Direct Answer (TL;DR)
Brilo AI product recognition may fail to detect a product when the agent's knowledge sources, training examples, or entity rules don’t include the product name, SKU, or a matching description. Typical causes are missing or out-of-date product data in your knowledge base, low NLU confidence thresholds, or mismatched entity extraction rules. To fix it, update the product data source (your CRM or catalog), add representative call examples for training, and adjust confidence or routing rules so ambiguous calls are routed for review. If the issue persists, enable a human handoff on low-confidence matches to collect correct identifiers and retrain the model.
Why didn’t the agent find my product? — The product name, SKU, or synonyms aren’t present in the Brilo AI product data or training examples.
Why does the agent hear a product but not act? — The NLU confidence for the detected entity is below the configured threshold and is therefore treated as unknown.
Why are similar product names confused? — Entity extraction and intent overlap; add disambiguation examples and clearer entity rules to improve separation.
Why This Question Comes Up (problem context)
Buyers ask this because enterprise phone automation depends on accurate product lookup for order status, eligibility checks, or policy verification. In healthcare, banking, and insurance calls, failing to detect a product (or plan, SKU, or policy) can lengthen calls, increase escalations, and create compliance-sensitive errors. Brilo AI customers often surface this question during rollout or after catalog updates because integrations and training data must stay synchronized with live product catalogs.
How It Works (High-Level)
Brilo AI product recognition uses NLU to match caller language to product records, entity extraction to isolate SKUs or names, and a confidence score to decide whether to act or escalate. When configured, Brilo AI looks up products against your connected data sources (your CRM, catalog, or knowledge base) and applies routing logic based on confidence and intent.
In Brilo AI, product recognition is the combined process of matching caller utterances to catalog entries using entity extraction and lookup rules. Intent recognition is the classification of the caller’s goal (for example, “order status” or “claim inquiry”) that determines which product checks run.
Guardrails & Boundaries
Brilo AI will not perform a product action when confidence is below the configured threshold, when required identifiers are missing, or when an integration lookup fails. A confidence threshold is the minimum NLU score required before the agent updates records, authorizes actions, or provides product-specific guidance. Do not rely on product recognition alone for high-risk decisions; configure explicit human escalation or verification steps for sensitive workflows. Avoid broad wildcard matching that can introduce false positives—prefer controlled synonym lists and normalized SKUs.
Applied Examples
Healthcare: A caller asks about a specific medical device model; Brilo AI searches the connected product catalog. If the catalog lacks the model name or synonym, Brilo AI prompts for the device identifier and, on low confidence, routes the call to a nurse or specialist for verification.
Banking: A customer references a specific savings product by nickname. If the nickname isn’t mapped to the bank’s product codes in the CRM, Brilo AI will not complete a product-specific action and will ask clarifying questions or escalate based on routing rules.
Insurance: A policyholder mentions a plan feature; Brilo AI tries to match the feature to a policy SKU. When matches are ambiguous, Brilo AI is configured to collect policy numbers and hand off to an agent to avoid incorrect benefit confirmations.
Human Handoff & Escalation
When Brilo AI cannot confidently recognize a product, configure the workflow to:
Ask a short verification question (for example, “Can you confirm the SKU or policy number?”).
Offer to route to a human when entities are missing or confidence is below the threshold.
Create a structured ticket with the extracted fields and NLU confidence for the agent, so the human agent has context and can retrain the model with corrected examples.
Brilo AI supports routing to live agents or alternate workflows when configured in the call scenario rules.
Setup Requirements
Provide a product data source: connect your CRM, product catalog, or a regularly updated knowledge base.
Supply training examples: upload representative call transcripts or canned utterances that mention product names, nicknames, and SKUs.
Define entity rules: list canonical product names, SKUs, synonyms, and disambiguation patterns for Brilo AI to extract.
Configure confidence thresholds: set the NLU score that determines automatic actions versus handoff.
Set routing rules: create low-confidence paths to human agents or verification IVR prompts.
Enable logging and feedback: make sure Brilo AI stores misrecognitions for periodic review and retraining.
Test with pilot calls: run staged calls across scenarios and refine the knowledge base and entity lists.
Business Outcomes
When product recognition is reliable, Brilo AI voice agent call handling reduces average handle time for routine lookups, decreases avoidable escalations, and improves first-call resolution for transactions like eligibility checks or order status. Fixing recognition issues early reduces manual rework and helps secure consistent customer experiences across healthcare, banking, and insurance interactions. Better recognition also produces cleaner structured call data for analytics and model retraining.
FAQs
How do I know if product recognition failed or the caller used an unfamiliar name?
Review the NLU confidence score and the extracted entity. Low confidence with no entity suggests the caller used an unfamiliar name; a present entity with low lookup hits suggests a missing catalog entry.
Can Brilo AI learn new product names automatically?
Brilo AI can improve over time with supervised retraining using corrected examples you provide. Configure a feedback loop where human agents tag missed products and add them to the product list for ingestion.
What should I do if product data changes frequently?
Automate syncs from your canonical catalog or CRM and schedule regular re-ingestion. Keep synonym lists and SKU mappings in a single source of truth to minimize drift.
Will Brilo AI act on a product match without a policy number or SKU?
That depends on your confidence threshold and workflow rules. For high-risk actions, require an explicit identifier before allowing the agent to take action; otherwise route for human verification.
Next Step
Review your connected data sources and make sure your canonical product catalog is available to Brilo AI (your CRM or catalog).
Collect and upload representative call examples and synonym lists to improve entity extraction and training.
Configure confidence thresholds and low-confidence routing in your Brilo AI call scenarios, then run a pilot and collect misrecognition logs for retraining.