Finding the right approved content in Engage just got easier. Engage search now understands what you mean, not just the exact words you type. This means Engage users can find what they're looking for faster, even when their search terms don't match a document's exact wording.
What It Is, How It Works, and How to Control It
Deterministic AI — what we use Consistent, predictable outputs — the same category of technology as Google Search and Netflix recommendations. No text generation, no randomization, no hallucination. | Generative AI / LLMs — what we do not use Produces novel text on the fly (like ChatGPT). Can hallucinate or fabricate information. AI Assisted Search does not use generative AI of any kind. |
What Is It?
AI Assisted Search upgrades how Engage finds content for your advisors — combining traditional keyword search with AI-powered similarity search, so advisors surface relevant documents even when their query doesn't exactly match the words in a document.
How It Works
Three deterministic, AI-trained components — none of which generate text or make judgment calls:
1. An embedding model converts each document's meaning into a numeric vector when it's uploaded. This is deterministic — the same document always produces the same result.
2. Vector similarity search blends keyword matching with semantic similarity, finding relevant content even when the exact words don't appear in the document.
3. Results surface the most similar, already-approved documents from this index. There is no randomization, no generated text, and no possibility of hallucination.
What the AI Does Not Do
• Generate, summarize, or interpret content for advisors
• Use large language models (LLMs) or produce on-the-fly AI output
• Make compliance decisions or surface unapproved content
• Use advisor or client personal data
• Change the advisor interface — the experience looks identical to today
Firm Controls
AI Assisted Search is enabled by default. If your firm would like it turned off, contact your Red Oak Account Team at any time.
Risk Profile
Because this is deterministic — not generative — the risk profile differs significantly from LLM-based features:
• No hallucination or fabricated output
• Only existing, already-approved documents are surfaced
• Consistent, auditable results
• Same compliance controls you have today
AI Risk Governance Document
AI Governance Framework for Deterministic Search
4U Engage's AI-Enhanced Search feature improves how advisors and distribution partners find content on the platform. It uses deterministic, pre-trained machine learning models — not a generative AI model and not an in-house trained model — delivered through Amazon Web Services (AWS) infrastructure that already operates within Red Oak's existing AWS Customer Agreement. This approach lets 4U benefit from proven, enterprise-grade search infrastructure while Red Oak maintains full control over what content is indexed, how search behaves, and whether the feature is used at all.
Our framework for AI-Enhanced Search consists of:
Existing Infrastructure, Not a Third-Party AI Vendor: 4U's AI capability runs entirely within Red Oak's own AWS environment, using AWS Bedrock (Titan embedding model) and AWS OpenSearch — both already covered by Red Oak's AWS Customer Agreement. There is no new third-party AI vendor relationship to evaluate.
Deterministic Architecture, Not Generative AI:
The feature combines traditional keyword search with vector similarity search to rank existing, already-approved content — it does not generate, summarize, or interpret content.
The same document, and the same query, always produce the same vector — there is no randomness, and hallucination is not possible because no new content is ever created.
Content is embedded into vectors when it's indexed (uploaded or updated). At search time, the advisor's query text is embedded using that same model so the two vectors are comparable — this is the only real-time AI step, and it stays inside Red Oak's own AWS environment.
Validation Methodology and Ongoing Monitoring:
A defined benchmark set of test queries and expected results is used to measure search precision, recall, and performance before and after changes.
Quarterly review of search quality against these baseline metrics, including the platform's historical zero-result rate.
Annual review of the AWS Bedrock terms of service governing the embedding model.
Any change in AI vendor, embedding model, or data inputs — or the addition of generative AI components — triggers a new governance review before deployment.
This framework lets 4U take advantage of enterprise-grade AWS infrastructure already used across the platform, while keeping full visibility into what data is processed, how search results are produced, and validation steps appropriate to a deterministic — not generative — AI capability.
Frequently Asked Questions
What does this mean for our clients and their data?
AWS does not use the content processed through this feature to train, fine-tune, or otherwise modify any AI model. AWS Bedrock's Terms of Service explicitly prohibit using customer inputs or outputs to train the underlying foundation models.
The content sent for processing is limited to content already approved and visible to advisors on the 4U platform — never client or advisor personal information.
Advisor search queries are converted into vectors using the same AWS Bedrock embedding model used to index content — this keeps query and content vectors comparable. That conversion happens inside Red Oak's own AWS environment; query text is not sent to any AI vendor or environment outside Red Oak's AWS account.
All AI processing for this feature runs inside Red Oak's existing AWS account and region — the same environment already used for the rest of the 4U platform. No new third-party data store or AI vendor is introduced.
AWS maintains SOC 2 Type II, ISO 27001, and FedRAMP authorizations covering the infrastructure this feature runs on; current reports are available on request. 4U is covered by its own SOC 2 Type II report (distinct from Red Oak's corporate SOC 2), which includes this feature.
AI-Enhanced Search is enabled by default. Firms that would like it turned off can do so at any time by contacting their Red Oak Account Team.
Implementation is done through configuration — indexing content and enabling or disabling the search behavior — not through training a model on your firm's data. Nothing configured for this feature is shared across clients.
What AI service is used? What is the policy for changing or upgrading it?
AI-Enhanced Search is powered by AWS Bedrock's Titan embedding model, combined with AWS OpenSearch for vector similarity search. Both operate inside Red Oak's own AWS account under Red Oak's existing AWS Customer Agreement — this is not a relationship with an external AI vendor in the way a large language model API would be. The same embedding model and configuration are used to convert both indexed content and advisor search queries into vectors, since two vectors are only comparable if they were produced the same way. Any change to the AI service provider, the embedding model, or the data sent to it requires a documented review and re-approval before it reaches production.
What are the potential risks (bias, inaccuracies)? How will these risks be mitigated? What is the monitoring process in place?
Because this feature ranks existing content rather than generating new content, the primary risk is that search results are less relevant than intended for certain queries or document types — not fabricated or biased new content. Red Oak has built a benchmark set of test queries and expected results to measure search quality (precision and recall) before and after any change, and reviews these results quarterly against baseline metrics, including the platform's historical zero-result rate. Formal bias and fairness testing across content types and terminology is an area of ongoing work, and any findings will inform adjustments to the underlying search configuration.
How will users be informed about the platform's capabilities and limitations, and how will feedback be gathered? Who's the Owner?
The 4U product team will make information about this feature available through the Support Center and account communications, and will brief distribution partners as part of regular account reviews. Feedback can be shared with your Red Oak Account Team at any time. The 4U Distribution Product Manager owns this feature and its governance documentation, with review from Red Oak's Information Security team.
How will incorporating AI impact your current processes/workflows?
AI does not change your content approval workflow. Only content that has already been approved through your firm's existing compliance process is indexed and searchable — AI-Enhanced Search does not introduce a new approval step or bypass existing ones.
AI does not generate or alter content. It only changes which of your already-approved documents are surfaced in search, and in what order, based on relevance to the advisor's query.
Your advisors' experience is unchanged. There is no new interface, prompt, or AI-visible interaction — search looks and works the way it does today, just with better results.