Business Impact

From brief to script: How Disker and Marvik built an AI creative assistant with Anthropic Claude

Share

How Disker brought Claude-powered scriptwriting into production, increased script approval rates to 96.2%, reduced AI operating costs, and accelerated engineering development with Claude Code.

The challenge: Scaling creator content without losing authenticity

Influencer marketing depends on a delicate balance: brands need creators to communicate specific messages, follow campaign guidelines, and represent products accurately, while creators need the freedom to produce content that feels natural to their audiences.

As campaigns grow to involve hundreds or thousands of creators, managing this process becomes increasingly complex.

Each creator needs to understand the brand's brief, translate it into a compelling video concept, and develop a script that reflects both the campaign requirements and their own creative style. Brands then need to review submissions, identify missing requirements, and request revisions when necessary.

For Disker, an influencer-marketing platform that helps brands and agencies manage creator campaigns, this presented an opportunity to transform the creative workflow.

The goal was not simply to automate scriptwriting. It was to build an AI assistant that could understand detailed campaign requirements, adapt to each creator's voice, support an interactive creative process, and help deliver content that brands could approve with fewer revisions.

The solution: Creta, a Claude-powered creative assistant

Disker partnered with Marvik to develop Creta, an AI scriptwriting assistant integrated directly into Disker's influencer-marketing platform.

Powered by Anthropic Claude through the Claude API, Creta helps creators transform brand briefs into personalized video scripts through a conversational experience.

By combining campaign requirements with creator-specific context, including previous video transcriptions, account insights, and reference content, Creta helps creators develop scripts that reflect their individual style while remaining aligned with brand objectives.

Two creative experiences, one intelligence layer

Creta supports two modes, selected according to the campaign's needs.

Classic mode: Ready-to-record scripts

Creta uses Claude to generate complete video scripts, including the dialogue the creator will deliver. Creators can collaborate with the assistant conversationally, request adjustments, and refine scripts before submitting them for review.

Suggestor mode: Guided creative freedom

Instead of writing the dialogue, Creta provides creative directions, content ideas, and structural suggestions informed by the campaign brief and the creator's previous content.

This allows creators to retain greater control over their exact wording while benefiting from AI-generated guidance.

Both modes incorporate mandatory campaign requirements, balancing creative flexibility with brand compliance.

Why Anthropic Claude?

Disker and Marvik evaluated leading AI models before selecting Claude Sonnet 5.5 as the production model for Creta's core conversational and scriptwriting capabilities.

Claude stood out for its ability to follow complex instructions in detailed campaign briefs, maintain context and consistency across multiple conversation turns, and produce natural, high-quality creative writing in Spanish.

These capabilities were critical to building an assistant that could balance brand requirements with each creator's individual style, making Claude a strong fit for Creta's production environment.

Building and deploying Creta

Marvik worked alongside Disker's engineering team to design, develop, integrate, and deploy Creta within Disker's existing platform.

The implementation focused on three areas.

Engineering the contextual data pipeline

Disker and Marvik built the workflows that process campaign briefs and supporting documents, extracting mandatory messages, product information, creative guidelines, and restrictions into structured requirements.

The implementation also incorporates creator-specific information, including previous video transcriptions, account insights, and reference content, providing Claude with the context needed to personalize its responses.

Building the conversational experience and revision workflows

The teams developed and integrated the Claude-powered conversational workflows that allow creators to generate, refine, and save versioned scripts directly within Disker's platform.

The implementation supports real-time streaming responses and iterative conversations. When a brand rejects a script, the assistant can incorporate the feedback into a revised version, simplifying the review and revision process.

Validation, monitoring, and production reliability

Disker and Marvik implemented automated validation to evaluate generated scripts against structured campaign requirements before automatic approval.

The system uses independent AI models for validation, alongside Langfuse for prompt management, tracing, and cost monitoring.

Automated quality evaluations and a human feedback process support continuous improvement, helping Disker maintain reliability as the platform evolves.

Measurable business impact

Creta's production deployment has improved script approval rates and reduced AI operating costs while maintaining a fast creative workflow.

Higher script approval rates without slowing down creation

Within the first few weeks of Creta's production deployment, script approval rates increased from 85.2% to 96.2%, an 11 percentage point improvement.

With other iterations of the solution, creators typically reached a script ready for review within approximately 5–8 minutes. By combining Claude-powered script generation with personalized creator context, structured campaign requirements, and automated validation, Creta enables significantly more brand-aligned scripts without extending the creative process.

Approximately 27% lower cost per conversation turn

By implementing Anthropic's prompt caching, Disker reduced the average cost per conversation turn from approximately $0.037 to $0.027, a reduction of 27%.

Prompt caching allows Creta to reuse previously processed campaign instructions and conversation context across interactions, reducing the cost of running Claude without compromising the creator experience.

With creators typically requiring two to three conversation turns per script, this optimization improves the cost efficiency of the platform as usage scales.

Accelerating engineering with Claude Code

Beyond the customer-facing application, Disker and Marvik used Claude Code throughout Creta's development.

Disker initially experimented with other AI development tools before standardizing on Claude Code across its engineering organization.

For a team of approximately 8–10 engineers, the adoption of Claude Code has coincided with substantial improvements in development throughput and shorter code integration cycles.

Creta development

  • Monthly merged pull requests increased from approximately 8–13 to 30–58.
  • Pull requests merged per engineer more than doubled.
  • Median time to merge decreased from approximately 20 hours to 1 hour.

Across Disker's engineering organization

  • Monthly merged pull requests increased from 78 to 275, eventually reaching 493 per month.
  • 100% of engineering pull requests are now created using Claude Code.

These results illustrate how Disker has increased engineering throughput while integrating Claude Code into its development workflow.

Looking ahead: Scaling human creativity with AI

Disker's collaboration with Marvik demonstrates how Anthropic Claude can support creative workflows while preserving the individuality of the people behind them.

By bringing Creta into production, Disker has improved script approval rates, reduced AI operating costs, and created a more efficient experience for creators and brands.

At the same time, its adoption of Claude Code illustrates how a relatively small engineering team can increase development throughput and accelerate product innovation.

Together, these outcomes highlight the value of combining Anthropic's technology with specialized AI engineering expertise to build and scale production AI applications.

Customer Testimonial

"We chose Claude because it offered an integrated solution for both our development and operational teams. It also came highly recommended by developers we trust, and we found it very intuitive to use"

Antón Chalbaud — Founder & CTO, Disker

keep exploring

News, Insights & Impact

View all
View all

Every AI journey starts with a conversation