Lookalike Audiences: Find Prospects Like Your Clients
You already know who your best clients are. They sit in a CRM, they pay on time, they renew, and they look surprisingly similar to each other. A lookalike audience is the ad platform's attempt to go find more people like them. Feed it a list of your closed deals, and it scans its user base for people who share the same buying signals, then shows your ads to them.
Done well, this is one of the cheaper ways to expand reach without dropping into cold, untargeted prospecting. Done badly, it quietly spends your budget on people who resemble your customers in age and job title but will never buy. The gap between those two outcomes is almost entirely about what you feed the algorithm and how you size the result.
This guide covers where lookalikes work in B2B, how to build a seed list that produces good matches, how to set them up on Meta and the realities on LinkedIn, and the specific mistakes that turn a smart tactic into wasted spend.
What a lookalike audience actually does
The platform takes a source audience you provide (the "seed"), models the shared traits of those people, then ranks its own users by how closely they match. You pick how broad to go, and it builds an audience of the closest matches at that size.
The traits it models are mostly invisible to you. It is not just "marketing managers at SaaS companies." It blends behavioral signals, interests, page-visit patterns, and engagement history into a similarity score. That is the strength and the weakness. The model finds patterns you would never spot manually, but it also has no idea which of your seed customers were actually profitable. It treats a one-time buyer who churned the same as your best three-year account, unless you tell it otherwise by what you put in the seed.
This matters more in B2B than in e-commerce. A consumer lookalike off 50,000 purchasers is dense and reliable. A B2B lookalike off 200 closed deals is thin, and a chunk of your seed may be personal email addresses the platform cannot match to a profile. The mechanics are the same; the data hygiene bar is much higher.
Where lookalikes fit in a B2B funnel
Lookalikes are a top-of-funnel, prospecting tool. You are reaching people who have never heard of you, chosen because they pattern-match to your customers. That means two things for how you use them.
First, judge them by prospecting metrics, not closing metrics. A lookalike campaign that produces cheap, qualified leads at the top is winning even if the immediate cost per closed deal looks high, because those leads still have to move through your funnel. Compare lookalike cost per lead against your other cold channels, not against retargeting.
Second, they pair best with a warm-audience layer underneath. Lookalikes bring new people in; retargeting and nurture sequences convert them later. If you run lookalikes with no follow-up system, you are paying to introduce people to your brand and then letting them forget you.
The channels that matter here are Meta (Facebook and Instagram), where classic lookalikes are strongest, and LinkedIn, where the equivalent feature works differently. We will take them in turn.
Building a seed list that produces good matches
Everything depends on the seed. A weak source audience produces a weak lookalike no matter how clever the algorithm is. Here is how to build one worth modeling.
Use outcomes, not just leads. The instinct is to upload everyone who ever filled in a form. Resist it. Your strongest seed is closed-won customers, ideally filtered to the profitable ones. If you have enough volume, seed off your high-LTV segment specifically. The platform will then look for people who resemble buyers, not people who resemble form-fillers. This single choice separates lookalikes that bring qualified leads from lookalikes that bring tire-kickers.
Hit the minimum size, then some. Most platforms need at least 100 matched people in a source audience, and they will tell you the bare minimum is enough. It rarely is. A few hundred is shaky; 1,000 to 5,000 matched records gives the model enough signal to find a real pattern. If your closed-won list is too thin, widen the definition carefully: add high-intent leads (demo requests, pricing-page visitors) rather than dropping your quality bar to anyone who downloaded a PDF.
Clean the data before you upload. Match rates live and die on data quality. Use the email address the person gave at work, include phone numbers and names where the platform accepts them (more fields means a higher match rate), and strip out test records, competitors, and obvious junk. Your first-party data is the raw material here, and a messy CRM export produces a messy seed.
Segment by what you sell. If you have distinct products or customer types, build a separate seed and lookalike for each. A blended seed of two very different customer profiles tells the model to find the average of two things, which usually resembles neither. One clean segment beats one large mixed one.
| Seed source | Signal quality | When to use it |
|---|---|---|
| All form fills, ever | Low | Only if you have no better volume |
| All leads (MQLs) | Medium | When closed-won volume is too thin |
| Closed-won customers | High | The default starting point |
| High-LTV customers | Highest | When you have enough buyers to segment |
Setting up a lookalike on Meta
Meta is where lookalikes are most mature and most useful for B2B, even though it is not a "B2B platform" the way LinkedIn is. The targeting on Meta is consumer-grade, so a good lookalike off a strong seed often outperforms manual interest targeting by a wide margin.
The setup, in order:
- Create the source as a Custom Audience. Upload your cleaned customer list, or pull from a Custom Audience already built off website visitors, lead form opens, or video viewers. A customer-list source is usually the strongest for prospecting.
- Build the Lookalike Audience from that source. Pick the country or region you sell into. The lookalike is country-specific, so build separate ones per market.
- Set the size. Meta lets you choose from 1% to 10% of the population in your chosen country. 1% is the tightest and most similar to your seed; 10% is the broadest and loosest.
- Layer light filters on top. You can narrow a lookalike with a job-related interest or seniority filter. Keep it light. Over-filtering a 1% lookalike can shrink it to nothing useful.
- Exclude existing customers and recent leads. Add your customer list as an exclusion so you are not paying to re-introduce yourself to people already in your pipeline.
Start at 1% to 2% and only widen once that audience is performing and starting to fatigue. A common pattern is to run 1% and 3% to 5% as separate ad sets and let the data show which size your funnel converts.
The LinkedIn reality
People expect LinkedIn to have the best lookalikes because it has the best B2B data. The feature exists, but it behaves differently and the economics are different too.
LinkedIn's version builds a similar audience off a matched audience source (a contact list or website-visitor segment). It needs a larger minimum source to activate, and the resulting audience leans broad. In practice many B2B advertisers get more control on LinkedIn from precise audience targeting by job title, function, seniority, and company attributes than from its lookalike feature. You can describe your ideal buyer directly, so the modeling matters less.
LinkedIn also has "audience expansion," a checkbox that loosens your targeting to include similar members. Treat it with suspicion for B2B. It often pulls in adjacent-but-wrong people and inflates spend. For most precise campaigns, leave it off.
The practical split: use Meta for lookalike-driven prospecting where the algorithm does the work, and use LinkedIn's native firmographic targeting where you can spell out the buyer yourself. If you are weighing the two channels broadly, the tradeoffs go beyond audience tools, and it is worth reading a full comparison of LinkedIn and Meta for B2B before splitting budget.
Sizing: tight or broad?
The size choice is the lever most people get wrong. Here is the honest version.
A 1% lookalike is the closest match to your seed. Smaller audience, higher similarity, usually higher relevance per impression, and it fatigues faster because you exhaust it sooner. Good for a strong, narrow seed and a clear offer.
A broad lookalike (5% to 10%) trades similarity for reach. More people, looser match, more room for the algorithm to optimize against your actual conversion events if you have them firing reliably. This can work well when your conversion tracking is solid, because the platform can re-tighten the loose audience using real outcome data. It works badly when your tracking is thin, because then nothing corrects the looseness.
There is no universally right answer. The deciding factor is how good your conversion signal is. Strong tracking and a high-volume pixel let you go broad and let the machine optimize. Weak tracking means stay tight and lean on the seed's quality. Test two sizes side by side and judge by cost per qualified lead, not cost per click.
Common mistakes that waste the budget
Seeding off the wrong list. The single biggest error. A lookalike off all leads, or off newsletter subscribers, models the wrong people. Garbage seed, garbage audience. Always start from buyers.
Treating "matched job titles" as success. A lookalike that contains the right job titles is not automatically a good audience. The model optimizes for similarity to your seed, not for your conscious idea of the buyer. Judge it on lead quality downstream, not on whether the demographics look right.
Never refreshing the seed. Customer lists go stale. A seed built 18 months ago models who used to buy. Refresh source audiences quarterly so the lookalike tracks your current best customers, especially if your ICP has shifted.
Forgetting exclusions. Without excluding current customers and open opportunities, you pay to advertise to people already talking to your sales team. It also pollutes your performance data.
Running lookalikes with no warm layer. Cold prospecting brings people in once. If there is no retargeting or nurture catching them afterward, most of that introduction is wasted. The creative also has to do real work here, and weak paid social ads will sink even a perfect audience.
Going too broad with weak tracking. A 10% lookalike with no reliable conversion signal is close to spray-and-pray. Earn the right to go broad by getting your tracking right first.
FAQ
How many people do I need in my seed list? The technical minimum is usually 100 matched records, but that produces a weak audience. Aim for at least 1,000, and ideally several thousand, for a stable model.
Are lookalike and Custom (saved) audiences the same thing? No. A Custom Audience is people you already have a relationship with (your list, your site visitors). A lookalike is new people the platform finds who resemble that Custom Audience. You build the lookalike from the custom one.
Do lookalikes work for B2B, or only B2C? They work for B2B, mainly on Meta, when your seed is built from real buyers and you judge results by lead quality. The thinner your customer data, the harder they are to run well, which is the usual B2B constraint.
Should I turn on LinkedIn's audience expansion? For most precise B2B campaigns, no. It loosens your targeting and tends to pull in adjacent people who do not fit, which inflates spend. Keep your targeting deliberate instead.
How often should I rebuild the audience? Refresh the source seed roughly quarterly so it reflects your current best customers. The lookalike updates from the seed, so keeping the seed current keeps the audience current.
Will a bigger lookalike always mean cheaper leads? Not reliably. Broader audiences cost less per click but often less qualified, so cost per click drops while cost per qualified lead rises. Measure the metric that maps to revenue, not the cheapest surface number.
Before you launch: a short checklist
- Seed built from closed-won customers, not all leads, and ideally your high-LTV segment.
- At least a few hundred matched records, more if you can get them.
- Data cleaned: work emails, extra fields filled, junk and competitors removed.
- Separate seeds and lookalikes per product or customer type.
- Existing customers and open deals excluded from targeting.
- Two sizes tested side by side (tight vs broad), judged on cost per qualified lead.
- A retargeting and nurture layer ready to catch the new traffic.
Lookalikes reward good data and punish lazy data. If your CRM is clean and you know which customers are actually worth more, the algorithm can do real work for you. If it is messy, the model will faithfully find you more of the wrong people.
If you would rather not run this trial-and-error yourself, that is what we do. Send us your customer data situation and your current paid social results, and we will tell you whether a lookalike-driven prospecting layer is worth building for your funnel, and how to seed it. No long pitch, just a straight read on whether it fits.