What Actually Works on LinkedIn: 10 Lessons from AI-Optimized B2B Campaigns

Introduction

We spent the last year running LinkedIn ads for B2B companies using AI to handle decisions most agencies and in-house teams do manually — or don't handle at all. What follows is a direct account of what we've learned. Not theory. Real patterns, from real accounts, with real numbers.

Some of this will confirm what you already know. Some of it will contradict what your current agency is doing. Either way, we hope it's useful.

One key thing to understand first

I always say the same thing to anyone who asks me the cardinal question: "So what really works for paid channels?" No one knows what will work before you test it. Not the smartest agency in the room. Not you. Not us.

The companies that win on LinkedIn aren't the ones who figured out the perfect setup on day one. They're the ones who learn fastest. That's the thesis behind everything below. AI is wicked fast at learning, and these are the biggest lessons we've learned from using AI to optimise our customers' campaigns.

1. Pipeline influence vs. lead generation

Most LinkedIn programs are set up as lead generation campaigns. For a lot of companies, that's right. But for companies selling into large enterprises — Fortune 500, large financial institutions, big manufacturers — it often isn't. We see it take 100-200 touchpoints with a company before they decide to book a meeting. If you're running lead gen on that audience, you're asking for a commitment they're not ready to make.

Pipeline influence campaigns flip the model. Instead of driving form fills, the play is to take a target account list — companies your sales team is actively outbounding — and run LinkedIn exclusively for those people. The goal is to make your brand feel everywhere for those specific accounts.

One account generated $636K in pipeline influence on roughly $3,200/month in spend. The attribution wasn't last-touch LinkedIn lead. It was LinkedIn as the reason sales got in the door.

Ask yourself: do you know which of your LinkedIn programs are helping sales move faster on accounts they're already working — not just generating new MQLs?

2. Offers and incentives

There's an arms race happening. $50 gift cards. $100. Even $500 in some cases, just to take a meeting. Done badly, this is exactly what your sales team worries it is: people showing up for a gift card, not for your product. But done well, offers convert at a remarkable rate — and the leads are legitimate.

• Targeting has to be tight. Senior executives won't bother with $100 if your product isn't relevant to their actual problems. The offer doesn't fix bad targeting — it amplifies it in both directions.
• The best-performing offers are usually the ones most specific to the ICP. For technical audiences, engineering equipment consistently outperforms generic gift cards. For one cybersecurity company, a custom Lego set themed around their product line created a cohort of engineers who were genuinely excited to engage.
• Intrinsic offers also consistently outperform cash in the right context — framing the first call as an AI strategy session, or giving away benchmarks and research reports usually reserved for paying customers.

Ask yourself: is the feedback from your sales team about "bad leads" actually measured, or is it anecdote? And when did you last test an offer tailored to your ICP's interests?

3. Bidding

LinkedIn's default campaign setting is Maximum Delivery. What that means in practice: LinkedIn spends your money as fast as it can. A more sophisticated setup uses manual bidding — setting a max bid you're comfortable with, then watching it closely. The problem is that ad auctions shift constantly. Underbid and you'll pace at half your budget and wonder why nothing's working. Overbid and you're wasting money you didn't need to spend.

The right approach is to check bids daily: look at LinkedIn's own recommendations, compare them to actual pacing, and adjust to the lowest bid that still allows the full budget to run. Most accounts never do it.

Ask yourself: when was the last time someone checked your bid amounts relative to actual pacing? Not the dashboard summary — the actual bid settings per campaign.

4. Voice of customer

The best ad copy doesn't come from brainstorming sessions. It comes from your customers. One company in the compliance space shifted their entire creative strategy after reading through Gong call transcripts. One specific quote — a customer describing a pain point in their own words — went directly into an ad. It became their best-performing creative by 2x. Not slightly better. Twice as good as everything else in the account.

The bigger lesson: your ICP's language changes. What resonated six months ago may not be what's landing now. AI is making markets move faster, and the vocabulary for problems shifts with it. You need an always-on feedback loop from real conversations, not an annual research project you reference for nine months.

Ask yourself: how old is the language in your current ads? When did you last pull actual phrases from recent customer conversations and test them as hooks?

5. Audience testing

LinkedIn has over 40,000 job titles you can target, plus job functions, seniority levels, skills, company sizes, and industries. The permutations for any ICP are essentially infinite, which means it's very easy to pick the wrong one.

We see 30-100% improvements in conversion rates from testing different audience builds for the same ICP. Sometimes job function targeting crushes job title targeting. Sometimes skills-based audiences — built by looking at what skills people who've actually bought from you have in common — outperform everything else. You can't know until you test. Start with recent closed-won data: who bought in the last one to three months, and what were their titles, functions, and seniority levels?

LinkedIn Audience Network is checked by default in every campaign. That single setting sends your ads to third-party mobile apps and websites, not LinkedIn. One account had spent $7,400 with 99% of impressions served on apps like Solitaire. Zero ICP reached. Disabling it is a five-second fix that costs nothing.

Ask yourself: when your current audiences were built, was it based on recent closed-won data — or a best guess from when you first set things up?

6. LinkedIn objectives

When you launch a campaign, LinkedIn asks you to pick an objective: website visits, lead generation, brand awareness, engagement, video views. There are over ten options. Here's what LinkedIn doesn't tell you: the exact same ad, run to the exact same audience, can produce radically different results depending on which objective you choose.

A/B testing objectives across every main ad format shows the right answer is rarely obvious. For video campaigns, testing video views vs. engagement vs. website visits on the same creative consistently produces surprising results. The "logical" choice is often not the most efficient one.

Ask yourself: does your team have a documented rationale for why each campaign objective was chosen — or did someone pick the default and move on?

7. Ad formats

For top-of-funnel awareness, the most efficient formats are thought leadership ads (promoting organic posts) and video. Their primary job isn't conversion — it's building a retargeting pool of people who've engaged with your content. That warm audience becomes the highest-converting segment in the account when you later run a conversion campaign against it.

For bottom funnel, conversation ads and single image ads tend to be most efficient. Document ads are strong for mid-funnel MQL generation, and work best when the asset title names a specific outcome rather than a generic theme. One account saw its single image CPL drop from $682 to $118 after a creative refresh — same format, same audience, different message framing.

Ask yourself: do you know which ad format is your most efficient bottom-funnel converter right now — and are you actively scaling it?

8. Thought leadership content

Four content themes consistently outperform everything else when organic posts are converted into paid thought leadership ads:

• Proactively answer your ICP's objections. Don't avoid the hard questions about your product — go straight at them.
• Market evolution. What's changing in your category right now? Make your company part of the story, not the whole story.
• Show momentum. If your company grew 4x, say it. Be specific. Your ICP should feel a little FOMO if they're not already talking to you.
• Give things away. Not fluff — real IP. Benchmarks. Playbooks. Frameworks. Resources like this one.

The accounts hitting 2-7% CTR on thought leadership (vs. the LinkedIn benchmark of 0.3-0.5%) are almost always doing at least two of these four consistently.

Ask yourself: what percentage of your LinkedIn content is genuinely useful to your ICP vs. announcing what you do?

9. Metrics to optimise for

Most LinkedIn dashboards lead with CPM, and a lot of teams optimise for it. Most of the time, it's the wrong thing to be looking at. If you're running conversation ads, the metric you care about isn't cost per send. It's three things, in sequence:

• Open rate — how compelling is your sender identity and subject line?
• Click-to-open rate — how good is your message copy, offer, and CTA?
• Form completion rate — how much friction is in your conversion step?

The LinkedIn benchmark for click-to-open rate is 0.9%. With pain-first subject lines and specific incentives, some accounts hit 5.5% — more than 6x the benchmark. Optimise those three things independently. Break every metric into its sub-components, find the one that's underperforming, and fix that variable first.

Ask yourself: do you know where in the conversation ad funnel your conversion rate is lowest — the open, the click, or the form?

10. The real moat: rapid testing

Every item on this list is something we don't know the answer to in advance. The bidding strategy that works for one company fails for another. The offer that converts in one ICP tanks in the next. The audience that seems obvious turns out to be wrong.

The companies we've seen scale from $10K/month to $88K/month in LinkedIn spend didn't start with a brilliant strategy. They started with a solid hypothesis, tested it fast, found what worked, and doubled down while cutting what didn't. The job isn't to run ads. It's to build a learning machine.

Ask yourself: how many distinct experiments did your LinkedIn program run last month? And how fast did you know which ones were working?

The most common mistakes

• Running cold audiences with "book a demo" CTAs
• Leaving LinkedIn Audience Network on
• No retargeting campaign
• Audiences built from old data
• Reporting on CPM as a proxy for performance

Start this week

1. 1Open Campaign Manager. Check whether LinkedIn Audience Network is enabled on any campaign. Turn it off.
2. 2Pull your closed-won deals from the last 90 days. Note the job titles, functions, and seniority levels, then compare them to your current targeting.
3. 3Read the last 20 call transcripts from won deals. Write down exact phrases customers used to describe their pain, and test one as a hook this week.
4. 4Check your conversation ad metrics: open rate, click-to-open, form completion rate. Find the lowest number — that's where to focus.
5. 5Ask your agency (or check yourself): what's the current bid strategy for each campaign, and when was it last reviewed?