// CASE 06 · TRANSLATED · MATESUB
Before the AI writes the subtitle, watch who writes it today.
Translated serves 180,000 clients in 177 languages. Matesub was going to automate the slowest parts of subtitling and dubbing with AI. Before designing a single screen, we went to see how subtitlers in more than ten countries actually work.
180K
clients Translated serves
177
languages in its catalogue
10+
countries of the professionals we interviewed and shadowed
1
web-based subtitling tool informed by the research
01
Executive summary
AI could cut the time and cost of multilingual subtitles, which means more video content available in more languages. But subtitling is a craft with its own rhythm, tools and pain, and a tool that ignores it gets ignored. We mapped the industry as a system, followed professionals through their real workflow, interviewed them, and turned what we saw into a set of recommendations, journeys and UX prototypes that shaped the first web-based version of Matesub.
02
Four phases
➔ System map of the subtitling industry: actors, tools, handoffs
➔ User interviews and shadowing with subtitlers in more than ten countries
➔ Customer journey mapping of the as-is workflow, with the pain points
➔ User flows and wireframes for the prototype of the web tool
Tools & tactics
System mapping
Remote interviews and shadowing
Journey mapping
Wireframing and prototyping
03
Issues discovered
➔ The workflow was split across several tools, with manual handoffs between them.
➔ Timing and segmentation ate most of the hours and none of the skill.
➔ Quality checks lived in people's heads, not in the tool.
➔ Multilingual jobs multiplied the same manual steps per language.
➔ Professionals distrusted automation that hid what it had changed.
Four principles for AI tools
Automate the repetitive, keep the call
Show what the machine changed
One workspace, no handoffs
Design for the professional first
04
Results
Translated got a full research report, journey maps and UX prototypes, and used them to inform the development of its web-based subtitling tool; Matesub then shipped as a product. What I cannot claim: I do not have adoption numbers, and the product evolved past our prototypes. What stays: this was 2020, and the question of how people trust an AI that rewrites their work was already the design problem. It is the same question I now put to every AI feature, whether it writes a subtitle, a listing or a page.
05
Lessons learned
➔ Research in more than one country prevents designing for the one workflow you know
➔ Professionals accept automation when it shows its work
➔ A system map finds the handoff nobody had noticed
➔ AI products are still products: the craft around them is the design