Why “Personalized” Ecommerce Still Feels Generic in the Age of AI

Ecommerce has accumulated years of customer and product data, yet many shopping experiences still rely on broad segments and disconnected signals. As AI agents begin reshaping product discovery, the industry is confronting a deeper problem: personalization is only as intelligent as the commerce data behind it.

For more than a decade, personalization has been one of ecommerce’s biggest promises. Yet the experience often remains surprisingly predictable: a customer buys a product, browses a category, or searches for something once, and the retailer responds with recommendations based on what similar customers did before. That gap matters because consumer expectations have moved faster than much of the underlying infrastructure. McKinsey found that 71% of consumers expect personalized interactions and 76% become frustrated when they do not receive them. Companies with faster growth also derive 40% more revenue from personalization than slower-growing peers.

Now artificial intelligence is raising the bar again. In Asia Pacific, almost three-quarters of consumers are already using AI to discover, compare or learn about products, according to Deloitte. Meanwhile, 29% of consumer businesses in the region are already adopting agentic AI, a figure Deloitte expects to reach 76% within two years. The challenge is that AI cannot create contextual understanding from disconnected information. As retailers move toward AI-assisted and agentic commerce, the weakness may no longer be the recommendation engine. It may be the data architecture underneath it.

Personalization Has Often Meant Segmentation

Traditional ecommerce personalization has largely been built around purchase history, browsing behavior, demographics and predefined customer segments. These systems can be effective at predicting what a customer is statistically likely to click or buy, but they do not necessarily understand why that customer is shopping at a particular moment. That distinction is becoming increasingly important.

“Most ‘personalization’ today is really segmentation i.e. rules-based recommendations based on thin, siloed data. Retailers rarely have product, customer, and transactional data joined in one place, so agents default to generic outputs,” Ashwin Puri, CBO and Co-founder of Graas, told AsiaTechDaily.

The problem is not necessarily that retailers lack data. Many have enormous amounts of it. The problem is that customer profiles, product catalogs, transactions, inventory, pricing and fulfillment information can remain distributed across different systems. A retailer can therefore know a great deal about a customer without having a coherent picture of the customer’s current context.

Generative AI and agentic systems change the nature of the recommendation. A conventional recommendation engine might conclude that someone who bought running shoes could be interested in another pair of running shoes. An AI shopping agent can potentially interpret a much more complex request: finding a lightweight running shoe for a specific budget, suitable for a particular type of training, available in the customer’s size and deliverable by a particular date. That requires more than customer history.

It requires customer intent, product intelligence, inventory, pricing, availability and fulfillment to work together. Deloitte describes this shift in Asia Pacific as a move from static segments toward continuous, real-time personalization across channels. But the firm also identifies technology foundations, implementation challenges, governance, security and trust as barriers to agentic commerce. This is why better AI models alone may not solve ecommerce’s personalization problem. A model can reason across information only if the relevant information is accessible, consistent and connected. McKinsey has similarly argued that scaling personalization requires a stronger underlying technology stack spanning data, decisioning, execution and measurement.

Product Data Is Just as Important as Customer Data

The personalization conversation has traditionally focused heavily on understanding the customer. AI shopping introduces an equally important requirement: understanding the product. An agent needs to distinguish between products based on specifications, attributes, variants, compatibility, price, availability and substitutes. It must also understand how products relate to one another. Google’s Shopping Graph illustrates the scale of this challenge. Google says its system contains more than 50 billion product listings, with around 2 billion updated every hour, and is increasingly being connected to AI-powered shopping experiences.

This points to a broader change in ecommerce infrastructure. Product information is no longer simply catalog content displayed to consumers. It increasingly needs to become machine-readable commercial intelligence that AI systems can reason over. That makes product-data quality, consistency and connectivity strategic issues rather than back-office concerns.

Asia’s Fragmented Commerce Makes Context Harder

The problem becomes even more complicated across Asia. Consumers in the region move between marketplaces, brand websites, social commerce, physical stores, messaging platforms and increasingly AI interfaces. Different countries bring different languages, consumer behaviors, payment systems and retail structures.

At the same time, Asia Pacific is becoming an important testing ground for agentic commerce. Deloitte expects the region to account for around two-thirds of the world’s new retail sales over the next five years, powered by more than 4.3 billion shoppers. The region’s fragmented commerce environment therefore creates a difficult technical question: how can an AI system provide contextual recommendations when the customer’s commerce journey itself is spread across multiple platforms and systems? The answer may increasingly depend on connecting information rather than simply collecting more of it.

The Emerging Race Is for Commerce Context

This is where technologies such as knowledge graphs become relevant. Instead of treating information as isolated records, a connected commerce model can represent relationships between customers, products, transactions, preferences, inventory and other commercial signals. For example, knowing that a customer bought a product is useful. Knowing that the customer bought it for a particular purpose, prefers a particular specification, has rejected certain alternatives and can receive a replacement tomorrow creates substantially more context.

The market is already beginning to invest in this infrastructure. In August, Graas announced a $17 million Series B and the acquisition of Singapore-based Trustana, an AI-native product-data platform. Graas said the acquisition would strengthen its Commerce Knowledge Graph and its agentic commerce platform by combining product and customer data. The significance extends beyond one company. It reflects a broader shift toward building the data layer required for AI-native commerce.

AI Could Make Personalization More Powerful, and More Wrong

There is also a risk in assuming that more context automatically means better personalization. An AI system operating on incomplete or outdated data can make a recommendation that sounds highly personalized while being commercially wrong. It could suggest an unavailable product, misunderstand a customer’s current intent, recommend something already purchased, or infer a preference that no longer applies. That makes the next stage of personalization partly a trust problem.

Retailers will need to balance relevance with transparency, data quality, privacy and consumer control. The objective cannot simply be to maximize how much an AI system knows about a customer. It must be to ensure that the context it uses is accurate and appropriate. The ecommerce industry’s personalization problem was never simply a shortage of algorithms. Retailers have spent years accumulating customer, product and transaction data. The deeper challenge is making that information work together.

“AI can meaningfully close that gap, but only once the underlying data layer is unified – that’s our bet behind combining data into a single Commerce Knowledge Graph,” Puri told AsiaTechDaily.

As AI agents become more involved in product discovery and eventually purchasing, generic recommendations may become less acceptable. Consumers will expect systems to understand not just who they are, but what they need, what is available and why a particular product makes sense in that moment. The next competitive advantage in ecommerce may therefore not come from having the most sophisticated AI model. It may come from having the most coherent understanding of commerce context. Personalization can finally become genuinely personal, but only when the data behind it can understand each other.

Quick Takeaways

• Ecommerce personalization still often means segmentation, with retailers relying on rules and fragmented customer data rather than real-time context.
• AI is raising consumer expectations by making it possible to interpret shopping intent instead of simply predicting clicks or purchases.
• The biggest constraint may be data architecture, not AI models. Customer, product, transaction, inventory and fulfillment data often remain disconnected.
• AI shopping agents need deeper commerce context to make useful decisions, requiring product intelligence and customer information to work together.
• Asia’s fragmented ecommerce landscape makes contextual personalization harder, as consumers move across marketplaces, social commerce, brand sites and physical stores.
• Product data is becoming as important as customer data, because AI agents need accurate information about specifications, availability, pricing, variants and compatibility.
• Knowledge graphs and connected commerce data could become critical infrastructure for AI-native shopping by linking customers, products, transactions and preferences.
• More personalization also creates new risks. AI can produce highly confident recommendations that are wrong if the underlying data is incomplete or outdated.
• The next ecommerce advantage may be context, not just better AI models. As Ashwin Puri told AsiaTechDaily, AI can close the personalization gap only when the underlying data layer is unified.