This is a hypothetical case by SasaDaily. The process and figures are for demonstration only, and actual results vary by team and product.

Scenario: 3 People, 10 Videos a Week

A 3-person e-commerce team consists of the owner handling product selection and customer service, one editor responsible for listing products, and a part-timer assisting with shipping. Short videos are the main traffic source, with a goal of producing 10 product short videos weekly.

Previous approach: The editor managed everything — from finding product selling points, writing scripts, editing, subtitling, to voiceover — taking about 2 hours per video. At 10 videos a week, that’s 20 hours, leaving no time for other tasks.

Breaking It into Three Steps: AI Does the Heavy Lifting, Humans Make Decisions

The team split the process into three parts, with AI handling execution and humans making judgments:

Step 1: Script — AI writes 5 versions, humans pick 1

Inputting product name and selling points, AI generates 5 different styles of voiceover scripts (pain point, unboxing, scenario, critique, recommendation). Humans don’t write scripts, only select the smoothest version and do minor edits before finalizing.

Step 2: Captions — AI applies template, humans check word breaks

Captions follow a fixed template (font size, outline, position). AI formats the script into subtitles using that template. Human work is just checking for word breaks, typos, or if captions block the product in the video.

Step 3: Voiceover — AI generates Chinese voice, humans confirm tone matches visuals

AI produces the Chinese voiceover. Humans listen once to check if tone fits the on-screen character and verify there are no misread words. Only problematic segments are re-recorded.

Numbers: Production Time per Video Reduced from 2 Hours to 40 Minutes

  • Script: originally 40 minutes (writing) → 10 minutes (selecting + fine-tuning)
  • Captions: originally 40 minutes (manual timing) → 10 minutes (review)
  • Voiceover: originally 30 minutes (recording + editing) → 10 minutes (listening + confirmation)
  • Editing: 10 minutes → 10 minutes
  • Total: about 2 hours → about 40 minutes

At 10 videos per week, that cuts work from 20 hours down to about 7 hours. The editor finally has time to handle listings and customer service.

Two Decisions Still Made by Humans

Product Selection: Choosing the wrong product means even a well-edited video won’t convert sales. AI can help analyze trend data, but the choice of which product to focus on is always up to humans.

Pre-publishing checks: typos, prices, product page links, and ensuring different versions in the same batch don’t have identical thumbnails. AI cannot fully handle these, so a final human review before publishing is the most critical 5 minutes in the entire workflow.

Related Links

  • AI Business Case｜2026/09/24: How a 4-Person Short Video Studio Uses YouTube Create: AI does initial editing on material separated into three piles, with humans handling hook, branding, and publishing details
  • AI Business Case｜2026/10/11: How Asana Reduced Browser Agent Cost to $0.47 Per Use by Focusing on Context, Cache, and Screenshot details