Fei Wang
24 March 2025
As technology rapidly advances, consumer expectations are evolving just as quickly. Today, hyper-personalized, context-aware, and highly scalable customer engagement is essential for delivering tailored and coherent shopping experiences throughout the customer journey.
However, bridging the gap between customer engagement on paid social and retention channels and the on-site e-commerce experience remains challenging. Key issues contributing to this disconnect include fragmented data sources across marketing and site experiences, slow evaluation and feedback cycles, and the lack of a centralized, context-aware model to unify the customer journey.
Unique combination of my tech and retail experience
During my 12-year tenure at Amazon as a Principal Engineer and founding engineer within the Alexa organization, I focused extensively on understanding customer intent, maintaining contextual continuity, and translating customer behaviors into actionable insights. This experience provided me with a profound understanding of how innovative technology can transform e-commerce and drive significant business growth.
In 2021, I joined Saks OFF 5TH as CTO, overseeing technology for this prominent omnichannel fashion retailer. Over the next three years, I immersed myself deeply into every facet of the business—from merchandising and marketing to digital customer experiences—gaining first-hand knowledge of the unique pain points that retailers and brands face.
Spangle LPM (Large Product Model) and Agentic Framework
With recent breakthroughs in large language models (LLMs) and agentic systems significantly lowering training and inference costs, it is now feasible to combine advanced AI with domain-specific e-commerce requirements. This combination enables the creation of a truly contextual, scalable, and individualized personalization system, making every customer touchpoint enjoyable and relevant.
After six months of dedicated innovation, we're proud to introduce our proprietary Large Product Model (LPM). This groundbreaking model treats each product as a distinct token and considers user interactions with these products as sentences, predicting the next token to deliver highly personalized recommendations. Our model incorporates every critical aspect of a product—images, descriptions, categories, reviews, product structured attributes and real-time user interactions—building a large multi-modality model by leveraging existing models alongside our proprietary data, training and finetuning the model with our innovative approach to align with the needs of the e-commerce domain.
Complementing the LPM is our innovative agentic framework, which captures real-time user behavior and employs rapid iteration and continuous feedback loops to discover and apply optimal solutions dynamically. This unique blend of contextual intelligence and iterative learning is redefining customer engagement.
Proven Results
The results speak for themselves. Our recent implementations with select clients have achieved impressive outcomes, including:
- 51% increase in conversion rates
- 46% improvement in shopper engagement
- 18% rise in Average Order Value (AOV)
- 2x improvement in Return on Ad Spend (ROAS)
The Future of Shopping
As AI continues to evolve, we foresee a shopping landscape increasingly characterized by interactive, distributed, and even AI-driven shopping experiences. At Spangle AI, we remain committed to ongoing innovation in AI and infrastructure to consistently delight customers—whether they're human or AI-driven shoppers.
