AI advertising examples: what real B2B campaigns published, and what they left out
Real, sourced AI advertising examples from B2B and consumer brands, with the published result, who published it, and what each one leaves out.

The most useful AI advertising examples for B2B are the ones with a named company, a stated baseline and a published result: Siemens testing LinkedIn Accelerate and Google Performance Max, IBM producing over 1,000 marketing variations with Adobe Firefly, and Vanguard Institutional testing AI-written LinkedIn copy. Every one of them reports clicks, leads or engagement. None of them reports pipeline.
You've probably been sent at least one of these in a deck. Someone on the leadership team saw "26 times higher engagement" or "77 times more leads" and wants to know why your own numbers look nothing like it. It's an awkward question to answer, because the number is real and it measures something other than what your board asks you about.
This article collects the examples that hold up, with each brand, what it did, the result as published, who published it and when. Then it reads each one the way a CFO would, and says what a B2B paid team can copy from it on Monday. It's written for the marketing leader who has to decide which of these stories to believe and which to fund. Consumer examples are included where they teach something, and they're labelled.
Six AI advertising examples that survive a source check
Most "AI in advertising examples" lists repeat the same stories without links, and several of the links they do give are now dead. We kept an example only if the figure could be read today on the brand's, the platform's or the vendor's own page, or in trade press quoting the company directly. Six passed, plus LinkedIn's own aggregate test data.
| Example | B2B or consumer | What the AI did | Published result | Who published it | What's missing |
|---|---|---|---|---|---|
| Siemens on LinkedIn Accelerate, 2024 | B2B | LinkedIn's AI built and optimized the campaign | 2.4x website visits; lead form completion 5.66% vs 2.56%; cost per lead on par | LinkedIn case study | Lead quality beyond job title and country, pipeline |
| Siemens Digital Industries Software on Performance Max, 2022 | B2B | Bidding, audiences and placement across all Google inventory | 77x lead volume, 97% lower cost per lead than search only | Google (Think with Google) | Lead quality, pipeline |
| IBM "Let's Create" with Adobe Firefly, 2024 | B2B | Generated over 1,000 marketing variations from text prompts | Engagement 26x higher than estimated | Adobe blog | What the estimate was, how engagement was defined, any conversion metric |
| Vanguard Institutional with Persado, 2022 | B2B (financial services) | Generated and tested LinkedIn message variants | 15.76% higher conversion rate vs control | Persado (sponsored) | What "conversion" meant, sample size |
| Klarna, 2024 | Consumer fintech | Image generation and an internal copy tool | $6 million lower image production costs; 25% lower supplier spend | Klarna via Marketing Dive | Any effect on ad performance |
| Coca-Cola "Holidays Are Coming", 2024 | Consumer | Fully AI-generated TV spot | Strong pretest score, then public backlash | Adweek | Sales impact |
Read across the last column and the pattern is plain. The published results stop at the first measurable step after the ad, and that's where the platforms and vendors stop too, because that's the data they own. Teams that want to read an AI test past the form need ad results joined to CRM opportunities, and Scalewell sets that join up so those tests end in a decision about pipeline.
Siemens on LinkedIn Accelerate doubled form completion at the same cost per lead
Siemens ran an A/B test of LinkedIn Accelerate, LinkedIn's AI campaign creation experience, against its Classic campaigns for two objectives, website visits and lead generation. According to LinkedIn's April 2024 case study, Accelerate drove 2.4 times as many website visits with a 58% lower cost per click and a 79% lower CPM. On lead generation, form completion rose to 5.66% from 2.56%, while cost per lead stayed on par with Classic.
That last line is the honest one. Cheaper clicks and higher completion rates are what an AI campaign tool optimizing for the objective you gave it should produce. Cost per lead not moving tells you the gain stopped at the form: each lead cost Siemens about what it cost before.
Lance Murphy, VP Global Demand Strategy at Siemens, said the team "also saw improvements in the demographics of the leads for both job title and country targeted." That's the most B2B-relevant sentence in the case study, and it's still a description of who filled the form, not what happened after.
What a B2B team can copy: run the AI campaign type as a split test against your current build with the same budget and objective, which is exactly what Siemens did. Then take the leads from both arms into your CRM and compare opportunity rates before you move budget. The humans in the loop on ad spend piece covers where the human check sits.
LinkedIn's wider numbers point the same way. In June 2024 LinkedIn said advertisers using Accelerate saw a 52% lower cost per action than Classic campaigns, based on 29 A/B tests. By October 2024, with 67 tests, the figure had become "up to a 42% lower cost per action". As the test pool grew, the headline figure fell and became an "up to" number, which is normal and worth remembering when an early figure is quoted to you as settled. LinkedIn has since retired Accelerate as a standalone campaign type, as of August 2026, and moved its automation into Classic campaigns, BrandRocket reported, so read these figures as a record of a 2024 product.
Siemens on Performance Max got 77 times the leads against a weak baseline
Siemens Digital Industries Software tested Google Performance Max for its Solid Edge for Startups Program, with the agency iProspect, as a 2022 pilot in the US, Canada and the UK. Google's one-page case study, published in German on Think with Google, reports that lead volume grew 77-fold and cost per lead was 97% lower than with a search-only campaign. Siemens then rolled the campaign type out to the DACH markets, and Lance Murphy, then VP Global Marketing Execution, said the company planned to extend it to all campaigns.
So why not copy it straight away? Because the same case study explains the baseline: the broad search campaigns had been producing too few qualified B2B leads in a competitive market. A 77-fold gain over a campaign that barely worked says more about the starting point than about the method. And Performance Max optimizes toward whatever conversion you count, across YouTube, Display and Search, so if the conversion is a form fill, it will find you form fills.
It's a bit like a sales rep who doubles their meeting count by booking anyone who picks up the phone. The number is real. You still want to see the calendar.
What a B2B team can copy: use the automated campaign type, but feed it a conversion that means something. Google Ads offline conversion import or enhanced conversions for leads lets you count qualified leads or opportunities instead of raw submissions. Without that, a broad AI campaign is free to fill your CRM with whoever fills forms most cheaply. Our AI performance marketing guide covers how to set the outcome you feed the model.
IBM generated over 1,000 variations and reported engagement, not conversion
IBM used Adobe Firefly, Adobe's generative image model, in an early pilot for a marketing campaign that CIO&Leader reported as its "Let's Create" brand campaign. According to Adobe's May 2024 post on its IBM partnership, IBM used simple text prompts to generate over 1,000 marketing variations "in a matter of minutes," and the campaign "performed well above IBM's benchmark, with engagement 26 times higher than estimated." CIO&Leader's March 2024 report adds that 20% of respondents were identified as C-level decision makers.
The production speed is the part you can rely on, because IBM could count the variations directly. The 26x figure is harder to use. It's measured against an estimate nobody states, neither source defines engagement, and the vendor published it.
The 20% C-level figure is the interesting one for B2B, because it says the variations reached the buying committee. It doesn't say any of them bought.
What a B2B team can copy: use generative tools to make more variations for a test you've already designed, with one idea per variation, rather than flooding the ad set and letting the platform pick. A creative brief written before generation keeps the variations testing something. Our sibling piece on AI generated ads for B2B covers where generated visuals start to look cheap.
Vanguard Institutional found the words that converted on LinkedIn
Vanguard Institutional, the institutional business of the investment firm Vanguard, used Persado's AI to test its LinkedIn messaging. Persado's January 2022 case study says the work raised conversion rates by 15% and notes that LinkedIn was the only social platform the division used. A sponsored Persado article on Adweek gives the detail: eight AI-generated variants, representing 64 permutations of the message, were tested against Vanguard's own control copy, and the winner delivered a 15.76% higher conversion rate.
This is the most methodical example on the list, because it had a control and isolated the words. It's also the oldest, and every source is published or sponsored by the vendor, with no sample size and no definition of conversion. Treat it as proof that structured copy testing works on LinkedIn, and as no proof of a size of effect you should expect.
What a B2B team can copy: the method costs nothing. Break your best-performing ad into components (opening line, proof point, call to action, emotional tone), have an AI model write variants of one component at a time, and test against your current copy. Teams doing this for Google search ads can start from our ChatGPT prompts for Google Ads.
Klarna cut production cost, a consumer example with a B2B lesson
Klarna, the consumer payments company, said in May 2024 that AI helped it cut sales and marketing spend by 11% in the first quarter while running more campaigns. As reported by Marketing Dive, generative AI tools such as Midjourney, DALL-E and Firefly saved $1.5 million in the quarter, image production costs fell by a further $6 million, and the image development cycle dropped from six weeks to seven days. An internal tool, Copy Assistant, was used for 80% of copywriting.
Notice what Klarna measured: cost and time. It didn't claim the AI images performed better, only that they cost less and arrived faster. For a B2B team with a small creative budget, that's the realistic version of the AI creative pitch, and it's the version a CFO will accept, because the saving shows up in an invoice rather than a dashboard.
Coca-Cola shows the brand risk, a consumer example worth knowing
Coca-Cola remade its 1995 "Holidays Are Coming" TV ad with generative AI in 2024. Adweek reported that the spot earned a perfect 5.9 score from System1 in both the US and the UK, then turned into a public backlash, with critics saying it drained "the warmth and joy out of a holiday classic."
The lesson for B2B is narrow but real: a pretest score measures how an audience reacts to the ad, and it can't measure how people react to learning how it was made. Your buyers are less likely to start a pile-on, but a visibly synthetic face in a LinkedIn ad for a trust-based product sends its own message. Keep generated imagery for backgrounds, objects and concepts, and use real people where trust is the product.
AI ad case studies share three habits that flatter the result
Published AI advertising examples tend to flatter the result in three repeatable ways, and knowing them makes every new case study faster to read.
- The baseline was weak. Siemens' search-only campaign was underperforming before Performance Max arrived. A big multiple usually means a low starting point.
- The metric stops at the platform. Every example above reports clicks, completions, engagement or cost. The platform can see those; it can't see your CRM unless you send the data back.
- The publisher sells the tool. LinkedIn, Google, Adobe and Persado published these results. That doesn't make them false, and the Siemens LinkedIn case even admits cost per lead didn't move. It does mean you won't see the tests that went nowhere.

Here's a quick way to see what that gap costs. Imagine two hypothetical campaigns, A and B, each spending $20,000. Campaign A produces 200 leads at $100 each, and 4% become opportunities, so that's eight opportunities at $2,500 each. Campaign B produces 100 leads at $200 each, and 15% become opportunities, so that's 15 opportunities at about $1,333 each. A case study written about A would report twice the leads at half the cost. Your sales team would rather have B.
That isn't a hypothetical worry. None of the published AI examples above would catch a gap like that, because none of them measured as far as opportunities. Our AI marketing ROI piece shows how to set up the measurement.
Run your own AI example before you cite someone else's
The only AI advertising example your CFO will fully trust is one from your own account, measured in your own CRM. You can build one in a quarter.
- Pick one job. Choose campaign building (LinkedIn's campaign automation or Performance Max), creative variation, or copy testing. Each of the examples above did one of these, not all three.
- Set up a split. Run the AI version against your current build with the same budget, audience and objective, as Siemens did on LinkedIn.
- Send outcomes back. Connect qualified leads or opportunities to the platform through LinkedIn's Conversions API or Google Ads offline conversion import, so the AI optimizes toward something sales cares about.
- Read it at opportunity level. After enough time for leads to become opportunities, compare cost per opportunity between the arms, not cost per lead.
- Write it up honestly. Note the baseline, the metric, the time window and what didn't move. You'll have a better case study than most of the published ones.
For a wider view of where AI fits across a paid team's work, the generative AI in advertising guide covers the tools and the governance.
Where to start
Pick the one example above that matches a job your team does every week, and run it as a split test measured at opportunity level this quarter. Scalewell helps paid teams set up these tests and the CRM measurement behind them, so the next AI example in your board deck is your own.
Frequently asked questions
What is an example of AI in advertising?
A clear B2B example is Siemens letting LinkedIn's AI build its campaigns in a 2024 A/B test, which more than doubled lead form completion while cost per lead stayed level. Others with published figures include Siemens on Google Performance Max, IBM generating marketing variations with Adobe Firefly, and Vanguard Institutional testing AI-written LinkedIn copy.
How are companies using AI in advertising in 2026?
As of November 2026, companies use AI in three main ways: platform AI that builds campaigns and sets bids (LinkedIn's campaign automation, Google Performance Max and AI Max, Meta Advantage+), generative tools that produce images, video and copy variations, and AI models that test and rewrite messaging. Most published results cover the first two.
Do AI-generated ads perform better than human-made ads?
No published, independent B2B study settles it. Vendor case studies, such as IBM's work with Adobe Firefly, report higher engagement, while Klarna's public figures cover only lower cost and faster production. The fair answer is that AI makes testing cheaper, and the test tells you which ad performs better in your account.
Is AI-generated advertising risky for a brand?
It can be. Coca-Cola's 2024 AI holiday ad scored well in pretesting and still drew a public backlash once people focused on how it was made. For B2B brands, the bigger day-to-day risks are generic-looking visuals, inaccurate claims in generated copy, and synthetic people in ads that depend on trust.
Which AI advertising tools do B2B teams use most?
B2B teams most often start with the AI built into the platforms they already buy: LinkedIn's campaign automation, Google Performance Max and Meta Advantage+. Generative tools such as Adobe Firefly and Midjourney handle visuals, and general AI assistants such as ChatGPT and Claude draft copy and analyze performance. We found no public survey that ranks them reliably for B2B.