Google Vertex AI Search + Optimizely Integration Guide

This guide walks through the integration process, from prerequisites to production deployment. You'll learn about Optimizely's Commerce Search v3 architecture, the data requirements for each AI tier, and how to avoid common pitfalls during implementation. This Google Vertex AI Search Optimizely integration represents a modern B2B commerce search solution for organizations seeking semantic product search ecommerce capabilities.

Prerequisites

Before starting the integration, you'll need:

• Optimizely Configured Commerce version 5.2.2509 sts (June 2025 release or later)
• Opti ID configured for your environment
• Language tags properly set up
• Spire websites (Classic sites aren't supported)
• Unique-email validation enabled

Commerce Search v3 is the official built-in integration for Configured Commerce. Customized Commerce can also integrate, but Classic sites won't work with this approach. This Optimizely Configured Commerce search capability requires proper setup before proceeding.

One requirement that's easy to overlook: you'll need to submit a manual provisioning request to Optimizely before you can enable the feature. Plan for this lead time in your project schedule.

Understanding the Architecture

Commerce Search v3 uses what Optimizely calls a "blended search architecture." The AI handles high-impact relevancy work like product discovery, category browse ordering, and autocomplete. Non-relevancy lookups, like quick order queries or basic product data retrieval, can stay on Elasticsearch v7.

This split makes sense. You get Google's semantic search where it matters most without overhauling every search function in your system.

Relevancy-based queries route through an Optimizely commerce search service that connects to Google Cloud Vertex AI Search for Commerce. Behind the scenes, this is Google's Retail Search product, which they've rebranded several times. You might see older references to "Retail Search" or "Discovery AI for Retail" in documentation. Understanding this Google Retail Search integration history helps when troubleshooting documentation references.

Implementation Steps

1. Review Your Custom Extensions

Start by auditing any custom search extensions in your current setup. Some may need updates to work with the new architecture.

2. Set Up ODP Integration

While Optimizely says the Optimizely Data Platform isn't strictly required for basic Commerce Search v3, we've found it's practically necessary if you want meaningful results. ODP feeds behavioral signals (clicks, cart additions, purchases) into the AI, which directly affects search quality. This ODP Optimizely Data Platform integration is essential for advancing through the tier system.

Create integration jobs to sync:

• User events and historical orders
• Products and warehouses
• Customers and user profiles
• Customer-user mappings

Schedule real-time incremental updates for active data, with scheduled "safety net" syncs as backup.

Privacy note: ODP-collected tracking data gets synced to Google. Update your privacy statements and consent language to reflect this third-party sharing.

3. Sync Your Product Catalog to Google Cloud

Your catalog data needs to meet specific quality thresholds:

• Products with valid, accessible URIs: 95%
• Products with detailed descriptions: 90%
• Titles with at least 2 words: 80%
• Duplicate titles: Under 50%
• Searchable attributes: At least 5

4. Enable in Admin Console

Once your data pipelines are running, enable Commerce Search v3 through the Admin Console for your test environment first.

5. Configure Vertex AI Search Settings

Set up your search configuration including dynamic faceting preferences, query expansion rules, and autocomplete behavior.

6. Modify Frontend Search UI

Update your Spire frontend to work with the new search responses. The API patterns differ from Elasticsearch, so plan for UI adjustments.

7. Set Up Analytics and Feedback Loops

Connect analytics to track search performance. The AI learns from user behavior, so these feedback loops directly affect future result quality.

Understanding Relevancy Tiers

Vertex AI Search for Commerce uses a tier system. Each tier brings better search capabilities, but requires more data. Understanding Commerce Search v3 tier requirements is critical for planning your implementation.

Tier 2: Relevance Popularity

• Over 100,000 text search or browse events in the last 90 days
• At least 95% product join rate
• At least 95% attribution tokens on search events
• 70% of search requests with associated user events

Tier 3: Revenue-Adjusted Ranking

• Over 200,000 search events in 90 days
• Over 250,000 detail-page-view events following searches
• Conversion ratios: add-to-cart/detail-page-view ≥ 0.02, purchase/add-to-cart ≥ 0.025

Tier 4: Personalized Revenue-Adjusted Ranking

• Over 100,000 attributed search events in 90 days
• Visitor ID matching above 10% (last 7 days)
• User ID presence above 1% (last 7 days)

Our experience shows that most B2B commerce sites need several months of event collection before reaching Tier 2. Plan accordingly, and don't expect immediate improvements in search quality after launch.

Common Mistakes to Avoid

Expecting immediate results. User-event impact isn't instant. Vertex AI needs enough ingested events to train before ranking changes meaningfully. If you launch and don't see improvements within the first week, that's normal.

Assuming autocomplete works like traditional search. Commerce Search v3 autocomplete suggests search terms, categories, and brands. It does not suggest specific products or website content. If your users expect product-name autocomplete, you'll need a custom approach.

Skipping ODP integration. Technically it's optional. But without behavioral data, you're stuck at the lowest tier with basic capabilities. The smarter discovery everyone wants requires those user signals.

Ignoring catalog data quality. If your product descriptions are thin or your titles are duplicates, no amount of AI will save your search results. Fix your data first.

Forgetting about quick order flows. The blended architecture keeps quick-order lookups on Elasticsearch. Make sure you've properly configured which queries route where, or you'll have inconsistent search experiences.

Missing regional considerations. Commerce Search v3 currently runs from North America only. If your primary user base is in Europe or Asia, factor in latency implications.

Testing and Verification

Working with commerce teams has taught us to focus on these areas during testing:

Query expansion testing. Run searches for long-tail and ambiguous queries. Verify that the system returns relevant results even when the exact keyword isn't present.

Facet behavior. Test dynamic faceting by searching for different product categories. The system should surface contextually relevant filters, like showing "lens color" for sunglasses searches.

User event attribution. Verify that user events are properly synced from ODP to Vertex AI. Check your attribution tokens are being passed correctly.

Tier status monitoring. Monitor your tier status in the admin console. Track whether you're meeting the event thresholds for your target tier.

Regression testing. Compare results for your top search queries before and after the migration. Document any unexpected changes.

Known Limitations

Be aware of what Commerce Search v3 doesn't support:

• Customer-specific part number search at the customer level
• "Search within" functionality
• Instant attribute sync
• Mobile app integration (planned for future)
• Local partner development (use sandbox instead)

If any of these are critical for your business, you'll need workarounds or should wait for future releases.

Migration Timeline

If you're currently on Commerce Search v1, here are the key dates:

• Support ends: February 1, 2026
• Migration deadline: June 30, 2026
• Infrastructure support ends: July 1, 2026

The typical migration path goes v1 → v2 → v3. Start planning now if you haven't already.

Conclusion

Integrating Google Vertex AI Search with Optimizely Configured Commerce brings semantic search capabilities to product discovery, but success depends on catalog data quality and behavioral event volume. The tier system means you'll need patience and proper ODP integration to reach the AI's full potential.

The payoff can be significant. Google reports retail implementations seeing up to 2% conversion increases, and reduced "no results" pages help customers find what they actually need.

Configuring this integration involves balancing data pipelines, tier requirements, and workarounds for unsupported features like product-name autocomplete. If your team is planning a Commerce Search v3 migration or trying to determine whether your event volume will support the higher tiers, we can help you map out the technical requirements and identify any gaps in your current setup.