Every AIVA deployment has one question at its centre that's rarely asked out loud: when should a human take over? Get this wrong in one direction and AIVA routes everything to an agent, producing no efficiency gain. Get it wrong in the other direction and AIVA handles things it shouldn't, producing angry customers who feel they were denied a human when they needed one.
Getting escalation right is the single biggest determinant of deployment success. Here's how we think about it.
The wrong way to think about escalation
The most common failure mode is configuring AIVA to escalate when it's uncertain about an answer.
The intuition is reasonable: if AIVA doesn't know something, send it to a human who does. In practice, this creates a system where AIVA handles only the most routine queries — the ones so standard that a FAQ page could answer them — and routes everything moderately complex to agents. You've built an expensive FAQ search engine.
The problem is that AIVA's value isn't just knowing answers. It's handling the conversation — managing the back-and-forth, gathering context, pulling data from systems, executing actions. AIVA can handle a complex query it doesn't immediately know the answer to by asking clarifying questions, fetching relevant data, and reasoning toward a resolution. It should do that, not escalate.
Escalation is not a solution to AIVA not knowing something. It's a solution to specific human requirements that AIVA can't meet.
When to actually escalate
We've found three categories of situations that genuinely require human involvement.
Emotional distress. When a customer is upset — genuinely upset, not just mildly frustrated — the conversation needs a human. AIVA detects distress signals: raised tone indicators in voice transcription, specific emotional language patterns ("this is unacceptable," "I want to speak to a manager," "I've been waiting for weeks"), and repeated unsuccessful resolution attempts in a single session. When these fire, we escalate immediately, regardless of whether the underlying query is one AIVA could technically handle.
The reason: an upset customer isn't asking a question. They're expressing a feeling. AIVA is very good at answering questions. It's not a substitute for a human who can validate someone's frustration.
Legal, compliance, and account security. Anything involving account authentication beyond standard verification, billing disputes over a threshold (configurable by the customer), fraud concerns, legal requests, and regulatory queries. These require human judgment, human accountability, and in many cases documented human decision-making. AIVA identifies these categories and routes them immediately.
Explicit human request. If a customer says they want to speak to a person, they get a person. We don't try to talk them out of it. We confirm the transfer and hand off.
The three escalation triggers — emotional distress, compliance/legal, explicit request — are baked into every AIVA deployment and aren't configurable to off. They exist because getting them wrong has consequences that outweigh any efficiency gain.
The middle zone: complexity without escalation
Between "AIVA can handle this" and "this needs a human" there's a large category of complex queries that AIVA should handle, not escalate.
A customer who wants to reschedule an appointment, check insurance coverage, get directions, and confirm the fee structure is asking four things, not one. It's complex. It's not emotional (unless handled badly). AIVA can work through it with the right integrations, the right configuration, and the right patience.
The rule we use internally: escalate for who the customer needs (a human), not for what they need (an answer). Most complex queries still just need an answer. Complex ≠ human.
Calibrating for your context
Every deployment needs its own calibration. A healthcare company's escalation triggers are different from a logistics company's. The right emotional distress threshold for a lending platform is different from one for a fashion retailer.
We recommend a two-week calibration period where you review escalation logs daily. Look for patterns: are specific query types escalating that AIVA should be able to handle? Are there situations where AIVA didn't escalate but should have? Adjust triggers based on what you see.
The goal isn't zero escalations. Some escalations are correct. The goal is that when AIVA escalates, the human agent gets a call that actually needed them — not work that AIVA should have handled.