AI call routing system: a virtual receptionist for a multi-department retailer
A conversational front door that answers inbound calls, resolves simple questions and transfers the rest to the right department.
The problem
A retail business with several departments wanted incoming calls to reach the right team on the first attempt, without a human receptionist screening each one.
Challenges
Routing correctly at the first attempt
The business has several departments, and a wrong transfer costs the caller time and the team an interruption. The system had to work out the right destination from what the caller said, without a human receptionist screening the call first.
Understanding needs in the caller's own words
Callers rarely describe their problem in the words a menu expects. The build had to map varied, natural phrasing to the correct department line, rather than forcing people to guess which menu option is closest to what they want.
Avoiding transfers that were never needed
Many calls ask about opening hours, policies or other common questions. Transferring those wastes a department's time, so the system needed reliable company-specific knowledge to settle simple questions itself and transfer only when a person is genuinely required.
How the system works
Conversational triage
An ElevenLabs voice agent answers and asks callers what they need in natural conversation, in place of a fixed phone-tree menu.
Department routing
From what the caller says, the agent makes a blind transfer to the right department line. This implementation covers seven departments.
Knowledge layer
An embedded prompt database gives the agent company-specific details such as opening hours, policies and common answers, so simple questions can be settled without a transfer.
Design decisions
Conversation instead of a phone tree
We replaced the fixed menu with an open question about what the caller needs. Menus push people to press zero or hang up, which also distorts any analytics built on them. Natural language routing avoids that behaviour and feels friendlier.
Blind transfer to the department line
Once the destination is clear, the call is handed straight to that department's line. We kept the hand-off simple because the value is in choosing the right team, and a lighter transfer keeps the stack easy to run and maintain.
An embedded prompt database
Company details such as hours, policies and common answers sit in a knowledge layer the voice system can draw on. Keeping them in one place means simple questions get consistent answers and the information can be updated without redesigning the routing.
Result in production
Callers reach the correct department without working through menus, and straightforward questions are answered on the spot.
Callers reach the correct department without working through menus, which saves them time and reduces mis-routed calls.
Straightforward questions are answered on the spot, so departments receive fewer calls that never needed a person.
Routing no longer depends on a human receptionist screening each incoming call.
Who this pattern fits
We built it for a retailer, and it suits any organisation with comparable call volumes or coordination needs across several teams. The stack is stable, can be self-hosted where necessary, and can be run by a small team.
Tools used
ElevenLabs
Speech synthesis for the voice agent, chosen for production-grade latency and voice quality.
Twilio
Telephony carrier for inbound and outbound calls, SMS and call recording.
Blind transfer
Hands the call to the chosen department line as part of the routing stack.
Embedded prompt database
Holds company information the agent uses to answer simple queries.
Related
Frequently asked questions
What happens to simple questions like opening hours?
They are answered directly. An embedded prompt database holds company-specific details such as hours, policies and common answers, so the voice system can settle straightforward questions without transferring the caller to a department at all.
How does the system decide which department to transfer to?
It asks the caller what they need in natural conversation and chooses the department from what they say. The call is then handed over with a blind transfer to that department's line, covering every department in this build.
How is this better than a standard IVR menu?
Callers say what they need in their own words rather than working through a menu. That tends to route them faster and avoids the abandonment that makes traditional IVR data hard to trust.