Voice and conversational AI

Agents that take the call, and systems that read every call your people take.

Two things, on one discipline. A voice agent that answers the line at any hour, follows a caller who switches from Hindi to English mid sentence, and has the ticket written before they hang up. And a call analytics system that reads every call your team takes, scores both sides, and hands them a review queue in risk order instead of a sample.

See it in our work

At a glance

Languages
Hindi, English and Hinglish in production
Channels
Phone, WhatsApp and chat
Recordings
In your cloud and region
Fallback
A person, whenever the caller asks

What it covers

On the line, and behind every call.

  • Real conversation handling

    Barge in, pauses, interruptions, and language switching inside one sentence. The difference between an agent people stay on the line with and one they hang up on.

  • Phone, WhatsApp and chat on one core

    A ticket from a call and a ticket from a chat are the same object. One set of rules, one log, one place to look.

  • Warm transfer with context

    When a person is needed, the agent hands over to a named person with the conversation carried across, along an escalation chain with a fallback.

  • Every call kept with its record

    Recording, transcript and summary attached to the ticket, read by the person who resolves it. Their corrections are the accuracy measurement.

  • Every call read and scored

    A system that reads every call your team takes, on the customer side and the adviser side, and ranks them by risk. A sample only finds what you suspected; reading everything finds the call nobody thought to open.

  • Checks in plain English, coaching from the call

    A new quality or compliance check is one sentence, deployed once and run on every call from then on. Each score is pinned to the moment in the transcript that produced it, so coaching points at the exact thing.

How we do it

Tuned on real calls, rolled out in stages.

  1. 01

    Sit in on the line first

    We listen to how support or sales actually runs before designing anything. The opening line, the detours, the moment people give up, the check that only a person would catch.

  2. 02

    Adaptive, not scripted

    The agent decides what to ask next from the whole conversation, the way a trained support person does, rather than walking a fixed tree that cannot reconsider.

  3. 03

    Your QA becomes the checks

    For call scoring, your own scripts and procedures come across as plain English instructions, marked three ways: said word for word, covered in your own words, or left to the adviser. Your QA lead reviews calls beside the system and tells us where it is wrong, with the reason.

  4. 04

    Staged rollout, tuned on real calls

    Internal staff first, then a slice of real callers, then the whole line. Every stage is judged on real calls, and the corrections people make feed tuning from launch day on.

  5. 05

    A flag is a candidate, not a verdict

    Every score carries its evidence. A person confirms it, the team keeps its own judgement, and nothing is reported to a regulator or anyone else on its own.

What you get

What goes live on your number.

  • A voice agent on your existing number, and the same agent on WhatsApp or chat
  • Ticket creation and routing into your systems, with status updates to both sides
  • Warm transfer to named people, with a fallback
  • Call analytics on your recordings: every call scored on both sides, a risk ordered review queue, a quality index per adviser
  • Recordings, transcripts and summaries in your own cloud and region, personal information stripped before any model sees them
  • An accuracy measure built into the resolution or review workflow

Where we have done it

Real engagements, named.

Questions

Questions we get asked.

We do not want a voice agent. We want to understand the calls we already take.
Then it is the call analytics system on its own. It connects to your recordings, reads every call, applies your checks, and gives the team a queue in risk order. No agent on the line, and nothing changes for your callers.
Which languages?
Hindi, English and Hinglish in production today, with Sarvam for speech. Other languages depend on the speech models available for them, and we test before we promise.
What happens when the caller refuses to talk to a machine?
They reach a person, and the query is logged either way. Nobody is trapped with the agent.
Is call scoring a compliance tool or a coaching tool?
Both, and the client usually decides which matters more once they see the first results. Life Insurance Direct moved the brief from whether an adviser crossed a line to whether they stopped dead at it and lost a customer. The checks were rebuilt around that.
What does it run on?
Plivo or your telephony for the line, LiveKit for the live conversation, Sarvam for hearing and speaking, Claude for deciding what to ask next, WhatsApp Cloud API for chat. Call analytics runs in your cloud account and region, with personal information stripped before any transcript reaches a model.

What clients say

In their words.

01 / 04

Life Insurance Direct
Before Metricsense, checking calls was a manual process: we selected files for review. Now the platform analyses every call, and we review by risk. It has identified customer experiences we would not have picked up before, and it lets us scale our quality assurance program without increasing headcount, with our team still the human in the loop. It helps us protect our customers, our business and our agents.

Russell Cain

Founder and CEO, Life Insurance Direct

Also under AI engineering

Where it leads next.

All of AI engineering

Not sure this is the right place to start?