Generative AI in 2025: Moving Beyond PPT to a Productivity Revolution
Pick up your phone now, open an e-commerce site, and search for “charger.” The first brand you’ll likely see is Anker Innovations. If you want to compare prices or ask about specs, you’ll click on customer service. Little do you know, these simple actions involve multiple AI large model capabilities.
At the 2025 Amazon Web Services (AWS) China Summit, Gong Yin, Chief Innovation Officer of Anker Innovations, shared how the company uses AI to innovate smart products and improve operational efficiency with AWS technology. Anker has built:
- A real-time knowledge base powered by large language models with 50+ AI agents
- Vela, a multimodal AIGC content production platform
- A smart advertising system integrated with Amazon SageMaker, covering over 90% of in-platform ads
- A product development system using deep learning and large models
What’s revolutionary? Not the tech itself, but its practical application.
- Vela has generated over 1.2 million images
- AI resolves over 70% of customer service tickets
- Over 20% of in-platform ads are fully managed by AI
- The AIME platform hosts 300+ active AI agents with millions of usage instances
This is not an isolated case. The summit showcased numerous generative AI applications across industries, including TCL, WPS, Lalamove, Kingdee, Hehe Information, and Fosun Pharma.
“More and more enterprises want to embrace AI,” said Chu Ruisong, Vice President of Amazon and President of AWS Greater China.
Key Insights from 1,500 Generative AI Projects
- 82% success rate from PoC to mass production (vs. 41% industry average)
- No longer just a “concept”—generative AI is a proven productivity tool
Why Do Some AI Projects Fail?
Common pitfalls from AWS’ 350+ global experts:
- Unclear goals or wrong use cases: Some problems don’t require generative AI at all.
- Higher-than-expected costs: Poorly optimized technical 选型 (model selection) leads to waste.
- Lack of strategic commitment: Treating AI as a “fad” rather than a core innovation driver.
The AWS Approach: A Four-Step Framework for Success
- Scenario Assessment
- Evaluate seven key dimensions: team, timeline, risk, data, ROI, budget, feasibility
- Example: Fosun Pharma uses AI for medical content generation; TCL leverages global infrastructure for product innovation
- Rule of thumb: Use generative AI for creative tasks (content, interaction), not for solved problems (face recognition, OCR).
- Technology Selection
- Choose the right model, not just the “hottest” one
- 34% of enterprises use 2 models, 41% use 3 (Jefferies report)
- Case study: Base44, a 9-person startup, chose Claude via AWS over OpenAI for cost efficiency, achieving 250k users and $189k profit in 6 months.
- Mass Production Optimization
- Complex engineering challenges: data labeling, model tuning, cost-performance balance
- Example: A cultural industry group improved dialogue attribution accuracy from 70% to 90% using Amazon Bedrock and Claude 3.5, while reducing token costs.
- Performance Monitoring
- Track quality, performance, and user feedback in real time
- Avoid “money pits” with dashboards for latency, throughput, and hallucination rates.
Generative AI in 2025: Key Trends
- Market growth: China’s MaaS market surged 215.7% in 2024 (IDC)
- Multi-model strategy: 80% of Chinese enterprises will adopt hybrid models by 2027 (Gartner)
- Cost reduction: Training costs are 1/20 of 2022 levels, enabling industrial-scale applications
Challenges Ahead
- The “easy wins” are gone—deep integration requires hard work
- From hype to reality: AI must solve real business problems, not just chase trends
The Bottom Line
Like the steam engine revolutionizing industry, generative AI is reshaping how we work. While the journey is long (30 years for the internet, 20 years for cloud computing), the foundation for AI as the “new infrastructure” is being laid today.
References:
- Stanford HAI 2024 AI Index Report
- IDC China MaaS Market Report
- AWS Generative AI Innovation Center Case Studies
- The Information Global AI Adoption Survey
