Webinar Recap: Build Full Campaigns With AI

Most marketing teams still hand a campaign off between an agency, an editor, and a sound person before it's finished. Every handoff adds a week and a review cycle.

Modern teams are already leveraging AI in some parts of their production process, but those with the strongest efficiency gains and the most dramatically improved output use it throughout their entire process, building scalable, AI-powered systems rather than stacking point solutions.

In our latest live workshop, Luke Harries (Head of Growth, ElevenLabs) and Aneri Amin (Product and Growth, ElevenCreative) showed what that looks like in practice: building a full campaign live, from a product photo to a finished vertical video, using AI voice, music, image, and video models in one workspace. Key takeaways below followed by the full recap and video clips.

Key Takeaways

• Run AI through the whole production loop, not just one step of it. Teams producing hundreds or thousands of creatives a week run ideation, production, distribution, and analysis through one connected system, without trading files between tools and people at every step.
• Localization is a measurable growth lever. Dubbing that preserves tone and timing, rather than a translated caption, is what took ElevenLabs' own paid ads past a plateau, and the same mechanism drives lower cost per lead and cost per click on localized B2B ads generally.
• Hire for prompting and taste before you hire a strategist. The highest-leverage new role on a creative team is someone who can direct an AI agent toward a specific outcome. Let the agent handle model selection; hire humans for judgment.

Why creative production is under pressure

Demand for creative volume has exploded. Performance marketing increasingly runs on how many variants a team can produce, whether that's different hooks, different ICPs, or different languages, and the gap between leading and lagging teams is stark: some are producing hundreds or even thousands of creatives a week, while others are still doing one or two a month.

Localized B2B ads see a meaningful drop in cost per lead and cost per click, plus higher purchase intent. The mechanism matters: dubbing that conditions on the original performance, rather than retranslating a transcript from scratch, preserves tone and timing across languages. The result still sounds like the brand, not a translated caption read aloud.

There's also a legal unlock underneath the technical one. Music built in partnership with artists, labels, and publishers, rather than scraped audio, is what gets legal and compliance to sign off on AI-generated music and voice in ads at all. For a lot of brand teams, that approval is the actual blocker, not model quality.

And model quality itself has moved fast. Video models crossed a quality bar late last year that made ElevenLabs comfortable putting AI-generated video into its own paid campaigns, and image models are still moving: GPT Image 2.5, which shipped the day before this session, is already live inside ElevenCreative.

Demo: Building a campaign live in Flows

Scenario: Turning a single product photo into a finished, on-brand vertical ad usually means briefing an agency or freelancer, waiting on a script and voiceover, then a separate round for music and edit: days of back-and-forth before anything is ready to run. We compressed that into one live prompt, starting with two product photos and a brand color. Our goal was a finished hero shot, script, voiceover, and music, without touching a timeline editor.

What was shown:

• Opening Flows, ElevenLabs' node-based, agent-driven editor, and uploading the two product photos as the starting nodes.
• Writing one abstract prompt describing the ad (clean hero shot, brand color applied, a short script) and naming a specific model only to keep the live demo predictable; normally the Flows agent picks the right model on its own.
• The agent generating the hero image, applying the brand color, then writing and generating a voiceover and a piece of music, all from that single prompt.
• A composition node layering the generated clips, voiceover, and music into one finished piece, the same node teams use to combine multiple clips, avatars, or sound effects into a single ad.
• The finished cut: a slow push into the product against the plum background, timed to an ASMR-style voiceover landing on the beat of the music, generated end to end in a few minutes without the session ever pausing.

Why it matters: A year ago, getting this result required knowing which specific model to use for each asset and how to prompt each one well. The Flows agent has that model selection and prompting judgment built in, so the marketer directs the outcome ("make this feel like X") rather than operating each model by hand. 

The new creative team: from idea to analysis

This is the same loop ElevenLabs runs internally for our own paid campaigns:

• Ideate. A creative strategist looks at performance data and competitor activity and comes up with concepts.
• Produce. An AI creative producer, a new role focused on prompting and taste rather than manual editing, turns concepts into finished assets. ElevenLabs now has about 15 of these in house, and the job description is public if teams want a template for hiring one.
• Distribute. A media buyer pushes the assets live across Meta, LinkedIn, and other channels.
• Analyze. The strategist, producer, and media buyer meet weekly to review results and plan the next week's content.

The sequencing advice for teams starting from scratch: hire the AI creative producer first. Once one or two are consistently producing strong content, add a dedicated strategist on top.

On whether ad platforms penalize AI-made creative: the presenter noted that platforms can tell which creative is AI-generated, on both organic and paid, but ElevenLabs runs its own paid campaigns with a mix of AI-made and traditionally produced UGC and sees no statistical difference in performance between the two. The variable that matters is whether real brand taste and a real hook went into the prompt.

Two teams already running this: Ramp and Clay

Ramp uses ElevenCreative to pre-produce voiceovers and script timing before a live shoot. When they brought in an NFL player for a shoot, the team had already worked out exactly what he should say and how long each beat should run, so the live shoot day was tighter and cheaper.

Clay uses the platform primarily for localization: producing an ad in English, then scaling it into French, German, and Spanish to support international expansion, without rebuilding each asset from scratch per market.

Both are the same underlying move from different angles: reuse the workflow instead of rebuilding it for the next market, the next athlete, the next campaign.

Best practices for building campaigns with AI

1. Start with audio when the message is the hard part. ElevenLabs' own team typically nails the script and voiceover timing first, then builds the visual storyboard around it.
2. Hire for taste and prompting, not manual production. The AI creative producer role is about judgment and iteration speed, and it's the highest-leverage hire before adding a dedicated strategist.
3. Localization earns its own ROI. The cost per lead and cost per click gains from localized ads, plus ElevenLabs' own plateau-then-breakthrough story, both point to dubbing and translation as underused production capacity, not just a market-entry requirement.
4. Let the agent handle model selection. Naming a specific model is useful for predictability in a live demo; day to day, describing the outcome you want and letting the Flows agent route to the right model produces better results with less prompting expertise required.
5. Care in the prompt is what determines performance. Real brand thinking behind the prompt, the hook, the taste, the direction, is what separates a strong AI-made ad from slop, regardless of which model made it.

What's next: The Creative Engine

We discussed a next step ElevenLabs calls Creative Engine: connecting the full loop above (ideation, production, publishing, and performance analysis) into one system that runs continuously rather than one campaign at a time, with automatic localization and variant generation for every piece of creative.

The case for it is ElevenLabs' own numbers. The team's English-speaking paid ads had plateaued around $20M a year; Tim, who leads paid ads at ElevenLabs and previously ran paid at Shopify, pushed localizing existing creative into new languages rather than only making net-new creative, and that's what moved the number: a 17% increase in conversions, millions of dollars in incremental revenue, and a Google Impact Award. 

ElevenLabs is running Creative Engine with a small group of design partners now, in two configurations: fully run in parallel to a customer's existing creative team, or handed to the customer's own creative team to operate through Flows, with briefs coming in and finished assets going straight to the ad platform.

Watch the full session

Watch the full workshop here, including the live build and audience Q&A.