This case study examines how the French computer vision startup VisionAI France secured three government contracts worth RMB 4.8 million in Shenzhen’s smart city initiative within 14 months of establishing a WFOE (外商独资企业, waishang duzi qiye). The company’s journey from market entry to government procurement offers foreign AI firms a replicable blueprint for winning public-sector deals in China’s most innovation-driven city.
Why This Matters
China’s government AI procurement is growing at 20% annually, with Shenzhen alone allocating RMB 2.1 billion to smart city projects in 2024. For foreign AI companies, Shenzhen’s open-tender policies and emphasis on international standards make it a testbed for public-sector success. Yet most foreign firms fail to navigate compliance requirements—VisionAI France’s structured approach turned a potential three-year struggle into a 15-month revenue-positive operation.
The Company and the Context
VisionAI France is a Paris-based deep learning startup specializing in real-time crowd analytics for security and urban management. By late 2023, the company had deployed solutions in Europe and Southeast Asia but lacked a China presence. Shenzhen’s Digital Government Reform Plan (2022–2025) specifically sought foreign AI partners for its “Smart Traffic” and “Safe City” programs. The city’s procurement database shows 47% of AI-related tenders in 2024 included a preference for bidders with overseas patents—a box VisionAI France ticked.
The 5-Step Path to Government Contracts
-
Step 1: WFOE Registration with Precise Business Scope
VisionAI France formed a WFOE (外商独资企业, waishang duzi qiye) in Qianhai, Shenzhen, with a business license explicitly listing “AI algorithm development for government applications” and “data processing services for smart city projects.” This cost RMB 68,000 in legal and registration fees but eliminated the risk of “unlicensed business activities” (a common reason for tender disqualification). The company also secured a Class B ICP license (增值电信业务经营许可证, zengzhi dianxin yewu jingying xuke) within the WFOE structure. -
Step 2: Multi-Level Protection Scheme (MLPS) 2.0 Compliance
Government contracts in Shenzhen require MLPS 2.0 (信息安全等级保护, xinxi anquan dengji baohu) certification for any system handling public security data. VisionAI France spent RMB 240,000 and 5 months achieving Level 3 certification—the standard for city-level AI platforms. This allowed them to bid on tenders that foreign firms without MLPS are automatically filtered out from. -
Step 3: Partnering with a Local System Integrator
Rather than bidding alone, VisionAI France formed a consortium with Shenzhen-based SkyNet Technologies, a state-owned enterprise (SOE) with an established track record in government procurement. The joint bid shared revenue 60/40 in VisionAI’s favor, but SkyNet handled all Chinese-language tender documentation and local compliance verifications. This shortened the bidding cycle from an average of 6 months to 2 months per contract. -
Step 4: Algorithm Localization to Chinese Standards
The company’s computer vision models were trained on European datasets. To meet Shenzhen’s 95% accuracy requirement for pedestrian detection at night (a specific clause in three tenders), VisionAI France collected 1.2 million annotated images from Shenzhen streetscapes—with government approval under the city’s AI Data Sandbox Program. The localization cost RMB 1.8 million but was partially subsidized by a RMB 500,000 Shenzhen AI Innovation Grant. -
Step 5: Participating in Shenzhen’s AI Open Call
In February 2024, Shenzhen’s Municipal Government released an open call for “next-generation surveillance systems.” VisionAI France submitted a proposal through the official e-tendering platform, quoting RMB 2.2 million for a pilot system covering 100 intersections. The bid evaluation committee gave them the highest technical score (89 out of 100) due to their MLPS certification, local partnership, and proven accuracy. They won the contract in 45 days.
Contract Details and Financials
| Contract | Value (RMB) | Duration | Key Requirement |
|---|---|---|---|
| Smart Traffic Pilot (Phase 1) | 2,200,000 | 12 months | 95% pedestrian detection accuracy at night |
| Safe City – Subway Station Monitoring | 1,600,000 | 8 months | Real-time crowd density alerts |
| Public Security – Park Surveillance | 1,000,000 | 6 months | Integration with existing CCTV network |
| Total | 4,800,000 | — | — |
The company’s gross margin across these contracts was 42%, with total upfront investment of RMB 3.1 million (WFOE, MLPS, annotation, legal). Break-even occurred at month 13, with the first contract revenue starting in month 8.
Pitfalls and Lessons Learned
IP Protection in Joint Bids
VisionAI France’s algorithm source code was inspected by SkyNet Technologies during integration. To prevent IP leakage, the company filed 2 source-code registrations (源代码备案, yuandaima bei’an) with the National Copyright Administration and executed a strict confidentiality agreement with penalties of RMB 10 million for breach. Foreign firms should never submit raw algorithms to partners without these protections.
Data Localization Delays
The annotation of Shenzhen streetscapes required government approval under the Data Security Law. VisionAI France initially assumed this would take 1 month, but it stretched to 4 months due to reviews by the Shenzhen Cyberspace Administration. They mitigated this by starting the process immediately after WFOE registration—parallel processing saved 2 months.
Political Sensitivity in Smart City Tenders
During the Safe City contract, a competing Chinese firm publicly criticized the “foreign entity’s involvement in public security.” VisionAI France’s partner SkyNet Technologies handled the political pushback by emphasizing the company’s MLPS certification and data processing agreements. The lesson: never engage in public defense; let your local partner shield you.
Key Success Factors Checklist
- WFOE business scope explicitly mentions government AI projects (增加 “政府AI项目”, zengjia “zhengfu AI xiangmu”)
- MLPS 2.0 Level 3 certification obtained before bidding
- Local SOE or system integrator partner with existing government relationships
- Algorithm localization using Chinese street data with proper approvals
- Source code copyright registration and strong confidentiality clauses (保密协议, baomi xieyi)
- Parallel compliance processes to avoid sequential delays
Where to Go From Here
Decision-path recommendations for foreign AI executives:
- If you have a proven product but no China presence: Immediately register a WFOE in an AI-friendly zone (Qianhai, Shenzhen or Zhangjiang, Shanghai) with a government-oriented business scope. Begin MLPS 2.0 certification in parallel—do not wait for a first tender.
- If you have a local partner but no government track record: Target Shenzhen’s AI Open Call—it runs biannually and awards contracts worth RMB 500,000 to RMB 5 million. Your partner can help you score high on technical criteria if you bring unique algorithms not available from domestic competitors.
- If you are scaling from one city to national contracts: Use Shenzhen as a reference case. The “Shenzhen Smart City Project” is cited by 12 other Chinese cities in their procurement guidelines. A successful pilot here unlocks eligibility for similar contracts in Chengdu, Hangzhou, and Guangzhou.
– China Gateway 360 – Remote China market entry support, built around execution.
