Reinventing Retail Discovery: How Connected Data and AI Can Turn Search Into Intelligence

A shopper searches for “a lightweight jacket for rainy weather,” asks an AI assistant for a birthday gift for a 10-year-old, or looks for products that work with something they already own.

The request sounds simple. The data required to answer it is anything but.

Modern retailers have vast amounts of information spread across product catalogs, categories, brands, customer interactions, purchase history, reviews, inventory systems, pricing platforms, search indexes, and, increasingly, vector databases that power AI applications.

Rather than focusing solely on rapid retrieval or storage, retailers increasingly need to understand relationships across these data points in real time to power more relevant recommendations and answers. This is becoming especially important as retailers increasingly extend traditional search and recommendation experiences with conversational shopping and AI agents.

An AI assistant can generate an impressive response, but its usefulness depends on its ability to access accurate, up-to-date, and contextually relevant retail data. A connected data foundation can help retailers make that shift.

The Retail Discovery Problem: Too Much Data, Too Little Context

Traditional retail architectures often use specialized systems for different jobs.

A transactional database stores product and customer data. A search engine handles keyword queries. A vector database or vector index stores embeddings for semantic search. Other platforms manage reviews, inventory, pricing, recommendations, or customer profiles.

Each system can perform its role effectively. The problem emerges when a shopper’s request crosses several of those domains at once.

Consider a customer asking: “I need a gift for a 10-year-old who loves science. Something under $1,000 that can be delivered this week.”

Answering that request may require the retailer to understand the relationships between:

Customer intent → Product Category → Product attributes → Age Suitability → Related Products → Price → Inventory → Fulfillment Options

A keyword search can identify products containing relevant terms. A vector search can identify products with semantically similar descriptions. But neither approach, on its own, fully captures the relationships among those products, customers, categories, complementary items, availability, and business rules.

The result can be a familiar set of retail problems:

1. Search lacks semantic understanding

Customers rarely describe products using the exact terminology used in a catalog. Natural-language searches can include preferences, use cases, constraints, and context that aren’t directly represented in product fields.

2. Recommendations lack relationship context

Two products can be semantically similar while having very different roles in a customer’s journey. A product may be frequently purchased with another product, compatible with a specific device, appropriate for a particular age group, or strongly associated with a specific use case. Those connections matter when generating recommendations.

3. Data becomes difficult to keep synchronized

When transactional data, search indexes, vectors, and recommendation systems reside on separate platforms, updates must be shared across them. As product catalogs, prices, inventory, and customer interactions change, synchronization pipelines become increasingly complex.

4. AI applications operate on incomplete context

LLMs can reason over information they receive. They can’t reliably infer relationships that aren’t present in the provided context. For retail AI, grounding the model with isolated product descriptions may be insufficient. The application needs access to the relationships surrounding those products.

From Product Search to Connected Retail Discovery

The next generation of retail discovery requires retailers to think beyond individual records. Products don’t exist in isolation.

• A product belongs to a category. It has a brand. It may be compatible with another product. Customers purchase it alongside other products.
• Reviews describe specific attributes.
• Inventory determines where and when it can be fulfilled.
• A promotion changes its commercial context.
• Customer behavior provides signals about relevance.

Together, these relationships form a connected representation of the retail business. This is where a graph-based data model becomes valuable.

A retail graph can represent entities such as:

• Products, SKUs, categories, brands, and attributes
• Customers and customer segments
• Purchases, clicks, searches, and other interactions
• Reviews and product feedback
• Stores, warehouses, and fulfillment locations
• Pricing, promotions, and availability
• Product compatibility and complementary relationships

Instead of asking only, “Which products are similar to this query?”, retailers can ask more contextual questions: “Which products are relevant to this intent, available to this customer, compatible with what they already own, frequently purchased together, and within their price range?”

That shift can make discovery substantially more intelligent.

Building a Unified Retail Data Foundation With Spanner Graph

Google Cloud’s Spanner Graph brings graph capabilities together with Spanner’s relational database model. It supports graph queries using GQL while maintaining interoperability with relational data through SQL. Google Cloud also supports combining graph traversal with vector and full-text search, enabling applications to leverage semantic similarity, keyword precision, and connected relationships.

For retailers, this creates an opportunity to bring several previously separate capabilities into a unified data layer. Target provides a real-world example by consolidating transactional data, graph relationships, vector embeddings, and full-text search in Spanner Graph, ultimately retiring its legacy Elasticsearch stack.

1. Create a connected retail ontology

The first step is defining the entities and relationships that matter to the business.

A retailer might model:

Customer → searched for → Product

Customer → purchased → Product

Product → belongs to → Category

Product → made by → Brand

Product → compatible with → Product

Product → frequently purchased with → Product

Product → available at → Store

Product → eligible for → Promotion

This ontology gives AI applications more than a collection of documents. It provides a structured representation of how the retail business works.

Spanner Graph can create property graphs from relational data using a declarative schema, allowing existing relational data to be mapped into a graph model without requiring a wholesale migration away from the relational model.

2. Combine graph relationships with semantic and keyword search

Different search techniques solve different parts of the retail discovery problem.

Full-text search is useful when exact terms and keyword relevance matter. Vector search is useful for understanding semantic similarity and natural-language intent. Graph traversal adds relationship context that connects products, customers, and other entities. Combining them can produce a much richer discovery experience.

For example, a shopper could search for: “comfortable shoes for walking around a city in rainy weather”

Semantic search can identify products matching the intent. Full-text search can provide keyword precision. Graph relationships can then identify products frequently bought with those shoes, as well as relevant accessories, compatible products, complementary products, and other contextual signals.

Google Cloud mentions vector-plus-graph patterns specifically for finding relevant entry points by leveraging semantic similarity, then traversing connected relationships. Spanner also supports hybrid full-text and vector search patterns for combining keyword precision with semantic relevance.

3. Give AI applications richer context with GraphRAG

Retrieval-augmented generation, or RAG, is often used to provide AI models with information from an enterprise data source.

For retail, basic vector retrieval can still leave important context behind. Imagine an AI shopping assistant recommending a camera. The most semantically similar products might include several cameras with similar descriptions, but the shopper could also need:

• Compatible lenses
• Suitable accessories
• Available stock
• Relevant price or promotion
• Products frequently purchased together
• Alternative products at a lower price
• Products available for delivery to the customer’s location

Graph-based retrieval can traverse those relationships before passing context to the model.

This creates a GraphRAG pattern in which vector search helps identify relevant information, and graph traversal helps uncover the relationships surrounding it. Google Cloud describes this combination as a way to provide more precise context for AI applications.

The result is an AI assistant that can generate responses grounded in a connected retail context rather than simply retrieving a collection of similar documents.

Example: From Search Query to Personalized Shopping Journey

Consider a retailer with millions of products and a customer searching for: “I need a complete home workout setup for a small apartment.”

A traditional keyword search might return fitness equipment containing terms such as “home workout” or “apartment.”

A connected discovery architecture can go further and identify home fitness products, evaluate available attributes and dimensions, discover complementary products, apply customer preferences and budget constraints, check availability, and rank options based on the signals and business rules available to the system.

The AI assistant could then produce a response such as:

A compact set for a small apartment could include an adjustable dumbbell set, a foldable exercise mat, and resistance bands. These products fit the available space and can be purchased together.

The underlying system has effectively connected:

Natural-Language Intent → Semantic Search → Product Relationships → Customer Context → Inventory → Commercial Constraints

That is a fundamentally different approach to retail discovery.

A Practical Architecture for Retail AI

A modern architecture can be organized into three complementary layers.

Retail data and enrichment layer

This layer integrates product catalog data with relevant metadata from systems such as PIM, ERP, CRM, commerce platforms, inventory systems, and customer interaction streams.

AI can also help enrich product information by extracting attributes, classifying products, improving descriptions, and identifying relationships that may otherwise remain implicit.

The goal is to create higher-quality and more connected retail data before it reaches customer-facing AI applications.

Unified graph, search, and operational data layer

For retailers already invested in Google Cloud, Spanner Graph can serve as the connected data foundation, bringing relational data and graph relationships together while supporting full-text and vector search capabilities.

This can reduce the amount of duplication and synchronization required between previously separate search, vector, and transactional systems.

For retailers, this can provide a more unified way to query product data, relationships, and operational data alongside search and vector signals.

AI and application layer

The final layer delivers the experience to customers. This could include:

• Conversational shopping assistants
• Personalized search
• Recommendation engines
• Gift finders
• Product comparison assistants
• Cross-sell and upsell experiences
• Intelligent merchandising
• AI agents that help customers navigate complex catalogs

The application retrieves relevant context from the connected data, applies the appropriate business and AI logic, and generates a response that reflects both the customer’s intent and the retailer’s actual data.

Modernizing Without Rebuilding Everything

Replacing a critical search or discovery platform in one step can introduce unnecessary risk. Retailers need their commerce systems to remain available while the new architecture is validated. A phased migration approach can reduce that risk.

1. Map the existing data model

Identify important entities, relationships, source systems, data owners, and business rules. Start with the use cases that deliver the clearest business value, rather than trying to model every relationship in the organization at once.

2. Build the graph alongside existing systems

Create data pipelines that continuously apply operational changes to the new environment. This allows retailers to develop and test graph-based discovery while keeping existing customer experiences operational.

3. Validate with real traffic

Use controlled traffic shifting or canary deployments to compare relevance, latency, data consistency, and application behavior against the existing architecture.

4. Retire redundant infrastructure

Once the new architecture has demonstrated the required performance and reliability, legacy search or synchronization components can be reduced or removed where appropriate. This incremental pattern helps retailers modernize their data foundation without turning the transformation into a single high-risk migration event.

Where Connected Retail Data Can Create Value

Retail discovery is only one application.

Once products, customers, interactions, locations, transactions, and other business entities are represented as a connected data model, the same foundation can support a wider range of use cases.

• Personalized recommendations: Identify products based on customer intent, behavior, and relationships rather than similarity alone.
• Product bundling: Discover complementary products and frequently associated purchases.
• Conversational commerce: Give shopping assistants structured context for more relevant and grounded answers.
• Merchandising intelligence: Analyze the relationships among products, categories, customers, and demand signals.
• Inventory-aware discovery: Combine product relevance with current availability and fulfillment relationships.
• Fraud and anomaly detection: Explore unusual patterns across customers, accounts, transactions, or other connected entities.
• Customer 360: Connect interactions across channels to create richer customer context.

The common requirement is the same: understanding the relationships between pieces of information, not simply storing them.

The Future of Retail Search Is Contextual

Retailers have spent years optimizing search relevance, recommendation models, and customer segmentation. AI is changing the expectations around all three.

Customers increasingly expect to describe what they want naturally and receive an answer that understands their intent, constraints, preferences, and context. Meeting that expectation requires more than adding an LLM to an existing search box. The underlying data architecture needs to understand the retail business as a connected system.

A graph-based approach can provide that foundation by connecting products, customers, transactions, content, inventory, and business relationships. When graph traversal is combined with relational queries, full-text search, and vector similarity, retailers can create discovery experiences that are both more intelligent and more grounded in real business data.

For retailers preparing to build conversational commerce, AI-powered recommendations, or agentic shopping experiences, the strategic question is increasingly shifting to “Does our data architecture give AI enough context to make the right decision?” The answer starts with connected data.

How Kartaca Can Help

Building an AI-powered retail discovery experience requires more than implementing a graph database.

Retailers need a data architecture that can connect product, customer, transaction, inventory, and content data while delivering the performance, scalability, security, and governance required for customer-facing applications.

At Kartaca, we help retail businesses design and implement cloud and data architectures that integrate operational data, analytics, and AI. With expertise across Google Cloud, data analytics, AI, cloud architecture, and application modernization, we help organizations identify the right data foundation for use cases such as personalized discovery, recommendation engines, conversational commerce, and intelligent retail applications.

From data integration and architecture design to AI-ready data foundations and production deployment, we help retailers move from fragmented data to connected intelligence.

Contact us today to explore what connected data could enable for your retail business.