LinkedIn Lookalike Audiences for B2B Expansion Targeting

LinkedIn discontinued Lookalike Audiences on February 29, 2024. Plenty of guides still describe it like it's live, which sends people down a path that ends nowhere.

So what actually happened? LinkedIn didn't just pull the feature and leave a gap. It replaced Lookalike Audiences with two different tools: Predictive Audiences and Audience Expansion. Neither one is a repainted version of the old system. Both run on different logic entirely, and that difference is the whole reason this guide exists.

One detail trips people up early: a contact list no longer works the way it used to. A contact list today feeds the newer tools as a source, not as direct seed material for a standalone lookalike build. Small distinction, big consequence for how you plan a campaign.

Audience Expansion versus Predictive Audiences

Start with the simpler of the two. Audience Expansion is a toggle you apply to your existing Matched Audiences, widening reach beyond the exact people on your list to others who look similar on the platform. It sits as an extension of the broader Matched Audiences family, alongside website retargeting, contact list uploads, company list uploads, event audiences, and Lead Gen Form audiences. To serve ads at all, a Matched Audience needs 300 matched profiles as a floor, but real scale takes a lot more: advertisers generally need at least 50,000 members to get solid results, and some see their best performance above 120,000.

Predictive Audiences works on a different engine. It's an AI system that studies your CRM data, your conversion events, and other first-party signals, then goes and finds the professionals most likely to convert on its own. LinkedIn's AI pulls together millions of engagement signals and lines them up against your customer data to surface the people who match your actual business goals. The key difference from the old Lookalike feature sits right here: Predictive Audiences learns continuously from live conversions happening in real time, instead of working off a static list someone uploaded once and forgot about.

Source audience quality determines whether either tool works

Both tools take whatever you feed them and amplify it, so garbage in still means garbage out, just at a much bigger scale than before. Feeding Audience Expansion a weak seed audience broadens that weakness across thousands of new impressions.

So what counts as a strong seed? Customers who closed in the past 12 to 24 months, high-value accounts showing signs of expansion, SQLs that converted fast, and site visitors who asked for a demo or pricing information. A seed built from customers above a certain contract value, on the other hand, points the system toward people who actually resemble your best accounts.

Predictive Audiences raises the bar even higher, because the model never stops learning. If you feed it a steady, high volume of MQL or SQL events each month inside a single campaign objective, the model converges fast. Accounts below that volume spend most of their time in exploration mode rather than actually finding good prospects, because the model simply doesn't have enough signal to work with yet.

This is where the Conversions API, or CAPI, becomes the detail worth getting right. CAPI setup isn't a nice-to-have anymore for anyone who wants Predictive Audiences to actually perform.

The same hygiene rules apply to the Matched Audiences that feed Audience Expansion. Contact list uploads typically match between 30 and 70 percent, depending on how clean the list is, while company list uploads typically match between 60 and 90 percent. And none of this holds up on a stale list. You need to refresh seed lists at least once a quarter, or monthly if your company is adding customers quickly.

Building the Matched Audience library that feeds expansion targeting

A working expansion program needs a full library of Matched Audiences pulled from every first-party source available, not just one or two favorites. Start with the Insight Tag, installed across the entire site, not just the landing pages. Once that's in place, the highest-value site segments break down by how recently someone visited and what they looked at: pricing and demo page visitors get a 7 to 14 day retention window since that's the warmest signal on the site, product and solution page visitors hold for 30 days, case study and testimonial viewers for 45 days, and blog and resource readers for 60 to 90 days since that interest cools more slowly. Exclusion audiences deserve just as much attention as targeting audiences. Exclude anyone who already converted, using a 180-day window, so you don't spend acquisition budget chasing people who already did what you wanted.

Contact list audiences come next. Uploading deals lost in the past 12 months and serving them case studies or news about new capabilities tends to bring a measurable share of that lost pipeline back to life.

Company list audiences, built for ABM, need company names, domains, and location fields, since LinkedIn matches across multiple fields to land on the right accounts. Integrating LinkedIn with CRM and marketing automation platforms like HubSpot, Marketo, and Salesforce keeps these audiences syncing automatically, so segments stay current without someone manually re-uploading a spreadsheet every few weeks.

Round out the library with event and Lead Gen Form audiences. Event audiences capture anyone who expressed interest in or attended a LinkedIn Event, run on 365-day retention, need a minimum of 300 members, and work well for turning event attendees into pipeline. Lead Gen Form audiences target people who opened or submitted a form, also on 365-day retention with the same 300-member minimum, and they're especially useful for re-engaging people who almost converted but didn't quite finish. This library is the foundation everything else in this guide depends on. Without it built out, Audience Expansion and Predictive Audiences have nothing solid to work from.

Using Audience Expansion versus Predictive Audiences, and combining them with ABM targeting

Audience Expansion and Predictive Audiences don't fight for the same job. They solve different problems at different points in the funnel, and the strongest programs use both on purpose rather than leaning on whichever one got set up first.

Audience Expansion earns its keep at the top of the funnel, when the goal is reaching professionally similar people who haven't encountered the brand yet. Lookalike-style audiences lean toward scale over precision, so they work best paired with a clear goal like lead generation or brand awareness, rather than asking them to do the work of closing a deal.

Predictive Audiences fits once an account has solid conversion volume and CAPI infrastructure in place, because the model needs live signal to beat manual targeting. And it keeps getting better over time for programs with growing CRM data, since the model compounds on accumulated history instead of starting over with a static list each time.

The real leverage appears when Predictive Audiences is combined with ABM-style Account Targeting, where a company uploads a target list of up to 300,000 companies through Matched Audiences. Once a target list crosses 5,000 companies, Predictive Audiences paired with Account Targeting comes out ahead.

Two newer company-level facets sharpen ABM precision without adding more manual upkeep. Company Growth Rate adds another filter on top of an uploaded account list, so ABM campaigns get more precision without the maintenance burden a static list usually carries.

Structuring campaigns and budget across expansion and retargeting audiences

Expansion targeting and Matched Audience retargeting are not the same line item in a budget, and treating them as one leads to trouble in either direction: too much spend chasing net-new reach when warm audiences could move faster, or the opposite, where retargeting eats the whole budget and nobody's bringing in fresh demand.

A three-layer framework helps sort this out. High-intent demand capture covers Google Search plus LinkedIn retargeting aimed at high-intent audiences, like Matched Audiences built from pricing-page visitors, demo requesters, and known contacts late in the funnel, and takes roughly 40 to 50 percent of budget. Demand education takes another 30 to 40 percent, aimed at mid-funnel content that pushes ICP-fit prospects toward a decision, using website retargeting of product and solution page visitors along with contact list audiences earlier in the pipeline. Demand creation takes the remaining 10 to 20 percent, built around thought leadership aimed at senior decision-makers, using Audience Expansion and Predictive Audiences to reach net-new professionals who match the ideal customer profile.

The cost math on LinkedIn forces a certain discipline here. Cost per lead by itself is a misleading number to chase. The number that actually matters is cost per qualified opportunity, not cost per lead.

Restructuring the budget around the three-layer framework turned that around, with the shift showing up in results by the third quarter.

Campaign structure needs to follow the same logic. None of this holds up, though, if the measurement behind it is broken, which is exactly where attribution comes in.

LinkedIn's default attribution window and the pipeline it misses

Campaign Manager's default attribution settings were built with short consumer buying decisions in mind, not long B2B sales cycles, so if teams use those defaults without question, they undercount how much pipeline LinkedIn is actually driving.

For B2B deals with sales cycles stretching 90 days or longer, that window misses a large chunk of LinkedIn-influenced pipeline, since deals closing weeks or months after someone first saw an ad never get credited back to the campaign that started things.

Fixing this means connecting LinkedIn's conversion data to CRM opportunity records directly. Ad performance and revenue outcomes need to be visible at the account level, not just tracked lead by lead. Where native integration between LinkedIn and a CRM isn't enough, third-party attribution platforms can bridge the two systems.

The CAPI setup covered earlier does double duty here.

Reporting should split into two tracks built for two different audiences. Campaign operators need tactical numbers: audience-level performance, creative performance, cost per lead broken out by segment, and reach and frequency for expansion audiences. Growth leaders need the strategic view instead: pipeline influenced by LinkedIn, cost per qualified opportunity, revenue tied back to paid media, and an account-level look at which target accounts are engaging and how.

Creative and landing page failures that cancel out correct targeting

Getting the targeting right only gets the ad in front of the right person. What happens after that click depends entirely on the creative and the landing page, and both of those move faster, and break down faster, than most B2B teams give them credit for.

Creative fatigue sets in after roughly four to six weeks on LinkedIn. Refreshing creative monthly with new angles and new formats is what keeps a drop in performance from getting misread as the audience running dry when it's really just the ad going stale. Different Matched Audience segments need different creative, too. Serving the same ad to a warm pricing-page retargeting audience and a cold Audience Expansion audience wastes the premium that comes with reaching someone already familiar with the brand.

The landing page has to match the audience just as closely. Sending an expansion audience built from high-value customers to a generic homepage erases all the specificity that went into building that audience the moment someone clicks through. ABM campaigns run into the same problem when a targeted account-list audience lands on a page that never mentions the account's industry, its specific challenge, or where it sits in the buying process. That mismatch wastes the specificity that justified paying the LinkedIn premium for ABM targeting.

Reading the full funnel rather than one platform metric in isolation is how to figure out whether a weak campaign is a targeting problem, a creative problem, or a landing page problem. And a strong conversion rate on the platform paired with weak pipeline in the CRM points to an attribution or lead quality issue, since the model may be optimizing for a conversion event that doesn't actually predict real pipeline.

Managing ongoing expansion targeting

Running Predictive Audiences and Audience Expansion well isn't a task that ends at launch; it's a continuous job that compounds in both directions. Managed well, each campaign feeds the next one better data and better results. Managed poorly, stale seed lists, broken attribution, and creative nobody's refreshed quietly drain performance without setting off any obvious alarm.

Every campaign that runs leaves behind real evidence: which audience segments converted, which creative angles actually worked, which accounts on a target list engaged. The next campaign has to be built on top of that evidence for it to pay off. Predictive Audiences benefits most from this pattern, since a model fed months of clean, high-volume conversion data converges faster and performs better than one that gets reset every time a new campaign launches.

Seed lists need refreshing quarterly at minimum, but companies adding customers fast should do it monthly. Creative needs a monthly refresh too, so fatigue doesn't quietly degrade the signal the model is learning from. Audience size needs regular monitoring, since a Matched Audience that drops below 300 matched profiles stops delivering altogether, flagged in Campaign Manager with a "Not delivering - Audience too small" warning. Attribution needs regular reconciliation between Campaign Manager and the CRM, to catch the pipeline that the default attribution window misses. And suppression lists need upkeep too, so you pull current customers, recent churns, and already-converted contacts out of acquisition campaigns on an ongoing basis.

That adds up to a real question for any growth leader running LinkedIn right now: does the team actually have the time, the infrastructure, and the continuous attention this takes? Or is the program running on settings that were right when it launched and haven't been touched since?