At Arfadia, we've implemented this strategy across hundreds of campaigns, consistently achieving 30-50% reductions in customer acquisition costs while maintaining high conversion quality.
The algorithmic process begins when you upload customer data to an advertising platform. That could be things like emails addresses, phone numbers, or data on website visitors gained through tracking pixels. That's the point at which the platform's machine learning system takes over and analyzes these users across hundreds of dimensions, what type of content they engage with, when they are most active online, what they buy, what devices they use, demographic information.
A lookalike audience represents one of the most advanced applications of artificial intelligence in digital advertising. The data comes from thousands of points of informations from your existing customers (referred to as the "seed audience") to find behavioral patterns, demographics similarities, and interest relationships that human marketers can't possibly find.
The real power of this technology is its ability to make non-intuitive connections. We've seen campaigns in which algorithms found that people who buy organic dog food are also early adopters of smart home technology, a correlation that no human marketer would have ever guessed. This ability to process patterns is why Meta's lookalike audiences were 8.2% CTR (versus the platform average 0.90%).
Different platforms use different algorithms. Facebook applies universal user embeddings deep learning models, and Google uses collaborative filtering like recommendation systems. LinkedIn tapping professional attributes, which differs for B2B targeting. All three platforms perform these in slightly different ways but they essentially follow three steps: query of data from your seed audience, compute the attributes influencing its behavior, and find the matching segment.
The complexity of these systems is evolving rapidly. Today's algorithms scrutinize more than 98,000 different attributes to make a decision on a user profile, according to industry research studies. Because of this granularity of analysis, targeting precision is now possible, a few years ago was not.
i"Lookalike audiences represent the most significant advancement in digital targeting technology over the past decade. When implemented correctly with high-quality seed data, they consistently outperform traditional demographic targeting by 3-5x while reducing acquisition costs by up to 50%. The key is understanding that quality always trumps quantity in seed audience selection."
— Tessar Napitupulu, CEO of Arfadia and Digital Marketing Expert with 20+ years experience
To date, Meta's incarnation is the most advanced and popular lookalike audience system. The product offers percentage targeting, anywhere between 1% and 10% of a given countries user base, 1% being the most similar to the seed audience.
A lookalike of the 1% in the US then consists of about 2 million individuals with the highest similarity scores, and 10% covers a more extensive 20 million. Facebook's 1% audiences always test the highest for us, with that performance decreasing as you move to larger percentages.
The platform needs at least a 100 people in your source audience (we recommend 1,000-5,000 for best practices). Optimal setting is a balance between having too much data so that there are too many effects and the classifier starts to learn the wrong patterns, and the classifier has too few data to learn statistically significant patterns.
When Google retired Similar Audiences in August and replaced it with Optimized Targeting, it reshaped the landscape significantly. Built on the AI model, the approach does start with your audience signals, but it goes beyond mere similarity matching to target customers that are most likely to convert based on campaign KPI's.
For marketers looking for the traditional lookalike, Google has Lookalike Segments available only in Demand Gen campaigns. These segments arrive as you'd expect, being 2.5%, 5%, and 10% narrow, balanced, and broad bolts (like Meta's own percentage bolts).
The change is also part of Google's larger movement towards automated audiences, instead of doing the heavy lifting of finding users that most resemble an advertiser's existing users in any given segment, the search giant is using machine learning to update the terms of the targeting on the fly, and adjust for performance quickly and regularly.
LinkedIn's change was less gradual: It entirely removed the traditional lookalike audiences in February. The company instead brought out its Predictive Audiences as a replacement, a step away from looking back at at profile histories and towards forward-moving AI.
And, Predictive Audiences, look at the work-related traits of your existing customers, from their job title, to company size, to industry to seniority level, to determine the LinkedIn members most likely to take similar action. This strategy works especially well for B2B campaigns, as professional context is often more important than typical behavior.
You're required to have at least 300 matched leaders in your seed audience, a far cry from the old 1,000-member minimum for data-based lookalikes.
Its doppelganger is especially granular in terms of control. Outside of typical narrow, balanced and broad settings however, TikTok enables advertisers to include or exclude the source audience and even create platform specific audiences for iOS versus Android.
It's the type of control that can prove vital when you are trying to reach a younger audience that has very specific devices and platform allegiances. TikTok suggests using 10,000 users or more in your seed audience for the best results.
Amazon DSP offers yet another appealing alternative, especially for e-commerce brands. Amazon's lookalike audiences use shopping activity data and purchase history to find users with purchase behavior similar to those within the Amazon environment.
Seltzer Goods, a seller of home goods, saw its wholesale revenue collapse during COVID-19. We pivoted their strategy to D2C sales with targeted lookalike audiences.
The strategy was to develop 1% lookalike audiences based on their best customers, not just all purchasers. We waited until our Facebook pixel had about 1,000 purchase events to gather good data, then launched campaigns using the post IDs of existing successful posts to maintain social proof.
The outcome went above and beyond all expectations: monthly revenue increased by 785%, return on ad spend soared to 9.68x, and the ripple effect carried organic traffic by 183%. The victory came from a patient scaling, when the ROAS goes over 5x, increased budgets only by 10-15% every few days (which avoided destabilizing of the algorithm, yet maintaining traffic).
A software company that has a fleet management solution was exhausted with rapidly rising costs with their Google Ads as the competition was fierce. Traditional keyword targeting was too costly with cost-per-lead > $200.
We did a lookalike on Facebook, but not on our whole website, on our CRM on existing customers and we targeted lookalikes vs our CRM. Seed audience 2,500 existing customers broken down by ACV and engagement level.
The campaign produced 300 qualified leads in five months, at a cost 60% less expensive than Google Ads. But most importantly, lead-to-opportunity conversion rate soared from 15% to 47% as lookalike targeting found people who genuinely needed the product instead of some browse-by-ers.
A high-end fashion retailer brought us in after experiencing some issues with Google Shopping performance as well as declining organic reach. Understanding their friction, we were able to rebuild their paid strategy with Performance Max campaigns, using strong testing-based lookalike segments at the time the feature was removed.
The strategy was rooted in visually telling a story and capturing a lifestyle that their target audience identified with. We established multiple lookalike audiences for different product categories, handbags, shoes, accessories, each optimized for different customer personas.
Results: $170,000 revenue from $35,000 in ad spend was a 4.8x ROAS and they broke their 2.5x ROAS goal in half. The victory was proof that product-based lookalike segmentation could drastically increase relevance and conversion rates.
Even small businesses can make effective use of lookalike audiences. A boutique fitness studio with 500 active clients used their most engaged (3+ per week class attendees) customers and created Facebook lookalike audiences from them.
Despite the qualified seed audience, the 1% lookalike brought in 127 new membership inquiries during three months, 34 per cent of which were converted to paid enquiries. The trick was an obsession with engagement quality over quantity in the seed audience selection.
We've looked at thousands of campaigns in dozens of different industries and lookalike audiences drive a stable and predictable 30-50% reduction in customer acquisition cost versus interest or demo targeting. This is thanks to enhanced relevance: You're reaching people who are highly likely to be interested in your products, rather than blanketing everyone with the same ad.
This cost-saving is more drastic in competitive sectors. Standard targeting methods have an average $100-200 cost per qualified lead, while it's common for B2B SaaS companies to be in the $20-50 range using lookalike audiences.
Statistical evidence supports this effectiveness. Industry benchmarks indicate that 1% lookalike audiences deliver 3-5x higher click-through rates (CTR) and 40-60% lower cost-per-action (CPA) than interest targeting.
Lookalike audiences not only bring in more conversions, but they bring in better conversions. Those acquired from lookalike targeting tend to have higher engagement rates, longer customer lifetimes, and higher lifetime value.
This quality enhancement stems from the fact that lookalike algorithms only target users that have real interest your merchandise, not just one-time buyers. When you create lookalikes from your top customers, the algorithm will hone in on its ability to not just deliver users that match your top users, but ones that become top users, too.
Real-world data confirms this pattern. Average order values were 23% higher for customers acquired through lookalike audiences versus other paid channels for one e-commerce client. And, six-month retention also increased from 32% to 48%.
As the traditional targeting mechanisms become saturated, you run out of interested users in your demographic or interest category targets, which have usually been used (in some manner) as proxies for interest and intent. With looking alike audience, it gives systematic way to scale with relevance!
This scalability is especially useful for companies in their growth stages that need more customer acquisition volume without losing efficiency. Then you can create several lookalike percentages (1%, 3%, 5%) to see which one best balances reach against relevance.
The expansion potential is substantial. In the United States, a 1% lookalike reaches about 2 million people, and 10% rolls up to 20 million. This spread will give you with the most precise potential of viewers based in your funds and growth targets.
Look-alike audiences support unified targeting approaches across different ad channels. You can create lookalikes on Facebook, Google, LinkedIn, and TikTok based on the same seed audience, ensuring message consistency and exploiting the relative strengths of each platform.
This uniformity makes for easy campaign management and attribution tracking. So that when these same audience segments see your message again on separate touchpoints, you're able to better calculate the combined effect of your advertising efforts.
Execution grows more efficient as you become proficient with one platform's lookalike system and can apply that learning to others. The strategy makes sense even when the tactics change from one platform to the next.
Unlike traditional, static demographic targeting, lookalike audiences get refined automatically over time as your customer base changes and expands. The algorithms are always updating on the latest customer data, learning more about the perfect elements of your audience.
This self-serving quality makes your targeting refined as you go without any further manual tuning required. The Lookalike algorithm learns these characteristics, and applies them to ensure a successful purchase funnel starting from the customer acquisition layer, which we all know to be Sales.
The learning curve increases as your business scales. The 100-user startup will have modest lookalike performance, but for the 10,000-user scale-up, it can deliver remarkably well-precise targeting based on rich behavioral signals.
The number one screwup we see: Using too little data to create lookalike audiences. The vast majority of advertisers dump their entire customer file on without regard to quality, recency or relevancy.
Best practice requires segmentation. High lifetime value customers generate fundamentally different lookalikes than purchasers who buy just one time. Recent customers produce different patterns than buyers from years past. Regional styles in rather than global combinations do best.
We found that it's best to segment seed audiences by value, by how users engage with your app, as well as by geography. A fashion retailer could make separate lookalikes for their handbag buyers, shoe buyers, and accessory buyers, for example, rather than lumping all customers into one audience.
Small seed audiences are fine, but you need to have enough data to target so that it is actually statistically significant. Facebook is most engaging for 1,000 to 5,000 people, while TikTok suggests 10,000 or above.
Methods become unreliable under these cutoffs. In this case, it suffers from insufficient data points to establish any meaningful pattern and the prediction is highly erratic and not scalable. On the other hand, a very large seed audience will outgrow the diversity that is useful for good pattern matching.
We allow a seed audience to hit recommended minimums before starting a lookalike campaign. For new companies, this could involve initially running engagement campaigns or lead generation campaigns to amass enough customer data.
Generating global lookalikes and not country-level makes an extreme underperformance, though. A 1% lookalike in Germany looks completely different than a 1% lookalike in Japan with the same seed audience.
The cultural tastes, how the platform used, and everyone's buying style will change from country to country. And what works best for U.S. consumers is not necessarily what will resonate with European or Asian audiences, even among what may look like similar demographic profiles.
We build lookalike audiences for every main geographical market and customize creative and messaging to local preferences while staying true to the brand.
Learning algorithms takes patience and doesn't happen in one fell swoop. A lot of advertisers see positive results and then immediately 2x or 3x their budgets, essentially "stopping learning" and causing decreased performance.
Scaling is done as 20% rule now that we keep the start budgets low and raise by 20% every 3-4 days with the condition that performance stays the same. This incremental scheme enables the maintaining of efficiency while adapting the algorithms.
When we want to make a jump bigger than this, we love increasing budgets through duplicate ad sets with new, high budgets instead of scaling our existing ones massively. This mode offers additional monitoring and with less impact to existing performance.
Luxury-purchaser lookalike audiences respond to different messaging than bargain-hunter lookalike audiences. Most brands serve the same creative to every audience, instead of optimizing message-market fit.
We create seed audience based creative briefs. If the doppelgänger is sourced from high net worth customers, we focus on exclusivity and a luxury product. For value-conscious segments, we showcase value propositions and promotions.
This congruence transcends the ad copy and entails visual identity, LP design, and value proposition. Everything should contribute to the feeling this particular segment of the audience wants.
Minimums differ by platform, but we suggest at least 1,000 people for Facebook, 1,000 for Google and 300 for LinkedIn. But those minimums are the absolute bar of operation for the algorithms and not an optimal result.
You should aim for 3,000-5,000 good users in your seed audience for optimal results. It is enough to create enough to recognise patterns yet retain audience coherence. Bigger seed audiences aren't always better, quality and relevance are more important than size.
Most platforms refresh ALO's for you, but even if they do that we still recommend manually updating every 30-60 days for best performance. This way the algorithm always take into account the new customer acquisitions and behavioral changes.
It will benefit fast growing businesses or seasonal brands which want to update daily. For stable businesses with little turnover in customer profiles, you can refresh this list every quarter. Balance freshness with learning phase stability is the key.
Yes, and that is the method we suggest for consistency. Take your seed audience and add them to Facebook, Google, LinkedIn and other platforms to develop a coordinated campaign that strengthens the message across various touchpoints.
But if possible make the creative and the message that much more targeted to each of the platform's special attributes as still consistent with your branded image. LinkedIn users are there for professional content, and TikTok for real, entertaining messaging.
If the performance is bad, it's often because of low-quality seed audiences, not enough data volume or a creative-audience mismatch. Check your seed audience segmentation, are you targeting your highest value customers or just any visitor to your website?
Think of geographic targeting range, budget scaling speed and campaign optimization settings. Every now and again, lookalike audiences have longer learning phases than other targeting methods; this is especially true for newer ad accounts.
With iOS and following privacy updates, limit is 30-60% less trackable user data for many advertisers, causing lookalike audiences to degrade. But a number of methods can help offset these effects.
Utilize the Conversions API for server side tracking, concentrate on email-based seed audiences over pixel-based data, and prioritise first-party data collection from owned channels. Platforms are also building privacy-enhancing alternatives that preserve targeting efficacy.
In general yes, unless you are using a specific campaign. When you exclude your seed audience, you avoid crossover with your retargeting campaigns and make sure you are targeting real new prospects rather than your current customers.
Yet for brand awareness efforts or when introducing new products to existing customers, inclusion can be a good idea. Try both techniques and see what fits your personal scenario.
The percent tells you how closely related users are to your seed audience and how big the resulting audience will be. The 1% audiences are of users that are most like your seed audience, but you're reaching less of them. Similar but much larger audiences are reached at 10% similarity.
We then generally start testing 3-5% for scale if performance still holds up there. 10% audience is good for raising awareness, but seldom will yield the best conversion effectiveness.
The development of lookalike audience targeting requires advanced tactics to implement it. Success is in getting more advanced rather than simple audience building to full optimization across data quality, platform selection through to creative alignment.
Start with data foundation excellence. Before you implement any lookalike campaign, audit your customer data for completeness, accuracy, and recency. Get rid of inactive users, repeat entries and low-value customers who may muddy the waters of pattern recognition. The performance of the algorithm was proportional to the quality of the seed audience.
Implement systematic testing protocols. We suggest controlled tests in which you isolate some variables: week one compares one size audience to another (1% versus 5% for example), week two segments seed audiences, week three optimizes creative. Test for at least 14 days and 50 conversions per variant for statistical significance.
Match creative strategy with audience features. Premium customers are responding to different messaging than discount shoppers, even if they are lookalike audiences. Create a brief for possible messengers which target the psychographics and motivations of each seed audience segment directly.
Gradually scaling the budget to avoid abrupt changes to the algorithm. Use the 20% rule: If performance is stable you can increase daily budgets by not more than 20% every 3-4 days. Significant budget boosts lead to the interrupting learning phase and tend to reduce effectively.
Lookalike audience features are evolving due to privacy regulations and platform developments. Apple's redesign of the iOS operating system decreased addressable users by 30-60%, and Google's deprecation of cookies will add more headwinds. But creative adjustments can keep it effective.
First-party data is the competitive edge. Brands who further invest in CDPs, email collection and deep relationships focused on the direct customer will have better lookalike this party signals fade. Some browser restrictions are bypassed via server-side tracking with Conversions API and Enhanced Conversions.
Artificial intelligence automation increases sophistication. Platforms are heading towards predictive systems, which predict future behaviours instead of targeting existing patterns. LinkedIn's Predictive Audiences are a prime example of this transition, with an emphasis on potential future activity as opposed to historical look-alikes.
Cross-platform coordination becomes essential. Because capabilities vary by platform, strong advertisers cross their lookalike strategies with different platforms. This methodology of operation allows for being agnostic to platform changes while optimising reach and frequency.
At Arfadia, we have not stopped developing our lookalike audience tactics as the terrain changes. The underlying truth will never change: discovering new customers who resemble your top ones has always been good business. The strategies for distribution are becoming more complex, but the fundamental value remains.
If you're just getting started with lookalikes, optimize, or have optimized in the past, concentrate hard on data quality, systematic testing, and patient optimization. These elements keep the campaigns that are working from failing, on every platform and in every vertical we've tested.
We use cookies to ensure the website runs optimally and to help us understand how you use our services. You can choose which categories to allow. Read our Privacy Policy.
Required for basic website functionality. Cannot be disabled.
Help us understand how visitors interact with the website. Data used anonymously.
Used to display relevant ads and measure campaign effectiveness.
Enables live chat, social media integrations, and language preferences.
