Scaling AI Content Production for a $50M Wellness Platform
How we cut voice-story production from 2 days to 15 minutes and scaled output to 200 stories/week
Executive Summary
Per-story production time, down from 1.5-2 days
Stories per week, up from an unreliable 100/week target
API cost per story, down from $50-$100 in specialist cost
From kickoff to a hardened production system
The Challenge
A $50M+ consumer health and wellness content platform needed to scale its voice-story production workflow without adding more writers, voice-over artists, designers, or engineering resources. The content team was targeting roughly 100 published voice stories per week, but the existing workflow could not reliably support that volume. Each story moved through a specialist-heavy process: a Product Manager researched the idea and wrote the brief, AI tools were used manually to create a rough outline, a writer created the final script over 2-3 hours, a voice-over artist recorded narration over 3-4 hours, a designer created banner artwork over 2-3 hours, and the team then managed revision loops until approval. Each story took approximately 1.5-2 days end-to-end with 4-5 specialists involved. The bottleneck was not just individual task time — it was the coordination overhead across multiple specialists, tools, and review loops, at an estimated cost of $50-$100 per story in India-based specialist production.
Key Pain Points
- ✕Each story required 4-5 specialists and 1.5-2 days end-to-end
- ✕Could not consistently hit the 100 stories/week target
- ✕5+ manual handoffs with revision loops per story
- ✕No in-house engineering team to build or operate automation
Our Solution
Amasa built an end-to-end AI content production engine that automated the core production pipeline while keeping human approval in place. A Product Manager submits an approximately 200-word brief and desired story duration through a custom web app. From there, the system generates the script using the xAI API, banner artwork using the Gemini API, and voice narration using the ElevenLabs API. A custom LLM evaluation framework acts as a quality gate before anything reaches a human, production tracking flows into Google Sheets, and a custom approval and publishing workflow closes the loop. Amasa delivered the first production story within approximately 3 weeks, then spent another 2 weeks hardening guardrails and edge cases — a fully hardened production system in roughly 5 weeks. The client, who has no in-house engineering team, now operates a production AI workflow through a simple web app, and Amasa continues to run and maintain the system in production.
Implementation Approach
Workflow Mapping & Design
Mapped the brief-to-publish pipeline and defined where AI generation and human approval belong
Pipeline Build
Connected xAI script generation, Gemini banner artwork, and ElevenLabs narration behind a custom web app
Quality Gating & Launch
Built a custom LLM evaluation framework as a quality gate and shipped the first production story in ~3 weeks
Hardening & Ongoing Operations
Spent 2 weeks hardening guardrails and edge cases; Amasa continues to run and maintain the system in production
The Results
The implementation delivered transformative results across all key metrics, with immediate impact on operational efficiency, accuracy, and customer satisfaction.
Impact Metrics
| Metric | Before | After | Improvement |
|---|---|---|---|
| End-to-end production time | 1.5-2 days per story | 10-15 minutes per story | ~99% faster |
| Weekly volume | Below 100 stories/week | 150-200 stories/week | 1.5-2x capacity |
| Team involvement | 4-5 specialists per story | 1 PM plus approval | Specialist-free pipeline |
| Cost per story | $50-$100 (specialist production) | $2-$5 in API cost | ~95% reduction |
Key Takeaways
- An AI production pipeline with a human approval gate can replace multi-specialist workflows without sacrificing quality control
- The biggest gains came from eliminating coordination overhead, not just individual task time
- A first production story shipped in 3 weeks; guardrail hardening made it production-grade in 5
- Companies with no engineering team can operate AI workflows through a simple web app when the system is run as a managed service
Inside the Engagement
The Workflow Before
Every voice story moved through a specialist-heavy relay before it could ship:
- The Product Manager researched the story idea and wrote the brief.
- AI tools were used manually to create a rough outline.
- A writer created the final script, typically taking 2–3 hours.
- A voice-over artist recorded narration, typically taking 3–4 hours.
- A designer created banner artwork, typically taking 2–3 hours.
- The team reviewed the assets and managed revision loops until approval.
The bottleneck was not just individual task time. It was the coordination overhead across multiple specialists, tools, and review loops – five or more handoffs per story, every story.
The New Pipeline
With the AI content production engine, a Product Manager submits an approximately 200-word brief and a desired story duration through a custom web app. From there, the system handles everything up to human approval:
- Script generation using the xAI API
- Banner image generation using the Gemini API
- Voice narration using the ElevenLabs API
- Quality gating through a custom LLM evaluation framework
- Production tracking through Google Sheets
- Approval and publishing workflow through the custom web app
Full Results
| Metric | Before | After |
|---|---|---|
| End-to-end production time | 1.5–2 days per story | 10–15 minutes per story |
| Script generation | 2–3 hours | 2–3 minutes |
| Image creation | 2–3 hours | 3–4 minutes |
| Voice production | 3–4 hours | 5–7 minutes |
| Manual handoffs | 5+ handoffs with revision loops | Brief in, approve, publish |
| Weekly volume | Could not consistently hit 100 stories/week | 150–200 stories/week |
| Team involvement | 4–5 specialists per story | 1 PM plus approval |
| Estimated cost per story | $50–$100 in specialist production cost | $2–$5 in API cost plus PM review |
The Impact
The new workflow helped the client scale from an inconsistent path to 100 stories per week to a production capacity of 150–200 stories per week – without expanding the specialist team. Dedicated writer, voice-over, and design involvement was no longer required for every story. External specialist spend was reduced, and internal team members were reassigned to higher-value work.
The system also gave a company with no in-house engineering team the ability to operate a production AI workflow through a web app. Amasa delivered the first production story within approximately 3 weeks, spent another 2 weeks hardening guardrails and edge cases, and continues to run and maintain the system in production.
The Takeaway
Amasa helped a $50M+ consumer health and wellness content platform turn a slow, specialist-heavy production workflow into a scalable AI content engine – reducing story production from 1.5–2 days to 10–15 minutes while increasing weekly output to 150–200 stories.
Quick Facts
Industry
Digital Media
Solution Type
Generative AI
Published
August 24, 2026
Technologies Used
Related Resources
Ready to Achieve Similar Results?
Let's discuss how we can transform your business with AI solutions tailored to your specific needs