Saved €170,000 and 1,400 hours creating multilingual fitness content

This make.com automation is part of a groundbreaking fitness app project that streamlines the creation of high-quality exercise descriptions and multilingual audio guides for exercise videos. The project addresses a critical bottleneck in producing exercise content by leveraging advanced AI technologies and optimizing workflow efficiency. Here’s how it works:

• Input from Exercise Database: The automation starts with pulling bullet points from a Google Sheets table titled "Exercises," where brief descriptions of the exercises are stored (e.g., "Lie on your back; place feet hip-width apart; arms relaxed by your sides; activate lower back, abdominal, and gluteal muscles, and lift the pelvis slowly toward the ceiling").

• Description Enhancement with ChatGPT: These brief points are sent to ChatGPT, which transforms them into comprehensive exercise descriptions in a consistent and professional tone. ChatGPT also translates the detailed text into 8 languages simultaneously, generating the output in a JSON array format.

• Voice Synthesis with ElevenLabs: The automation determines which trainer performed each exercise (data fetched from the "Exercises" table) and uses the respective cloned voice (created in ElevenLabs) to synthesize the exercise text in all 8 languages. This ensures a personalized and authentic experience for the end-user.

• Efficient Data Management: All translated texts are bulk-uploaded into a "Translations" table in Google Sheets, ensuring smooth and centralized management of multilingual content.

• Iterative Audio Processing: By iterating through the JSON array, the automation creates audio tracks for each language, significantly reducing manual workload and error-prone processes.

• Output and Storage: The finalized audio files are stored in Google Drive for seamless access. While Vimeo currently lacks API support for direct audio track uploads, these files will be manually attached to the corresponding exercise videos as a temporary measure.

Optimization and Results:

The original automation consisted of nearly 60 modules but was carefully optimized to ensure high performance, reducing unnecessary complexity and improving execution time.

By using ChatGPT to translate all texts at once and iterating through the JSON array for audio synthesis, the automation maximizes efficiency without compromising quality.

Impact and Time/Cost Savings:

For a library of 1,000 exercises, the traditional process without any AI tools would require:

• 30 minutes per exercise for writing detailed descriptions, totaling 500 hours.

• 1 hour per exercise for recording, editing, and finalizing audio files, totaling 1,000 hours.

Total traditional time required: 500 hours for text creation + 1,000 hours for audio production = 1,500 hours.

With an average hourly cost of €30 for manual work, this equates to €45,000 in labor costs. Additional costs for translation services (estimated at €0.10 per word for 1,000 exercises, each averaging 150 words per language) would amount to €120,000 for 8 languages. Furthermore, professional audio equipment and studio setups could add another €5,000.

In contrast, this automation eliminates:

• The need for manual translation services and audio hardware.

• Over 90% of the manual workload, reducing the production time to just the manual Vimeo uploads.

Estimated Savings:

Time: Saves approximately 1,400 hours of manual effort.

Costs: Saves an estimated €170,000 in labor, translation, and hardware expenses.

Even if we were to skip using a professional translation service and instead rely on tools like Google Translate, it would still require 15 minutes per exercise to manually verify and refine the 8 translations. For 1,000 exercises, this adds up to 250 hours, which corresponds to an additional €7,500 in labor costs (based on €30/hour) - so we would still save €57,500€ and 206 working days - resulting in us being able to launch about 9.5 months earlier!

Thank you AI & Thank you Mindvalley, the course really helped a lot 🥳