Show Resources

Here were the resources we covered in the episode:

How to get the best match rates from LinkedIn list uploads

Why aren’t lookalikes very valuable for targeting on LinkedIn?

LinkedIn’s description of Predictive Audiences

Join the LinkedIn Ads Fanatics community and get access to our 4 courses to take you from beginner to expert

Follow AJ on LinkedIn

B2Linked’s YouTube Channel

LinkedIn Learning Course

Contact us at Podcast@B2Linked.com with any questions, suggestions, corrections!

A great no-cost way to support us: Rate/Review!

Show Notes: Episode Summary

In this episode of the LinkedIn Ads Show, AJ Wilcox dives deep into the new feature of LinkedIn Ads: Predictive Audiences. Here are the key discussion points:

1. Introduction to Predictive Audiences:
   • Predictive Audiences are replacing the traditional lookalike audiences on LinkedIn.
   • They dynamically update and allow more control over audience matching using a slider.
2. Testing Predictive Audiences:
   • AJ shares insights from his tests comparing seed lists with predictive audiences.
   • He highlights a significant drop in audience size when applying specific targeting criteria.
3. Performance Insights:
   • Despite smaller audience sizes, predictive audiences showed higher click-through and conversion rates compared to standard audiences.
   • Discussion on how predictive audiences might be leveraging LinkedIn’s AI to focus on intent signals rather than traditional targeting facets.
4. Audience Composition Analysis:
   • Comparison of job functions, company sizes, and industries between seed lists and predictive audiences.
   • Notable shifts in audience composition, with predictive audiences often deviating significantly from the original seed list.
5. Listener Interactions:
   • AJ reads a positive review from a listener, highlighting the impact of the podcast.
   • Encouragement for listeners to leave reviews, join the LinkedIn Ads Fanatics community, and participate in discussions.
6. Practical Tips:
   • Recommendations for setting up predictive audiences with clean seed lists and appropriate guardrails.
   • Advice on leveraging audience insights for better campaign performance and understanding.

Call to Action: Join the LinkedIn Ads Fanatics community to deepen your expertise and engage with other LinkedIn Ads professionals. Subscribe to the podcast for weekly insights, and leave a review on Apple Podcasts to show your support.

Review and Connect: Listeners are invited to leave reviews, send questions or feedback via email at podcast@b2linked.com, or message AJ directly on LinkedIn. AJ’s DMs are open for any inquiries and suggestions.

Listeners interested in advanced LinkedIn Ads strategies and real-world performance data will find this episode particularly insightful.

Show Transcript

Move over, Ms. Cleo, you can take your crystal ball and shove it, because LinkedIn Ads now has predictive audiences. That’s right, we’re talking all about predictive audiences on this week’s episode of the LinkedIn Ads Show.

Welcome to the LinkedIn Ads Show. Here’s your host, AJ Wilcox.

Hey there LinkedIn Ads fanatics. As he said, I’m A. J. Wilcox. I’m the host of the weekly podcast, The LinkedIn Ads Show. I’m thrilled to welcome you to, if I don’t say so myself, the show for advanced B2B marketers. You’re going to evolve and master LinkedIn Ads and achieve true pro status.

I knew I wanted to have an episode all about predictive audiences, but I wanted to run a solid test first so I had some good data to share and results. I was going to compare the audience make up between the seed list that feeds the predictive audience and the predictive audience, which I’m still definitely gonna do. As I was building the campaigns, I noticed something pretty shocking, so I wanted to cover that first.

The LinkedIn Ads Show is proudly brought to you by B2Linked.com, the LinkedIn Ads experts.

That’s right, B2Linked is the ad agency 100 percent dedicated to LinkedIn Ads and we have been since 2014, you know, back before it was cool. We build custom strategies for every account we work with. You get to work directly with me and my local team. You won’t get a cookie cutter approach or standard account templates from us. If you want a white glove experience without the white glove price tag, we’d love to explore partnering together. And you can schedule your free discovery call at b2linked.com/discovery.

All right, first off in the news. I’m curious if any of you have activated the conversions API on LinkedIn through Google tag manager. I’ve been digging in really deeply to it. There’s going to be some really cool podcast episode or episodes about it in the future. I’ll warn you. I definitely think the topic is worthy of multiple episodes. But new to LinkedIn, when you go to actually set up the conversions API, there’s now an experience that walks you through how to set it up. Whereas it used to be that you were kind of on your own. I haven’t had a chance to fully test it, just I kind of peeked into what the flow looked like and it looks good , but I don’t know if it’s fully functional. I haven’t tried it yet. I I’m sure I will in the future, but I’ve got something cool coming for you if you’re interested in, in the conversions API through tag manager.

Also, when you run Thought Leader Ads, there’s a new metric that you’ll see. You might already see it. It’s probably mostly rolled out by the time you hear this. And it’s called click to member profile. You’ll be able to find it in the engagement view. And this is going to measure the people who actually click on the Thought Leader’s profile image or on their name from within the Thought Leader Ad.. And this is great for the Thought Leader. They can get follows and comments and things like that. Now we get to track those who actually go to check out their profile. It’s really cool data to be able to share with the Thought Leader.

Also a shout out to Tanner Stulte for bringing this to my attention. I didn’t realize it, but LinkedIn actually changed the name of LinkedIn Ads. It used to be called LinkedIn marketing solutions. Then someone had a really bright idea that I agree with to call it LinkedIn ads. But just in the last couple of weeks, LinkedIn has again, rebranded their ads product. So it is now LinkedIn For Marketing. Honestly, I think this is a terrible decision. I think LinkedIn, your customers already call it LinkedIn Ads.. They called it LinkedIn Ads back when you were still trying to call it LinkedIn Marketing Solutions. They’re still going to be calling it LinkedIn Ads.. I would highly suggest roll with the punches, call it what’s actually easiest for your customers to call it in protest of this. You’re still going to be hearing me call it LinkedIn Ads.

I wanted to highlight a review we got here on the show. Victoria Dondonville left, "Hi AJ, I love your show and I’m a regular listener almost daily now during my commute to work. You’ve taught me so much about LinkedIn Ads through each episode. I love how direct and concise your messages are. I feel like I’m getting so much out of my time I spend listening to your show. I just signed up for the B2Linked fanatics community, and I’m very excited to hopefully learn even more from you there. Keep up the great work." Senior Digital Marketing Strategist at UpGrow in San Diego. Victoria, thank you so much. That was such a kind and heartfelt note to me. Of course, when you’re podcasting, you never know what difference you’re making. You never know who’s listening or who it’s impacting. So a huge heartfelt thank you for leaving something so personal and really excited that I get to interface with you in the LinkedIn Ads fanatics community.

All right, for everyone else, do you have a question, a review, or feedback for the show? You can either message me privately on LinkedIn, my DMs are open and free, or you can email us at podcast@b2linked.com

You can attach a link to a voice recording from you, and I can play it right here on the show, or you can obviously just leave text. I’m happy to keep you anonymous, or shout you out as well. I hope you all know by now, I want to feature you. Alright, onto the topic at hand, let’s hit it.

So back in October of 2023, LinkedIn announced that they were replacing lookalike audiences with something new. Which I was 100 percent behind, because I thought that lookalikes were pretty useless, honestly. The reason why is that LinkedIn’s lookalikes were a snapshot, meaning that when you create that audience, it stays that way forever, and doesn’t update when the seed list changes, so it gets out of date. And we know that people are constantly moving in and out of the workforce, getting promoted, moving into positions where you might want to target them, or moving out of positions where they’re no longer interested. And you want your data to actually follow that dynamic nature of the audience. There was also no control over how tight or loose the resulting audience would be. Those of you who are familiar with lookalikes on meta ads, you know that this is one of the greatest features that Meta offers is the slider. You get to tell Meta how tight or how liberal to make your predictive audience.

So predictive audiences is the answer to this on LinkedIn. It’s dynamic. So if you build a list that changes like a retargeting list or a dynamic list synced to your CRM, it’s going to update every night and stay up to date. It also gives you the control over how liberal you want LinkedIn’s matching algorithm to get by using the same kind of slider. Pretty cool. Not quite as slick as meta, but I think we’re getting there.

All right, so here’s the test that I planned to run. My LinkedIn reps had been pressuring us to try predictive audiences for some time for a client. We thought that the very best test to roll this out for would be to an existing customer list to see how well it can find people who are most like the client’s past customers. We already had a customer contact list that we were using for exclusions, because, of course, why pay for clicks from people who are already your customer, right? And the client is a SaaS company who traditionally targets HR professionals. So, naturally, I expected that the job functions were mostly going to be made up of HR. The list was created through a third party integration with LinkedIn, and as such, it only had a 25 percent match rate.

Which, if we processed it manually, I know we’d be able to get it much higher. And if you’ve listened to episode 120 of the podcast, you’ll be able to too.

But it resulted in a 57,000 person contact list. So I decided to use that list as the seed for my predictive audiences. So I could compare the audiences. While I was at it, I actually used it to create two different sizes that I could compare. So the first predictive audience I created, I limited it to 200,000 people. And the second one, A little broader at 500,000.

My original goal was to use audience insights to compare the seed audience against the predicted audiences to see if I could tell what sorts of professional traits LinkedIn likes to base its predictions off of. So I’ll definitely share that analysis in a bit, but, I’ve got something else I want to share first. I took the 200,000 person predictive audience, and I put what I call guardrails around it. It’s just some basic targeting, just to make sure that the algorithm can’t go too crazy, and it’s going to focus around the ideal target audience that the client wants to go after.

So, in this case, it’s HR job function with company sizes 11 to 1000. Those were my guardrails. And it was kind of crazy, I noticed that as soon as I added the HR job function, that that audience size of 200,000 dropped like a brick from an airplane, down to 4 percent of its original size. Guys, that’s even worse than decimating. It’s less than half of a decimation. Okay. So, and then after I added company sizes, 11 through 1000, tt dropped even further to 2 percent of its original size. I was just wondering, did I make a mistake? What did I do wrong? What happened to this audience? I checked the same thing on my 500,000 person audience, and it was even worse. It dropped to 3 percent of its original size, just with the HR job function, and further dropped to 1 percent of the original audience. when I added the company sizes. Of course I needed to troubleshoot this, so I tried putting the same guardrails on the original audience, and sure enough, it dropped too, but not nearly as bad as the predictive audience. It dropped to 15 percent of its original audience. So we have the original audience where 15 percent fits in our ICP and the predictive audience dropped to 2 percent of its original size. Alright, so now I’m really scratching my head. I realize that this audience is very small, but I still wanted to test it out anyway. And I had about a $3,000 budget to allocate to it, so I would call this pretty healthy. And I was actually really pleasantly surprised with ad performance. It was great. Compared to our normal audiences that were usually getting a click through rate of about .64%, the predictive audiences got over 2%. The offer was going to gated content that we’d spent lots of money on, had really good historical data, And it was really cool to see the conversion rate was 219% percent higher than average.

All right, so how did predictive audiences expand the list that allowed it to cut so many people out of it as soon as I put the basic guardrails on? So I did a deep dive into the audience insights of the original audience versus the two predictive audiences to see if I could understand what predictive audiences was keying off of and what was most important to it.

First off, I had to look at job function. The original list only had 22 percent in the HR job function, and it actually had slightly more in business development, which actually makes sense because that’s what business owners and CEOs fall into. So obviously there are some of those folks in the seed list and actually quite a few, probably even more than HR people. But the rest were a smattering across operations, finance, IT and a whole bunch of others in like the lower single digit percentages. When I compared that to the predictive audiences version of job functions, the predictive audiences went way outside of this distribution. It expanded from 22% business development to 37% and finance jumped from 6% to 12%. So we have to ask ourselves, where was HR? In our original audience, it was 22%, but in the predictive audience it dropped down to 4%. What a drop. Unless you think that the larger of the predictive audiences did things any differently? No. The 500,000 person audience was quite similar. HR actually dropped from 22% down to 3% of total audience.

All right, so predictive audiences was obviously not trying to stick really closely to job functions. What about company size? The original list had about 69% of its audience in that range I told you about, the 11 to 1, 000 person companies. So how did predictive audiences fare in this range? I’ll tell you, not great. The predictive audiences dropped from 69% of the audience being in the company size range, down to only 34%. So predictive audiences ended up doubling the number of people outside of the company ranges that we cared about. So where did the rest of the company sizes go? I’ll tell you, it looks like it went upstream. Only 3% of the original audience was in the largest company size bracket of 10,000 plus. We consider these very much enterprise size companies. But the predictive audience went from 3% up to 22% in that largest company bucket. Pretty wild.

So it’s not looking like company size was a big comparison point in how LinkedIn builds the predictive audiences. How about industries? The top three industries that made up the seed list were technology, software development, and IT services. And together, the three of them made up 33% of the original seed list.

Looking at the predictive audience, that dropped from 33% down to 11%. So again, seems like predictive audiences doesn’t really care about industries all that much. What was interesting, it did skew towards companies with positive growth rates, which I thought was kind of interesting. The predictive audience had half as many negative growth rate companies as the seed list had, and had big increases in those company growth rates between 0 and 10%. The fastest growing companies of 20% plus, though, did drop. In the seed list, it was made up of 14%, but it drops in the predictive audience down to 7%, so half as much.

Then I moved on to seniorities, predictive audiences actually kept seniorities surprisingly consistent, but they added quite a few more into entry level than we had before. The rest were a pretty similar distribution, though. I’m guessing that the matching algorithm probably considers seniority level fairly closely.

But job titles, that’s a field I would expect the lookalike algorithm to rely heavily on. But as I compared the titles across all three, the seed, the 200,000 person list, and the 500,000 person list, they were all over the board.

There were two top HR titles that made up 23 percent of the seed audience. In the predictive audiences, they were reduced down to 3% or less. Ouch. It added a bunch more presidents, owners, that weren’t as prevalent as before. So again, I would imagine that job titles, probably not a big focus for the lookalike algorithm.

You know, when LinkedIn shows you a list of the facets that make up your audience, and when you add all of them up, they don’t add up to a hundred percent. Something else I noticed is that in all cases, predictive audiences added more unknowns into themselves than there were in the seed list. For example, company size of the seed list added up only to 91%, meaning that 9% were companies that were unknown. Predictive audiences only added up to 86%, so that increased it to 14% of the audience being unknown company sizes. The largest predictive audience was even worse, increasing to 28% being unknowns. The same effect showed up in all the other facets of comparison, like company growth rates and seniority. I wanted to go read the LinkedIn help article all about predictive audiences to see what they said about it. LinkedIn says that predictive audiences create an audience predicted to perform actions similar to those within your source data. To create a predictive audience, we combine your data source And LinkedIn’s AI to automatically generate a new custom audience to use in your campaign.

So the two things I want to point out here that are absolutely crucial are them saying that they’re creating an audience predicted to perform actions similar to those in your source data. And then finally, they mentioned that they take your source data and combine it with LinkedIn’s AI. Those are our two clues that probably explain a lot of what I’ve seen here. It doesn’t seem like many of the distributions of different targeting facets that I would think are actually used very heavily in the lookalike. And that’s because the predictive audiences is not a simple lookalike model like the last one. Like lookalikes and like audience expansion. It’s based heavily off of activity. The help article also says that your data source must have 300 or more members to create a predictive audience. But, really cool hack, you can select multiple sources within a data source to reach 300 people. So you can have smaller retargeting audiences, Things like form fillers, but interestingly enough, you can’t use this with contact lists like that. Your contact lists have to be over 300 alone. You can’t daisy chain a whole bunch of, you know, 60 person audiences to get there. There was also a note at the bottom of the article that audience expansion is disabled for campaigns that use predictive audiences. And I just need to say, yay, finally, something so useless was taken out of a default setting on something.

Now, I posted about predictive audiences this week on LinkedIn, and I had two people chime in with information that I thought was really interesting.

First off, Kerstin Haag, who’s Channel and Content Manager at the Zurich Tourism in Zurich, Switzerland. She shared that she’s been running some of these predictive audiences tests. She’s averaging a click through rate that is 2.5% compared to 1% being the baseline. And her conversion rates are 120% higher. This very closely echoes the performance that I’ve seen as well in my test. Seems like click through rates are higher and conversion rates are higher.

There’s also a comment from Miikka-Markus Leskinen, Digital Marketing for Fonecta in Finland. And Miikka said, "So LinkedIn has access to those LinkedIn profiles that are not easily targeted by audience attributes." These are, for example, very senior level decision makers who on LinkedIn have some very obscure profile information. They might be an engineer at some holding company, or a job title that really isn’t like sales at company. Many people don’t have a clear job function or seniority, but are very active on LinkedIn. Miikka, I’d never considered this before. That’s actually a really good model of how predictive audiences are probably working. Because when we see all of those unknowns start to creep up in the predictive audiences, There are a lot of members who are unknowns.

LinkedIn doesn’t know how to categorize them. And when we see them gaining and performance also gains, ah, it’s starting to make a little bit more sense. Mika also said that one other cool thing about predictive audiences is that only active members can be members of this audience. And I hadn’t heard that before. So I asked Mika for a source on it, but it seems totally plausible from what I’ve seen. And if so, I think that’s awesome.

All right, so for takeaways and advice for you running your own predictive audiences, what should you expect predictive audiences to rely on most heavily for matching? It seems like seniority and years of experience had the most consistency, and then maybe a little bit of job function and maybe titles to a lesser degree.

But if I had to guess, the vast majority of the picking is actually in intent signals that LinkedIn is picking up on and moving them over, despite what their targeting says. Overall, like I mentioned on episode 11, on targeting when I talked about lookalikes, I don’t see much of a purpose behind lookalikes on LinkedIn. It can be really worthwhile to test them, don’t get me wrong. And especially with these big increases in click through rate and conversion rate, you might be rewarded for trying them out. But one of the main things I don’t like about predictive audiences, or any other AI created audience, is it creates a black box that you get next to no insight into. When you have an audience like this that’s a black box, you don’t end up learning anything about who your audience is and what they like. Of course, you can break that audience down in audience insights and demographics, but And that’ll tell you, like, which of these facets have a higher click through rate. But those insights don’t carry down into the CRM, where I can start to measure things like conversion rates and lead quality. Which I think is one of the best parts about segmenting audiences on LinkedIn by campaigns. I call it that I’m building little focus groups that help me learn about who my audience is and what they care about.

All right, so the big question is, should you go and build a predictive audience? Totally up to you, but make sure to put some guardrails on it. And the thing that would have made my test a lot cleaner, I highly recommend starting with as clean of a list as you can possibly get, without noise. I would bet that if I uploaded a seed list that were mostly in the job function of HR, and in the company sizes that I cared about, the resulting audience would likely be a lot closer.

All right, if you’re not already a member of the LinkedIn Ads fanatics community, what are you waiting for? Get in there now! A very low monthly fee gets you access to all four of our courses that take you from beginner to expert. Plus, you get access to all of the top minds in LinkedIn Ads sharing what’s working for them, what they’re testing, and their results. And if you’re an upgraded member there in the SuperFanatics, you get on a weekly group call with me, and I’d love to get to talk to you about what you’re learning and point out things that you could be improving on. If this is your first time listening, welcome! Please hit that subscribe button. because I’m going to be dropping insights like this again next week, and the week after that. But if this is not your first time listening and you’re a loyal listener, thank you for coming back. Please do consider going to Apple Podcasts and leaving us a review. Not only am I going to shout you out, but I will tell you that is the most meaningful thing that you can do to me. It might make my poor ginger heart burst.

With any questions, suggestions, or corrections, please reach out to us at podcast@b2linked.com. With that being said, we’ll see you back here next week. And as always, I’m cheering you on in all of your LinkedIn Ads initiatives.

‍