A US brand is preparing its next launch while the product is still being refined in Shenzhen. Brand strategy, media planning and final approvals sit with headquarters. The working samples, engineers, packaging variants and component suppliers sit in China. Where should the video be produced?
The useful answer is neither “send everything to the global agency” nor “hire the cheapest local crew.” A better operating model separates control from physical execution: headquarters owns the rules and commercial judgment; a local node handles the work that requires access to the real product; both sides share the same brief, version data and approval states.
For brands connected to China’s supply chain, that reframes a Shenzhen production team. It is not simply a lower-cost vendor. It can be the node closest to product truth.
A global platform still needs physical nodes
On September 23, 2026, WPP opened Devon’s Point, a 34,000-square-foot production hub in East London. The company says the facility combines AI-powered virtual production, modular stages, rapid shooting and editing, creator spaces and post-production, while connecting those capabilities to WPP Open. WPP also says it plans equivalent hubs across every continent where WPP Production operates. The official announcement and Reuters’ independent report both confirm the facility and the rollout plan.
That is evidence of WPP’s strategy, not proof that the entire industry has already reorganized. But the direction matters: software can centralize coordination without eliminating physical production. An integrated platform becomes more useful when it can call on real spaces, crews, equipment and products.
Procurement data points to the same need for a mixed structure. On September 29, 2026, the World Federation of Advertisers published research based on 69 multinational companies with a combined $147 billion in global marketing spend. It found that production uses a balanced mix of global, regional and local arrangements, while 50% of ad-hoc creative work remains managed through individual national contracts. Fixed-fee or output-based compensation was the dominant standard model for production, at 61% of respondents. The full release is available from the WFA.
The conclusion is not that local is always better. It is that different kinds of work should not automatically be forced into one location or one commercial relationship.
The ONCE Core-and-Node Production Map
Before choosing a supplier, split the assignment by decision rights.
| Layer | Owns | Should not decide alone |
|---|---|---|
| Global control layer | Positioning, master claims, visual system, budget priorities, market goals, final approval and performance learning | Product details without access to the sample and engineering facts |
| Local production node | Sample verification, sets, talent and hands, product actions, lighting, capture, on-set changes, pickups and footage organization | Changes to market claims, compliance language or brand promises |
| Shared operating layer | Executable brief, shot IDs, SKU and packaging version, rights window, approval state, file delivery and change history | Treating “the file exists” as a substitute for clear ownership |
This is not extra bureaucracy. It moves each decision to the team that can make it accurately. Headquarters should not have to infer whether a sample is production-ready. A Shenzhen crew should not have to guess whether a claim is usable in the United States. The working contract between them is an executable video brief that connects each shot to the correct product version and approval condition.
Route the work with P-C-L
ONCE uses three variables to discuss where a content task belongs. P-C-L is a planning framework, not an industry standard.
- P — Physical dependency. Does the shot depend on a real prototype, factory, engineer, material, package or complex product action?
- C — Cultural dependency. Does the idea depend on local language, talent, creator instincts, lifestyle context or market-specific compliance judgment?
- L — Learning frequency. How often must the asset change in response to media performance, inventory, product revisions or customer feedback?
| Task profile | Recommended route | Typical work |
|---|---|---|
| Low P, C and L | Centralized production | Corporate explainers built largely from approved assets |
| High P, low C | China capture node under remote brand control | Materials, mechanisms, product actions, factory and packaging content |
| Low P, high C | Target-market creator or local team | Spoken content, lifestyle scenes and culturally specific social ideas |
| High P, C and L | Global control + Shenzhen capture node + target-market creator node | Always-on launch and performance programs that need both real products and local expression |
The framework also exposes the waste in two common defaults. “Shoot everything in the US” can create repeated sample shipping, slow engineering answers and pickups after a packaging or accessory change. “Shoot everything in China” can produce technically accurate footage that lacks the cultural or regulatory input required in the destination market.
The real saving is shorter error distance
A production node near the supply chain shortens the distance between a product fact and the shot that represents it. The team can verify SKU, ports, color, accessories and packaging before rolling. An unexpected product action can be checked with an engineer while the set is still lit. A revised component can be captured again with the same lens, lighting and motion path.
Those conditions can reduce sample movement, waiting, version mistakes and full reshoots. They do not support a universal savings percentage. Product complexity, sample value, talent, location scale and revision behavior will change the economics. Compare the total system instead:
Effective production cost = production fee + sample logistics + cross-time-zone waiting + version error + pickup risk + preparation for reuse
A cheap shoot that returns only one finished film may make the next campaign start from zero. A stronger node organizes footage for future aspect ratios, alternate hooks, product-detail edits and localization. That is how physical capture feeds an agent-ready footage library instead of becoming a folder of project leftovers.
Ask for a routing sheet before a day rate
Test the model with one SKU. List 10 to 15 real content tasks, score each one high or low on P, C and L, and assign decision rights to headquarters, the China node and the target-market team. Then ask suppliers to separate fixed preparation, one-time set and lighting work, repeatable product moves, version count, remote review, footage organization and pickup conditions.
If a task has high physical dependency, sending the product thousands of miles away does not become safer merely because a larger agency handles it. If a task has high cultural dependency, asking the team nearest the factory to own the entire expression does not become effective merely because execution is faster.
A mature global production model centralizes rules and learning while distributing physical and cultural judgment to the right nodes. For overseas brands with a China supply chain, ONCE’s Shenzhen team can own one precise part of that system: turning real products, engineering access and repeatable capture into high-value video assets that headquarters can review remotely and use over time.
