Summary of Key Points
AI-powered shopping is considered the next trillion-dollar market (expected to reach $5 trillion globally by 2030), with both giants and startups investing heavily. However, the current experience falls far short of expectations. While AI can handle products with clear parameters such as washing machines and tires, it struggles with items that require taste and emotional appeal, like clothing. Technical challenges include issues with inventory management and checkout processes. A more profound problem is that AI cannot understand human aesthetic preferences and the emotional value behind brands. To make significant progress, AI shopping needs to overcome these technical hurdles, enable brands to communicate effectively with machines, and develop accurate profiles of user tastes.
I. AI Shopping: A $5 Trillion Opportunity, but Still Can’t Buy a Simple Dress
The potential of AI shopping is impressive; McKinsey estimates it could generate $5 trillion in global transactions by 2030, with ChatGPT receiving 50 million shopping queries daily. Yet, the reality is that even simple tasks, like buying a suitable dress from Sézane for the author, prove difficult. The recommended styles are either outdated or out of stock, or users have to manually search for the links.
Why is this the case? According to Forrester analysts, most AI shopping tools were launched in a rush due to FOMO (fear of missing out), and their testing was inadequate. For example, Walmart’s Sparky chatbot is only slightly better than a regular search bar; cutting-edge models don’t even know if products are available, let alone handle direct orders. The promise of AI-driven shopping remains largely unfulfilled.
II. AI Shopping’s Strengths: Parameter-Based Products
AI isn’t completely useless; it excels in certain scenarios. For instance, when the author’s washing machine breaks down, providing the model and size to Gemini’s laundry service results in a suitable GE model being recommended. Walmart’s Sparky can also find the right tires based on the car model. These are products with clear parameters that AI can process quickly to save users time.
However, when it comes to more subjective items like clothing, perfume, or home decor, AI struggles. The author likes Sézane not for its materials or price but for the “French casual, effortless style” it embodies. AI currently lacks the ability to understand such emotional preferences and relies on random keyword recommendations.
III. Technical Challenges: Overcoming Barriers
A major obstacle in AI shopping is the lack of integrated data. Large language models (LLMs) can only access public information online and cannot obtain real-time inventory, member discounts, or checkout processes from merchants. For example, a dress recommended by ChatGPT might be from the previous season because the model doesn’t know about inventory updates.
Giants are working to address this issue: OpenAI and Stripe have developed “agent commerce protocols,” while Google and Shopify have launched “universal commerce protocols.” Shopify’s Catalog allows AI to see merchant inventory and pricing information. Checkout processes are also complex, with varying coupon and membership benefits, as well as different shipping requirements. OpenAI’s initial instant checkout feature with Shopify was discontinued due to technical issues and user reluctance. Google successfully launched such a feature in January but acknowledges the complexity.
IV. A Deeper Challenge: Understanding Brands and User Tastes
To improve AI shopping, we need to help machines understand brands’ emotional values and user tastes:
- How can brands make themselves understandable to AI? Traditional brands use visual elements (billboards, TikTok videos), but AI processes text. For example, Lululemon and Vuori’s tight pants have similar materials and prices, so AI must rely on online reviews and forums to distinguish them, which may not capture the core brand values. Brands need to actively communicate their emotions and values through written content online.
- How can AI understand user tastes? Current tools rely on users’ verbal inputs, but platforms like TikTok and Instagram use behavioral data to build taste profiles. Google’s Gemini can now integrate with Gmail and photos, potentially understanding preferences based on purchase history and vacation photos (e.g., a preference for French minimalism over Italian extravagance).
V. The Future of AI Shopping: Personalized Buying Assistants
If these challenges are overcome, AI shopping will evolve into personalized buying assistants. It won’t just be a keyword search tool but will understand your spending habits (e.g., your willingness to spend $300 on a shirt but not on a $50 candle) and brand preferences (e.g., always choosing Sézane over Reformation). It could even predict the equipment needed for a ski trip and place orders automatically.
This would transform the shopping experience: Current algorithms aim for mass engagement, leading to homogenized tastes, while AI-assisted shopping would provide more personalized recommendations. Brands that can effectively communicate their values to AI will reach their target audience more accurately.
Of course, this will take time, but AI is evolving rapidly. Maybe soon, users like the author will be able to rely on AI to find the perfect dress for a garden event.
---
This translation maintains the structure and tone of the original Chinese analysis, adapting the language to fit financial journalism standards while ensuring the accuracy and consistency of financial and business terminology.