Case studies
6-month ramp, 88% 12-month retention, ~48% onboard-to-production
First-party case studies from active Zipang production programs, named where the contract permits, anonymous where it does not. Each program runs through the same 5-gate screening funnel; only the rubric changes. 22 / 41 / 58 / 67 / 78% pass-through by gate, measured across the active book.
Shelf-edge computer vision annotation
100+ hypermarket retail network in France (via Transperfect Dataforce)
How Zipang built a 432-operator Indonesian annotation pod, sustained 90%+ accuracy on 3.4M monthly tasks, and held 88%+ 12-month retention through a 5-gate screening funnel.
432
Deployed
90%+
Accuracy
6 months
Ramp
208
Live
3.4M
Tasks / mo
88%+
12-mo retention
Video interpolation annotation
Transperfect Dataforce
How Zipang moved from 78% week-1 accuracy to 92.1% rolling-12-week sustained on a hard video interpolation task, using a 200-frame gold set and three-layer review (peer at frame N, supervisor at frame N+24, weekly gold set).
60
Deployed
92.1%
Accuracy
12 weeks
Ramp
20
Live
1.4M
Tasks / mo
86%+
12-mo retention
Trust & Safety content moderation
Global social platform (named customers: ByteDance/TikTok, Kuaishou, PUBG Mobile)
How Zipang ran a 180-operator moderation pod across high-risk T&S, spam/IP, UGC text/chat, creator compliance, and image/video queues, with a 96.4% precision/recall sustained number across 24 months.
180
Deployed
96.4%
Accuracy
30 days
Ramp
150
Live
2.1M
Tasks / mo
90%+
12-mo retention
Have a similar program to scope?
Most roles hit first productive shift in 7–14 days from signed SOW. 30-day replacement guarantee on fit and skill, 90-day on non-performance.
Case-study buyer questions
Common questions buyers ask before requesting a full operational case. Each one links to the source article or to /methodology.
Which case study should I read first?
If you are evaluating annotation accuracy at scale, start with the French retail AI case study (432 deployed, 90%+ sustained on 3.4M tasks/month). If you are evaluating ramp speed on a hard task, read the Transperfect Dataforce case study (78% to 92.1% in 12 weeks). If you are evaluating trust & safety / content moderation work, read the global social platform case study (180 operators across 5 queue types, 96.4% sustained).
Are the case studies anonymized?
Named where the contract permits (Transperfect Dataforce, ByteDance/TikTok, Kuaishou, PUBG Mobile), anonymized where it does not (the French retail client is referenced by the deployment descriptor '100+ hypermarket retail network in France'). Every program reference links to the underlying research note or the operative case study child page.
How do I get the full operational methodology for a program?
Each case study links to its source articles and to /methodology. For a scoping conversation, register at /employers#register with the program name; the Zipang team responds within 2 business days with the full operational case (rubric, gold-set cycle, supervisor ratio, ramp plan).
What is the replacement guarantee on a 200+ operator program?
30-day replacement on performance (sustained gold-set threshold miss), 90-day on non-performance (throughput, QA flags, shift-coverage failure). Replacement is the vendor's responsibility, not the client's. Replacement operators are pulled from the active shortlist pool fed by the same 5-gate funnel.
Curated by Yoseph Gratika, Founder & CEO (PT Lima Cakar Bumi). Last updated 2026-06-15. Case study methodology and rubric per the 5-gate screening funnel.