Brilo AI escalation rules determine when a Brilo AI voice agent stops handling a call and routes it to a human or alternate workflow. Escalation rules are driven by configured conditions such as low confidence scores, explicit caller requests for a human, safety keywords, or limits like maximum clarifying attempts or call duration. When an escalation triggers, Brilo AI passes context (transcript snippets, detected intent, and session metadata) so the human agent can continue without repetition. Administrators configure thresholds and transfer destinations in the agent’s escalation settings.

What triggers escalation? — Escalation rules determine when Brilo AI routes a caller to a human or different flow; administrators set the triggers and destinations.

How are transfers delivered? — Brilo AI performs a warm handoff or a cold transfer based on rules, and includes a summary and transcript excerpts for the recipient.

When does Brilo AI retry first? — Brilo AI can be configured to make clarifying prompts a set number of times before escalating automatically.

Enterprise teams ask this because escalation behavior affects compliance, customer experience, and agent workload. In regulated sectors—healthcare, banking, insurance—buyers must control when automated handling stops and human review starts. Escalation rules also determine telephony usage patterns and staffing needs during peaks. Clear escalation logic reduces caller repetition, prevents risky automated actions, and creates auditable handoffs for supervisors.

Brilo AI escalation rules are a set of conditional checks attached to an agent’s call flow. Typical checks include intent detection results, confidence thresholds, presence of safety or regulatory keywords, caller requests for a human, and configured time or retry limits. When a condition matches, the configured action (for example, call transfer, callback scheduling, or voicemail) executes and Brilo AI attaches handoff metadata so downstream systems or people have context.

Escalation rules are a configurable policy on an agent that lists conditions and corresponding actions. Intent detection is the runtime process the agent uses to label what the caller wants so rules can evaluate whether to continue or escalate. See the Brilo AI article on how the AI understands caller intent for details on intent detection and rule triggers: How does the AI understand what the caller wants?

Brilo AI enforces guardrails so escalation rules don’t create unsafe or confusing behavior. Common guardrails include maximum clarifying attempts before escalation, an absolute low-confidence cutoff that forces immediate handoff, limits on session context length, and prevention of automated actions on regulated requests unless human approval exists. Brilo AI also supports safe keywords that always trigger handoff.

Confidence threshold is the numeric cutoff you set that causes the agent to escalate when intent detection confidence falls below that value. Brilo AI will not complete regulated workflows that your account hasn’t explicitly authorized; instead, the escalation rule should route those calls to a human reviewer. For more on fallback behavior and uncertain-call handling, see: What happens when the AI is unsure?

Brilo AI supports warm handoffs and cold transfers. Warm handoffs send a summary, recent transcript excerpts, detected intent, extracted entities, and session metadata to the receiving agent or CRM so the human doesn’t ask the same questions again. Escalation can be automatic (confidence or keyword triggered) or manual (caller asks for a human). You can also configure callback handoffs where Brilo AI schedules a live-agent callback when no agents are available.

During handoff, Brilo AI preserves the conversation state and can include routing instructions (queue, agent skill, or phonebook destination). Administrators can choose to record the transfer metadata in your CRM or send it to your webhook endpoint.

To learn more about configuring voices, languages, and conversation settings used during these tests, see: Does the AI sound natural or robotic? and How does the AI handle accents and speech variations?

Properly configured escalation rules reduce customer repetition, lower average handle time for human agents, and ensure potentially risky or regulated interactions receive human oversight. For regulated buyers, rules create predictable audit trails and handoff metadata that support investigation and compliance reviews. Escalation rules also let operations scale AI-first support while preserving human review for exceptions.

How do I choose a confidence threshold?

Choose a threshold based on a combination of historical intent accuracy and business risk. Start conservatively (higher threshold for escalation) in regulated workflows, measure false positives and negatives, and iteratively lower the threshold as intent models improve.

What’s the difference between a warm handoff and a cold transfer?

A warm handoff sends context—including transcript snippets, detected intent, and extracted entities—to the receiving human or system. A cold transfer moves the call without context and is generally discouraged for caller experience.

Can Brilo AI escalate to a webhook or external system?

Yes. Brilo AI can invoke a webhook endpoint as part of the escalation action to create tickets, notify agents, or update CRMs, provided you supply the destination endpoint and authentication details in the agent configuration.

How does Brilo AI avoid escalating too often?

You can configure maximum clarifying attempts, cooldown timers, and combined conditions (for example, low confidence plus safety keyword) so the agent only escalates when business rules deem it necessary.

Will callers hear "AI" before a handoff?

You control agent prompts and disclosures. Many customers include a brief announcement before transfer, but Brilo AI does not force a specific disclosure—set the wording in the call flow prompts.

If you’re ready to configure escalation rules, start in the Brilo AI console by editing an agent’s Actions > Call transfer settings and run test calls to confirm the handoff metadata and behavior.