Retrieval-Augmented Generation (RAG)
The problem RAG solves
Section titled “The problem RAG solves”Large language models are trained on general knowledge. When you ask Opal about your campaigns, your approved messaging, or your latest assets, the model has no inherent access to that data. Without a bridge between the AI and your organization’s content, Opal can only generate generic responses that may be outdated or off-brand.
Retrieval-augmented generation (RAG) is that bridge. It enables Opal to search your organization’s data — campaigns, tasks, assets, and metadata — before generating a response. The result is output that reflects your actual content, not generic assumptions.
How RAG works
Section titled “How RAG works”RAG follows a three-step process:
-
Retrieval — When you make a request, Opal searches vector databases and keyword indexes to find information relevant to your query. This search spans structured data (campaign metadata, task status) and unstructured data (document text, image descriptions).
-
Processing — The retrieved data is analyzed and contextualized. Opal determines which pieces of information are most relevant and how they relate to your request.
-
Generation — The AI generates a response using the retrieved context. Because the response is grounded in your actual data, it is accurate, current, and aligned with your brand.
Opal’s RAG architecture
Section titled “Opal’s RAG architecture”Opal’s RAG implementation goes beyond simple document search. It uses several components that work together:
Unified content graph
Section titled “Unified content graph”Your campaigns, tasks, assets, and permissions are modeled as nodes and edges in a content graph. This structure preserves the relationships between items — a campaign contains tasks, tasks reference assets, assets have labels and metadata. Opal queries this graph to understand not just individual items but how they connect.
Multi-modal retrieval
Section titled “Multi-modal retrieval”RAG in Opal searches across multiple content types:
- Text — Document content, campaign descriptions, task notes
- PDFs — Uploaded briefs, guidelines, reports
- Images — Visual assets with metadata and AI-generated descriptions
- Video — Video content with transcriptions and metadata
- Metadata — Tags, labels, dates, status fields, and custom properties
Both vector search (semantic similarity) and keyword search (exact matching) are used to find the most relevant content.
Permission-aware access
Section titled “Permission-aware access”RAG respects your organization’s access controls at two levels:
- Query-level filters — Searches are scoped to content the requesting user has permission to see
- Post-query verification — Results are verified against permissions before being included in the response
This means a user cannot accidentally surface content they should not have access to, even through natural language queries.
Natural language queries
Section titled “Natural language queries”You do not need to learn a query language. The LLM translates your natural language request into optimized graph queries. Ask “What assets are tagged for our spring campaign?” and Opal constructs the appropriate search across your content graph.
Why RAG matters for marketing teams
Section titled “Why RAG matters for marketing teams”RAG is particularly valuable when your organization needs to:
- Maintain brand consistency — AI responses reference your approved messaging and brand guidelines rather than inventing new language
- Create personalized content — Pull customer segment data and product information to tailor messaging
- Scale content creation — Generate blogs, emails, ads, and social content grounded in your existing campaign data
- Ensure compliance — Reference approved legal language and regulatory guidelines stored in your content system
- Find existing assets — Query your DAM by tags, descriptions, or content to locate assets for reuse
Industries that benefit most
Section titled “Industries that benefit most”RAG delivers high value in industries where accuracy and compliance are critical: financial services, healthcare, retail, and enterprise marketing.
Data privacy and security
Section titled “Data privacy and security”RAG processes data internally through Optimizely’s secure infrastructure:
- Enterprise-grade security controls
- Permission-aware access at query and result levels
- Managed data synchronization
- GDPR compliance readiness
- No external data sharing — your content stays within Optimizely’s systems
1. A marketer asks Opal to recommend assets for a spring campaign. Without RAG enabled, what would happen?
Without RAG, Opal has no connection to your organization's content data. It can only generate generic suggestions based on its training data, not reference your actual campaigns, assets, or approved messaging.
Without RAG, Opal has no connection to your organization's content data. It can only generate generic suggestions based on its training data, not reference your actual campaigns, assets, or approved messaging.
Review this topic →2. A user with restricted access to certain campaigns asks Opal a RAG-enabled question. How does Opal handle permissions?
RAG in Opal is permission-aware at two levels: query-level filters scope the search to content the user can access, and post-query verification confirms permissions before including results in the response.
RAG in Opal is permission-aware at two levels: query-level filters scope the search to content the user can access, and post-query verification confirms permissions before including results in the response.
Review this topic →