Recruitment only counts when the data qualifies
A global AI data platform required short, authentic face-motion recordings from eligible participants. Each submission had to meet country, age, identity, recording and single-submission requirements. Our responsibility was a batch of 3,000 accepted videos within that wider program.
Recruitment therefore had to connect directly with qualification, recording guidance and quality control. A high response count alone would not complete the work: rejected videos had to be identified, followed up and replaced before the final batch could be accepted.
Project scope
Face-motion video data collection
- Purpose
- Face-motion data for AI training
- Contributor age
- 18 and over
- Eligibility
- Client-approved countries and platform KYC
- Submission rule
- One accepted submission per person
- Capture devices
- Computers, phones and tablets
- Delivery
- Video batch, QC and replacement records
How we approached the work
Recruit against the specification
We combined an established freelancer network with targeted freelance recruitment, using the client’s eligible-country list to screen contributors. Participants completed the platform’s KYC process, and the workflow included consent confirmation before recording.
Guide and inspect recordings
Recording guidance translated the brief into practical instructions. QC checked for a real participant, the required duration, clear facial movement, suitable lighting, an unobstructed face, appropriate capture settings and compliance with the supplied script or example.
Close the replacement loop
Rejected submissions were recorded and followed up with replacement work. Batch records connected QC decisions, replacement status and delivery lists, allowing the team to track progress toward the contracted accepted-video count.
From preparation to delivery
- RecruitEligibility and onboarding
Country screening, platform KYC and recording guidance
- CheckSubmission and quality review
Recording requirements, duplicate controls and batch tracking
- CompleteReplacement and acceptance
Resolve 187 rejected submissions and complete the accepted batch
What the client received
- 3,000 accepted face-motion videos
- Platform KYC and submission records
- QC check records
- Replacement recording records
- Batch delivery manifest
- Client-reviewed final delivery
The acceptance numbers, clearly defined
The initial submission contained 3,000 videos. Of these, 2,813 passed and 187 failed the client’s checks. First-submission acceptance was therefore 93.77%, calculated as 2,813 divided by 3,000.
The team completed 187 replacement recordings, bringing the final accepted delivery to 3,000 videos. The replacement process closed the rejected portion of the original batch; it did not turn the first submission into a 100% first-pass result.
This engagement demonstrates participant recruitment, video collection, QC and replacement coordination. It did not include speech recognition training data, multilingual transcription or measured improvements to an AI model.
- 2,813
- Accepted on first submission
- 187
- Rejected and replaced
- 3,000
- Final accepted videos
First-submission acceptance: 2,813 ÷ 3,000 = 93.77%. Final accepted delivery includes the completed replacement recordings.
Project questions
Does 3,000 refer to submissions or accepted videos?
The initial batch contained 3,000 submissions. After 187 rejected videos were replaced, the final client-accepted batch also contained 3,000 videos.
Which locations did the project allow?
The client allowed selected countries in Asia, the Americas, Europe, Oceania and Africa. This case study does not claim that our batch included every eligible country.
