Home / How we collect
How we collect
Three ways to collect
Dedicated teams, a distributed crowdsource network, or a hybrid of both. Pick the fit for your program. You own every byte.
3
Delivery models
35–30,000
Operator pod size range
100%
Your IP, always
Pick the fit
Three delivery models. One pipeline.
Each model is detailed on its own page. The right pick depends on IP control, ramp speed, and volume.
MODEL 01
Dedicated teams
Trained, vetted operators in your studios or ours. Maximum IP control.
- Full exclusivity. Your spec.
- 30-operator pod, default.
- Surgical, AV, defense, industrial.
- 4 weeks SOW to production.
Per FTE-month
MODEL 02
Crowdsource
Distributed operator network for scene diversity and fast scale-up.
- 8,000+ vetted operators.
- 12 countries, 28 languages.
- Long-tail, geographic spread, volume bursts.
- 3 days spec to first units.
Per task or per hour
MODEL 03
Hybrid
Dedicated core for the spine. Crowdsourced edge for diversity.
- 80% spine + 20% edge, by default.
- Unified QA across both layers.
- Production foundation models.
- 14 days SOW to first batch.
Custom SOW
How to choose
A simple decision matrix
If two columns describe your situation, hybrid is usually the answer.
Pick dedicated if…
- IP exclusivity is non-negotiable.
- You need operators trained to your spec.
- Your work is surgical, AV, or defense-grade.
- Cost per unit matters less than consistency.
Pick crowdsource if…
- You need diversity, geographies, or volume.
- The constraint is operator hours, not bar.
- You’re in fine-tune or benchmark mode.
- Speed-to-first-batch is the priority.
Pick hybrid if…
- You’re training a foundation model.
- You want a dedicated spine + scaled edge.
- You need one SOW, one dataset, one dashboard.
- Production reliability matters most.
What our partners say
The choice wasn’t dedicated vs crowd. It was which mix of the two. Roborax helped us run the math and shipped a hybrid program in two weeks.
Theo Hartmann
Head of Robotics Data, Northpoint AI
FAQ
Questions about how we deliver programs
Dedicated teams suit programs that need consistency, IP protection, and operational depth. Crowdsource suits programs that need volume and speed at lower cost. Hybrid suits programs that need both. We help you choose in the scoping call.
Yes. Many programs start with a dedicated pilot, then transition to hybrid for scale. We design programs with that flexibility in mind.
Dedicated teams produce higher consistency because operators work exclusively on your task and build expertise over time. Crowdsourced data has more natural variation, which can be an advantage for generalisation but requires tighter QA.
Every operator — dedicated or crowd — passes a task-specific quality gate before contributing to a live program. No unvetted operator data enters a client delivery.
Not sure which model fits?
Tell us the program and we’ll scope all three approaches with cost and ramp implications.