Before you shortlist: define your collection requirements
Most outsourcing engagements fail at requirements definition, not vendor selection. Before approaching any Vietnam vendor, your team needs a written requirements document covering the essentials: the modality (egocentric, fixed-camera, multi-sensor), the scenario specification (what tasks, what environments, what participant demographics), the annotation requirements (what labels are needed and to what standard), the delivery format (raw plus schema, or annotated), and the compliance requirements (GDPR, HIPAA, internal data governance).
A vendor who cannot scope against these inputs is not a managed program vendor - they are an annotation platform offering collection as an add-on. The distinction matters. Annotation platforms handle labeling at scale. Managed program vendors design and execute the physical collection process: hardware selection, participant recruitment, scenario scripting, QA, and legal compliance. You need the second type for a production-grade collection program.
If your requirements are not fully defined, the shortlisting process itself will surface gaps. But entering vendor conversations without any scenario specification wastes both sides' time and makes cost comparison meaningless - vendors will quote against different assumed scopes.
Step 1: Build a vendor shortlist for Vietnam
Vietnam has a large annotation vendor market but a small managed collection program market. The annotation market is well-documented and competitive. The managed program market - vendors who own hardware, recruit participants, design capture protocols, and deliver production-grade datasets - is substantially smaller. Conflating the two is the most common procurement mistake for teams new to Vietnam sourcing.
Identify vendors who own collection hardware (ask directly - the answer distinguishes managed program vendors from brokers), have participant recruitment infrastructure, and can provide a sample capture protocol document from a previous program. Sources for identifying candidates: direct referrals from AI and ML teams at APAC enterprises, industry analyst reports covering APAC data services, and LinkedIn search for "video data collection" plus "Vietnam" filtering for companies rather than individuals.
- Does the vendor own collection hardware or contract it to local partners?
- Can they provide a written capture protocol from a comparable previous program?
- Do they have a participant recruitment database or rely on open crowd platforms?
- Can they provide anonymized data samples from a previous program in your modality?
- Do they have a legal entity capable of executing a DPA and SCCs?
Step 2: Run the vendor technical evaluation
The core evaluation question is: has this vendor run a collection program operationally similar to yours? Generic capability claims do not answer this question. Ask each shortlisted vendor to walk you through their last three collection programs in operational detail - hardware configuration, participant recruitment approach, QA failure rates and handling, delivery timeline vs. contractual commitment. Vendors with genuine managed program experience answer these questions with specificity. Vendors overstating capability default to annotation case studies and cannot answer technical follow-up questions about collection operations.
Request a sample dataset from a comparable previous program. Review for: scene coverage consistency across the sample, sensor sync quality if the program was multi-modal, metadata completeness against their stated schema, and scenario diversity across the sample. A dataset drawn from only one environment or a single participant demographic is not from a well-designed program - it suggests the vendor ran a single-location shoot and is presenting it as a representative collection program.
Score vendors on two dimensions separately: operational capability (hardware, recruitment, QA) and compliance readiness (legal entity, DPA experience, de-identification pipeline). A vendor strong on operations but weak on compliance is manageable with the right legal structuring if they are willing to work through it. A vendor who dismisses compliance questions is a harder problem to solve.
Step 3: Scope and run a paid pilot
Never commit to production volume without a paid pilot at production-equivalent standards. The pilot must use your required hardware configuration, QA standards, and delivery format - not a simplified proxy. Vendors who cannot run a pilot at production standards cannot run a production program. Standard pilot size: 50-100 hours of collected footage. Pilot timeline: 2-4 weeks from scope sign-off to delivery. Budget the pilot as a percentage of the expected total program cost, not as a discount engagement - it is operational validation, not a sales demo.
The pilot reveals: actual QA throughput versus vendor claims, annotation or metadata gaps in the delivery format, coordination overhead for your internal program manager, and whether the vendor's capture protocol covers your scenario space adequately. QA throughput discrepancies between pilot and vendor claims are the most common finding - and the most important one to discover before committing production budget.
Step 4: Structure the compliance framework
For EU and US buyers, compliance structuring precedes production commitment. Required elements: a Data Processing Agreement covering GDPR Article 28 obligations (even if the vendor is not in the EU, the buyer's obligations require downstream coverage), Standard Contractual Clauses for Vietnam-to-buyer data transfer, consent form templates reviewed for PDPD and GDPR dual compliance, and documented data deletion procedures covering both in-program handling and post-delivery destruction.
This is standard commercial legal work - a Vietnam vendor experienced with EU and US buyers will have DPA templates ready and can name their legal basis for cross-border transfer without prompting. If a vendor has no existing DPA template or cannot explain their legal basis for the specific scenarios in your program, remove them from the shortlist. The compliance gap is not a negotiating position - it is an operational gap that will surface at the worst time if not resolved before production begins.
Step 5: Manage the production program
Production programs require structured weekly check-ins covering: recording sessions completed versus plan, QA pass/fail rates by scenario type, participant utilization versus quota, and delivery batch status. Establish a single point of contact on both sides before production begins - your program manager and the vendor's program lead. Define escalation paths for QA failures before production begins, not when the first failure occurs.
Require weekly delivery batches rather than end-of-program batch delivery. Weekly batches allow early detection of coverage gaps, systematic QA failures, or scenario drift before the full budget is committed. End-of-program delivery is the vendor-preferred structure because it hides problems until after the contract value is spent. Weekly delivery is the buyer-preferred structure because it preserves the ability to course-correct.
Step 6: Annotation handoff and dataset delivery
If your program includes annotation, confirm the annotation team has reviewed the collection protocol before recording begins - not after. Annotation teams who understand the capture protocol can design label hierarchies that match the footage structure, flag ambiguous scenario coverage before it becomes a labeling problem, and apply consistent judgments across edge cases that the protocol does not explicitly address. Annotation teams handed footage cold will make different interpretive choices at each ambiguity point.
Delivery format should be specified in the contract with no ambiguity: raw video plus metadata schema, annotated in your required tooling format, or both. Confirm storage handoff method and access controls: S3 bucket with defined IAM policy, GCP Cloud Storage with explicit access expiry, or vendor-managed SFTP with defined access window. Include a contractual data deletion confirmation requirement - a written statement from the vendor confirming deletion of all source footage from their systems after successful delivery verification.
What a successful outsourcing engagement looks like
A successful outsourced video data collection engagement delivers: a production-grade dataset covering your scenario specification with documented QA verification, compliance documentation sufficient for your legal team to confirm DPA obligations are met, and a vendor relationship with operational knowledge of your program that can be reengaged for the next collection cycle without restarting the onboarding process. The cost outcome - 40-60% below US or EU vendor pricing for equivalent quality - follows from the operational outcome, not the other way around.
The pilot is the critical investment in the process. It tells you whether the full program is worth running with that vendor, whether your scenario specification is buildable as designed, and whether the QA and delivery process meets your standards before you are committed to production volume. A pilot that reveals problems is a good outcome - it is far less expensive than discovering the same problems at production scale.
