AI in B2B Ads: What's Actually Working (After $250M in Spend)

Introduction

In this webinar hosted by Darshan from Factors, Kevin Lord Barry, the founder of Right Percent and author of "Do What Works: The B2B Ads Handbook", shares a practical, no-hype look at how AI is actually being used in B2B advertising. Drawing from over 12 years of experience and $250 million in managed ad spend, Kevin breaks down the exact tools, workflows, and strategies that work for high-growth accounts. Readers will learn how to build an AI-assisted creative pipeline, set up algorithmic targeting, and avoid common pitfalls when integrating AI into reporting.

About Kevin Lord Barry

Kevin Lord Barry is the founder of Right Percent, a B2B advertising agency he has run for seven years, and has over 12 years of experience in B2B advertising. He is the author of "Do What Works: The B2B Ads Handbook", which is currently the only B2B advertising book available on Amazon. Throughout his career, Kevin has managed over $250 million in ad spend, working with major clients such as Reddit, Brightwheel, and Get Your Guide.

The 80/20 Rule of B2B Ad Creative

Kevin shares his foundational framework for what actually drives performance in a B2B static ad, based on nine years of continuous testing.

• The visual headline: The text written directly on the ad image accounts for 80% of the ad's performance because it immediately signals to the target audience that the ad is for them.
• Supporting imagery: The actual image or graphic behind the text is secondary, serving only to support the visual headline.
• Ad copy: The body copy and description text surrounding the image make up the final 20% of the ad's performance and are far less critical than the headline.
• Positioning the offer: The visual headline must be treated as a way to position the core offer (such as a gated ebook or landing page) in as many different ways as possible to find what resonates.

Building an AI Creative Assembly Line

Kevin explains how he built a custom, automated ad generation system to replace the traditional copywriter-to-designer workflow.

• Vibe coding with Replit: Kevin built a custom ad assembly line website using replet.com (specifically replet.app), an AI-assisted coding tool that allowed him to build a complex site with API connections simply by talking to the AI.
• Splitting the AI roles: Kevin's rule of thumb is that you cannot generate the headline and the image in a single step; you must treat them as two separate AI employees—one for copy and one for design.
• Generating headlines with Gemini: The system uses Google's Gemini AI to read uploaded PDFs (such as a gated ebook on "how advertisers acquire advertisers") and automatically extract 10 distinct ad headlines.
• Top image APIs: For the design step, Kevin narrowed down his testing of image models to three top APIs: OpenAI's DALL-E (which released a new model during the week of the webinar), Meta Muse, and Ideogram.ai.

Overcoming Brand Guideline Limitations

While AI can generate images quickly, strict brand standards present a major hurdle for enterprise B2B companies.

• Training on brand guidelines: Kevin demonstrated that feeding Reddit's official brand guidelines into an OpenAI image generator produced ads that looked significantly more on-brand than generic outputs.
• The font and color hurdle: AI still fails when strict rules are applied, such as requiring a specific font like Arial Bold or exact hex codes for brand colors.
• Template-based automation: To solve the brand compliance issue, Kevin uses bannerbear.com to build rigid templates that automatically fill in text while preserving pre-approved fonts, assets, and layouts without relying on generative AI.
• Automated testing volume: For brands with flexible guidelines, Kevin's system features a toggle that automatically generates 9 new ads every week for rapid testing.

AI Video Editing and Resizing Tools

Video ads are highly effective but traditionally time-consuming to produce. Kevin highlights the specific tools his agency uses to streamline video workflows.

• Resizing with Gamma: Kevin uses gamma.app to easily convert square ads into vertical formats for platforms like Instagram Reels.
• Video editing with Async: Right Percent uses async.com to quickly parse long-form client assets, such as a 20-minute conference video, to find the best hooks and edit them into short-form ads.
• Static-to-video generation: Kevin uses runwayml.com to animate static ad images into high-performing video creatives.
• Safe-zone prompting: To prevent vertical video elements from being cut off by platform UI overlays, Kevin adds specific instructions to his AI prompts, such as "do not fill in the top and the bottom."

Algorithmic Targeting vs. List-Based Targeting

Kevin explains how B2B advertisers must navigate the shift toward platform-native AI targeting on networks like Meta, LinkedIn, and Google.

• Zuckerberg's targeting bet: Kevin references a recent interview where Mark Zuckerberg stated that Meta is actively prioritizing its AI computing power for ad targeting algorithms over selling it to third parties because it drives higher ad revenue.
• The 200,000 audience threshold: Kevin's rule of thumb is that algorithmic targeting works exceptionally well for broad audiences over 200,000 people (e.g., developers or restaurant owners), but fails for smaller, highly specific audiences.
• List-based targeting: For niche audiences under 50,000 people (such as CIOs at Fortune 5000 companies), advertisers should bypass algorithmic targeting and upload enriched, sales-curated account lists directly to LinkedIn and Meta.
• Metadata for cross-platform matching: For tight enterprise targets (e.g., 10,000 accounts), Kevin's agency uses a platform called Metadata to match and target LinkedIn-style professional audiences on Meta.

The Reality of AI in Reporting and Analysis

Kevin warns against over-relying on AI for reporting, drawing a sharp line between data analysis and data storage.

• Analysis vs. Source of Record: Kevin's take is that AI is excellent at analyzing clean data but terrible at acting as the primary source of record.
• The QA bottleneck: If an AI agent is given incorrect data, it will quickly send marketers down a costly rabbit hole; therefore, setting up AI reporting requires significant manual QA at the start.
• The shared Claude instance: For their Reddit account, Right Percent has a team of six people working inside a single Claude instance, feeding it clean data from four separate APIs (including HubSpot, Salesforce, and Google Analytics) to run ad-hoc reports.
• Human-in-the-loop guardrails: Kevin does not allow clients to interact directly with Right Percent's AI agents; all AI-generated reports and agendas must be vetted by a senior team member first.
• Creative fatigue tracking: Darshan demonstrated how Factors solves the context problem by tracking creative fatigue (such as CTR decay over 7 or 15 days) against historical benchmarks and sending automated Slack alerts when an ad needs to be replaced.

Conclusion

The webinar converged on the idea that while AI is a powerful tool for execution, it cannot replace the strategic context and decision-making of a senior marketer. Kevin Lord Barry and Darshan agreed that AI is shifting the marketing landscape by removing technical bottlenecks—allowing marketers to build landing pages on Replit or edit videos on Async without waiting for developers or designers. Ultimately, the competitive advantage for B2B brands remains unchanged: creating products and offers that are genuinely useful to businesses, and using platform algorithms to put those offers in front of the right people.