely good at. Your creative and your offer then become the targeting, which is where the leverage actually sits.

Broad with strong exclusions and two or three genuinely different creative arguments will outperform a carefully layered audience structure in a small market more often than not. It also gives cleaner reading, because performance differences come from the message rather than from overlapping audiences competing for the same impression.

When is a lookalike still the right call?

When the product has a narrow buyer inside a broad population, or when you are expanding beyond your own market. A niche B2B service, a high ticket purchase or a specific life stage all give the model something real to find. Crossing into a larger country restores the percentage dial to being a dial again.

There is one more case: when your seed is genuinely excellent and genuinely small. A few hundred high value customers with purchase values attached is a strong signal, and a one percent lookalike from that list is worth testing against broad rather than assumed to be better. Test it, keep the winner, and stop building the other six audiences.

That is the whole shift I would ask for. Spend the hour you were going to spend on audience architecture on the seed and the exclusion list, then put the rest into making another argument. If you would rather have that run properly, it is what my team does at Growth Bully.

Frequently Asked Questions

What size should a lookalike audience be in a small country?

Pick based on reach rather than on a percentage rule imported from larger markets. In a small market one percent and ten percent often describe nearly the same group, so start at one percent, check the estimated reach the platform reports, and if it is already close to your whole addressable audience, use broad targeting instead and put the effort into the seed.

How many people do I need for a good lookalike seed?

Roughly a hundred at the absolute minimum for the model to have anything to learn from, and a few hundred to a few thousand is the useful band. Beyond that, more names rarely improve the output. What improves it is purity: buyers beat leads, and leads beat visitors, every time.

Why do all my audiences perform the same?

Because they overlap. In a small market lookalikes, interest sets and broad targeting frequently describe much the same people, so the platform finds the same responsive individuals whichever door you open. Identical performance across audiences is a signal to simplify the structure and test the creative instead.

Is it audience overlap or creative fatigue?

Check frequency before you rebuild anything. If frequency is climbing while reach stays flat, you are showing the same ads to the same people through multiple audiences, which is a structure problem. If reach is still growing and performance is falling, the creative is genuinely worn out and needs a new argument, not a new edit.

Should I exclude existing customers from prospecting campaigns?

Yes, on almost every prospecting campaign. Otherwise you pay to acquire people you already have, and your cost per new lead looks better than it is because warm traffic is quietly propping it up. Keep customers, recent converters and open opportunities out, and handle them with separate campaigns built for that job.

Read next

Meta ads

Facebook Ads for Restaurants: A Complete Guide

How to run Meta Ads for your restaurant: campaign objectives, local targeting, creative that converts, and how to measure results properly.

27 Oct 2026 · 5 min read
Meta ads

What Is a Good Cost Per Lead on Facebook Ads?

Industry CPL benchmarks for Facebook Ads in 2026, plus the exact formula to calculate your own target cost per lead from your business unit economics.

27 Oct 2026 · 6 min read
Meta ads

What Is a Good ROAS for Meta Ads? 2026 Benchmarks by Industry

A good ROAS for Meta Ads is typically 3x–5x for e-commerce, but varies significantly by industry and margin. Here's the full breakdown.

21 Aug 2026 · 5 min read

Book 30 minutes.

I'll look at where your leads come from today and tell you straight whether I can help.

Rather build it yourself? The 30-Day Pipeline™ walks you through it in your own accounts.

The Meta ads checklist

The checks I run on every new account before spending a euro.

Check your inbox. The checklist is on its way.