Intent classification
Sonar maps incoming messages to your intent taxonomy: pre-sale, post-sale, support, billing, complaint, escalation, custom.
Feature · Routing
Routing by meaning, not menus
The Sonar intent engine in OrcaLinq reads what your customer actually wants — and decides whether to answer, ask, run a tool, or hand off.
What it is
Routing the conversation by meaning, not menu choice
Phone-era PBX assumes the caller can pick from a menu. Modern customers send messages with full context, screenshots, and voice notes. Routing has to read the message, not force the visitor through a flow.
Confidence scoring
Every classification is scored. Low scores trigger a clarifying question or hand off, never a guessed answer.
Context awareness
Prior conversation, customer profile, channel rules, and business hours all feed into the routing decision.
Policy enforcement
Refunds, complaints, regulatory keywords, and VIP tags route to humans regardless of confidence.
How it decides
Five outcomes, one routing layer
Every customer message goes through the same decision tree. The output is one of five actions.
- Answer. Confidence high, knowledge available, no escalation policy. AI replies.
- Ask. Need a clarifying signal before deciding. AI asks one targeted question.
- Run a tool. Connected tool can resolve (calendar, order lookup, refund). AI calls it.
- Hand off. Confidence too low, policy says human, or customer asks. The right person is paged.
- Escalate. Sensitive intent (legal, medical, security). Owner-defined endpoint is paged with a policy note.
Why it beats brittle flows
Bot builders age. Semantic routing learns.
Flowchart-based bots break when the visitor phrases the request differently. Semantic routing reads meaning instead of matching strings.
No brittle string matching
Customers don't memorize keywords. Sonar reads the underlying request even when the phrasing changes.
Updateable knowledge
Add a new product line, policy, or FAQ entry. The routing engine picks it up on the next message — no flow to redraw.
Auditable decisions
Every routing call is logged with inputs, score, and outcome. You can see why a conversation went where it went.
Routing FAQ
Frequently asked questions
How is semantic routing different from a chatbot flow?
Chatbot flows force visitors down predefined paths. Semantic routing reads what the visitor actually means and decides per-message: answer, ask, run a tool, or hand off.
What signals does Sonar use?
Message content, conversation context, customer profile, channel cues (e.g. WhatsApp template responses), business hours, knowledge availability, and any custom signals you configure.
Can I see the routing decision?
Yes. Every routing decision is captured in the Wake with the input signals, confidence score, and the policy that fired.
Can I override the routing engine?
Yes. Routing policies are versioned, and you can pin specific intents to specific endpoints, force handoff for keyword groups, or disable AI answering entirely while keeping intent classification.
Related pages
Keep exploring
See semantic routing decide a real customer message.
Bring one or two of the messages your team is fielding today. We'll show what intent score, confidence, and routing decision look like in your tenant.