A conversational AI IVR replaces "press one for sales, press two for accounts" with a voice that asks the caller what they need and acts on the answer. For a company of 10 to 250 people, the result is fewer abandoned calls, fewer misrouted calls and a written record of every call on the contact. This article shows how to turn a list of call reasons into a routing table, set up a voice agent on the platform to answer your main line, and decide which calls it should resolve, book or transfer.
Why keypad menus lose callers
A traditional IVR is designed around your departments, not around the caller. Someone chasing a late invoice does not know whether that is accounts, operations or their account manager. They press the wrong number, wait, get transferred and hang up. Existing clients learn the shortcuts. New enquiries, the callers you most want, do not know your structure and suffer most.
A conversational system reverses the burden. The caller says "I'm chasing an invoice" and the agent decides where that goes. It can also answer simple questions itself, so many calls never need a transfer at all.
Resolve, book or transfer: the routing table
Everything the agent does on the phone comes down to three outcomes. Build the table before you touch any settings.
- Resolve: questions with a fixed answer, such as opening hours, the address or the documents to bring.
- Book: calls that should end in a meeting, such as a first consultation, a valuation or a viewing.
- Transfer: anything that needs a named person or team, with the hours that team is reachable.
Take the reasons from a week of call notes or recordings, not from memory. For each one, write the destination and two or three example phrases callers actually use, such as "I want to sell my flat" under sales.
Before you start
- A phone number on the platform, either a new number or one you port across.
- The routing table above, with transfer numbers or users for each group and their working hours.
- Calendars for anything callers can book.
- The facts the agent needs to resolve simple calls, ready to load as knowledge.
- A short greeting and the voice you want callers to hear.
How to set it up
Create the voice agent and give it the routing table as part of its prompt:
- Go to AI Agents > Voice AI and open the Agent List tab.
- Click + Create Agent.
- Select Create Custom Agent, then click Continue.
- Enter an Agent Name and select a Voice.
- Select a Model and Language, and set the Timezone.
- Write the agent prompt: role, limits and objective.
- Configure the Welcome Message for inbound and outbound calls.
- Click Save.
- Open Knowledge Base and select the knowledge base.
- Open Actions, click + New Action and select an action type.
- Name the action, define its trigger condition and save it.
Add one Call Transfer action per team in the transfer column, an Appointment Booking action for each calendar in the book column, and a Send SMS or Trigger a Workflow action if callers should receive a confirmation. Test with Web Call and Phone Call from the Test panel until each route behaves, then put the agent on the line:
- Open the agent's Deploy tab.
- In the Phone section, select a phone number or Number Pool.
- Under Call Routing, choose Answer Calls Directly or Use as Backup.
- Under Working Hours, choose All Hours or Custom Schedule.
- Review forwarding, timeout, voicemail and fallback for the number.
- Save the deployment settings.
Choosing Use as Backup lets your team answer first and the agent pick up what they miss, which is a gentle way to start. Point the main number at the agent for a day, listen to every call and adjust the table before going full time.
What callers experience
The caller hears a short greeting and an open question. They answer in their own words. The agent confirms what it understood, then answers, books or transfers. After the call, a summary and the transcript sit on the contact record. That is the part a keypad menu never gave you: you can see why people called, which routes are busiest and which questions the agent could not answer.
Say the main line takes 1,200 calls a month. Under a keypad menu, suppose a large share of callers choose the wrong option and need a second transfer, and around 40 calls a week ask only for branch hours or a viewing slot. With a voice agent, callers would state the branch and the reason. Hours questions get answered directly, viewings are booked into negotiators' calendars and vendor calls reach the right negotiator with a summary. Those 40 weekly calls, around 160 a month, would no longer need a person at all.
Replacing the menu without a contact-centre project
Searches for vendors such as Omilia conversational AI usually come from the same need: let callers say what they want instead of pressing numbers. This article does not cover any vendor's product. For a firm your size, the job needs a voice agent that understands open speech, can transfer and book, and writes the result into your CRM. If you need queues for dozens of seats and wallboards, a dedicated contact-centre product may fit better; conversational AI for contact centres explains where that line sits.
What it costs you to run
Call minutes and AI voice usage are billed as used on the platform. The time cost is review: about two hours in the first week, then twenty minutes to an hour a week reading misrouted calls and unanswered transfers. Revisit the routing table whenever you add a service or change team structure, because an outdated table sends calls to people who no longer handle them.
Teams often copy their old menu into the agent: "Say sales, accounts or support." That rebuilds the keypad problem with voice and adds delay. Ask one open question and keep the routing logic on your side.
When a conversational IVR is the wrong choice
If you take very few calls, a person answering is cheaper and warmer. If your calls are mostly urgent, such as an emergency maintenance line, put a person first and use the agent only out of hours. If most inbound contact already arrives by chat or text, start there; the AI call centre guide covers how voice and text agents share the work.
Troubleshooting
Callers are sent to the wrong team. Two call reasons overlap or the table lacks the phrases callers use. Give each route a distinct description and add example phrases from transcripts.
Transfers ring out. The working hours you gave the agent do not match reality. Fix them, and tell the agent to take a message or book a callback outside those hours.
The agent transfers too often. The prompt does not permit it to resolve anything. List the questions it may answer outright.
Callers ask for a person straight away. Let them. Add "speak to a person" as a route that transfers immediately; forcing callers through questions creates complaints.
Summaries are too long to read on a phone. Ask for three lines: who called, why and what was promised.
Frequently asked questions
Can a conversational AI IVR keep my existing number?
Usually, by porting the number to the platform or forwarding it to a platform number. Check timing with your current provider and keep the old service live until the port completes.
Does it handle accents and background noise?
Modern speech recognition copes well with most accents; noisy lines cause more trouble. Have the agent confirm what it heard before acting on anything important.
Can it answer out of hours only?
Yes. Use a custom schedule or the backup routing option so your team answers during office hours. See AI phone answering and the AI phone receptionist guide for that pattern.
How do I know the routing is working?
Read the transcripts of transferred calls each week and check that the receiving team was the right one. For voice agents in lead handling, read AI call center agent.