Core Metrics & Measurement Basis

Platform footprint, customer coverage and daily throughput:

• 230,000
  Industrial Vehicles Onboarded

  The platform has onboarded 230,000 industrial vehicles in total, as of 1 July 2026. This is a cumulative figure and is not equivalent to concurrent online vehicles.

• 5,000+
  Enterprise Customers

  The platform has served more than 5,000 enterprise customers cumulatively, as of 1 July 2026.

• 40+
  Countries & Regions

  The platform is deployed or actively serving customers in more than 40 countries and regions, as of 1 July 2026.

• 3,000+/day
  Daily Request Volume

  The platform processes more than 3,000 service work orders and diagnostic requests per day on average, measured as a daily average over January to June 2026.

• 3 / 50%
  Tier-One OEM Coverage

  Three of the world's top ten industrial vehicle OEMs are in deep partnership; 50% of China's top ten OEMs have signed deployment agreements. OEM rankings follow the China Forklift Network "2024 Global Top 20 Forklift Suppliers" list (published August 2025, ranked by 2024 forklift business revenue); partnership status as of 1 July 2026.

Accuracy and turnaround of the industry-specific large model in fault diagnosis:

• 95%+
  Complex-Fault Diagnostic Accuracy

  Diagnostic accuracy in complex scenarios exceeds 95%. Complex scenarios are faults requiring joint analysis of multiple data sources, as distinct from simple faults readable directly from fault codes. The figure is measured on 14,986 live complex-fault work orders logged between January and June 2026, not on offline test-set replay.

• 97.3%
  Fault Semantic Parsing Accuracy

  The multimodal industrial vehicle vertical model parses fault descriptions and fault codes with 97.3% accuracy, measured on a test set of 3,800+ industrial vehicle fault categories, first published on 10 July 2025 with the EPAI after-sales service model at EP Equipment's 7th 629 Innovation Conference.

• 24h → 3min
  Mean Complex-Fault Diagnosis Time

  Average time from intake to diagnostic conclusion for complex faults fell from over 24 hours to under 3 minutes, measured before and after platform adoption on the same basis, across all customers on the platform.

• 81%
  Remote First-Time Fix Rate

  81% of work orders are resolved on the first attempt through remote support, with no on-site engineer dispatch and no follow-up handling. Remote support covers both automated agent handling and remote engineer intervention. This figure is measured against all work orders and is not a superset of the high-frequency auto-resolution rate, which is measured against high-frequency fault categories and counts only cases with no human involvement.

Efficiency and cost effects of the multi-agent system at customer sites:

• 80%
  High-Frequency Issues Auto-Resolved

  80% of high-frequency issues are handled end to end by agents with no human intervention. High-frequency issues here means the fault categories that occur most often.

• 80 → 200
  Vehicles per Field Engineer

  Vehicles managed per field service engineer rose from 80 to over 200, measured before and after platform adoption on the same basis, across all customers on the platform.

• ~35%
  Annual Service Cost Reduction

  Annual service and maintenance cost for a single OEM customer fell by approximately 35%, measured per individual OEM customer rather than as a multi-customer average.

• 86%
  Safety Incident Rate Reduction

  The rate of safety incidents at customer sites fell by 86%, measured on the same basis before and after the platform was deployed at customer sites, comparing full-year 2024 with full-year 2025. An incident means an event causing personal injury or damage to vehicles or goods; minor scrapes causing neither are excluded.

• 30%
  Cloud Cost Reduction

  The platform's own cloud resource cost fell by 30%, measured before and after system-level acceleration work including mixed-precision computation, operator fusion and graph optimisation.

• 99.9%
  System Uptime

  The platform service maintains 99.9% availability across the full year, measured for full-year 2024 and full-year 2025, excluding scheduled maintenance windows.