Skip to content
UVS

AI Voice Agents

Build

The call gets answered. On the first ring, at 2am.

Voice agents that answer, qualify and book against live availability — and hand a warm call to a person the moment the conversation needs one.

Typical stack

  • Deepgram
  • ElevenLabs
  • Claude
  • Twilio
  • LiveKit
  • PostgreSQL
  • Node.js

Written for: An owner-operator losing revenue to unanswered phones and unable to justify a night shift.

Who it is for

Is this built for a business like mine?

Built for businesses where the phone is the front door.

These are businesses where demand arrives by voice, arrives unpredictably, and goes to whoever picks up first. If that is not your business, a voice agent is an expensive novelty.

What we see

A burst pipe at 11pm calls three companies. The one that answers gets the job, and yours went to voicemail.

What we build

An agent that answers immediately, triages emergency from routine, and books against real technician availability and drive time.

The detail that matters

It will not book two jobs an hour apart across town — capacity is modelled, not assumed.

What we see

Reception is on the phone while three patients wait at the desk, and the no-show rate nobody owns sits above ten percent.

What we build

Booking, rescheduling and confirmation handling, plus recall calls that actually go out on schedule.

The detail that matters

Built to the data boundary you operate under, decided during discovery rather than after.

What we see

Service bays are a fixed capacity sold by the hour, booked over one phone line by an advisor who is also handing over keys.

What we build

Service booking that respects bay capacity, technician skill and parts availability at once, not just an open calendar slot.

The detail that matters

Three constraints on one booking, or it books work that cannot be done.

What we see

The phone rings through dinner service, and the booking that was not answered becomes an aggregator booking with commission attached.

What we build

Direct reservations taken against live PMS availability, plus group and event enquiries captured with the detail needed to quote.

The detail that matters

Availability is read live. A reservation for a room that is gone is worse than a missed call.

What we see

A portal enquiry lands while every agent is out showing property, and its value has largely decayed by the time anyone calls back.

What we build

Immediate answering that qualifies the enquiry, books the viewing against agent availability, and logs it before the agent is back in the car.

The detail that matters

Qualification is tuned not to gatekeep — losing the buyer who was worth the call costs more than an extra viewing.

And who it is not for

If your calls are long, consultative and different every time, a voice agent will frustrate your callers. We would rather staff that line with people — which we also do.

The problem

Do they understand what is actually going wrong?

Voicemail is a competitor’s lead-generation tool.

The economics are unforgiving: an inbound call is the highest-intent contact a business receives, and its value collapses to nothing within minutes. Yet answering every call means staffing for a peak that happens twice a week, at hours nobody wants, for conversations that are mostly the same four questions.

  • Calls outside office hours go to voicemail and are returned after the job has been booked elsewhere.

  • Peak periods produce four times the volume and the same number of people.

  • The generic answering service takes a message, which is not the same as taking a booking.

  • Nobody can say how many calls came in last Tuesday, or how many were missed.

What we build

What exactly would I be buying?

It answers, and it can actually finish the call.

Taking a message is not answering the phone. A voice agent is only worth having if it can complete the transaction the caller rang about — which means live availability, real qualification, and a clean route to a person when the conversation stops fitting the script.

Conversational, not a phone tree

Callers speak normally, interrupt, change their mind and go off-script. The agent handles that, because real callers do it constantly.

Booking against live capacity

Connected to your calendar, PMS or dispatch system, so what it offers is genuinely available and what it books is genuinely doable.

Handoff by rule

Transfer happens on conditions you define — call type, caller value, an explicit request — not on a confidence score the model calculates about itself.

Every call in writing

Recording, transcript, extracted details and outcome, in your CRM before the caller has put the phone down.

How it works

How does this actually function?

Sixty seconds, start to booked.

Latency is the whole experience. Above roughly a second of silence a caller assumes the line dropped, so the pipeline is built around that budget rather than around a feature list.

  1. Ring

  2. Greet

  3. Understand

    Transfer to human
  4. Check availability

  5. Book

  6. Confirm

Speech, reasoning and response are streamed and overlapped so the caller hears a reply, not a gap.

Availability is queried live. Nothing is offered that cannot be honoured.

A transfer carries the transcript so far, so the caller never repeats themselves to the person who picks up.

What changes

What is different afterwards?

What is different afterwards.

The after-hours call converts

The calls that previously became voicemail become booked jobs, and those are usually the highest-urgency calls you receive.

Booked work originating outside staffed hours.

Peaks stop being a staffing problem

Concurrent calls are answered simultaneously, so a storm week or a campaign spike does not need a hiring plan.

Concurrent calls answered without a queue.

The phone becomes measurable

Volume, outcome, duration and reason for every call — the channel that was previously invisible becomes the one you can actually manage.

Call reason and outcome, categorised, for every call.

How we deliver

How does this start, and what do I get at each step?

Six stages, and every one has an exit.

You can stop after any stage with something useful in hand. That is the point of naming the artefacts rather than the activities.

  1. 01

    Call audit

    We listen to your actual calls. Every voice agent that fails does so because it was designed from an imagined conversation rather than a real one.

    • Call type taxonomy from real recordings
    • Volume and timing profile
    • The call types we recommend never automating
  2. 02

    Conversation design

    The flow per call type, the qualification criteria, and the exact conditions that trigger a transfer.

    • Conversation flows
    • Transfer rules, written by you
    • Escalation and out-of-scope paths
  3. 03

    Integration

    Calendar, dispatch, PMS, CRM — whatever holds the availability and whatever needs the record.

    • Live availability integration
    • CRM write-back
    • Telephony configured
  4. 04

    Voice tuning

    Latency, interruption handling, accent and noise robustness, tested on real audio rather than in a quiet room.

    • Latency measured under load
    • Interruption handling verified
    • Voice selected with you
  5. 05

    Live pilot

    One call type, out of hours first, monitored closely with a person able to take over at any point.

    • Pilot report with recordings
    • Transfer rate by call type
    • Tuned transfer rules
  6. 06

    Rollout

    More call types and more hours, widened on pilot evidence.

    • Rollout plan
    • Monitoring dashboard
    • Runbook and fallback to human line

Questions

But what about the thing that worries me?

The questions people actually ask.

Will callers know it is not a person?

Many will, and the agent identifies itself if asked — we do not build agents that claim to be human. What callers consistently care about more is whether the call got resolved on the first attempt, which is the thing voicemail never does.

What happens when the caller asks something unexpected?

It transfers, on a rule you wrote rather than a guess. The person who picks up receives the transcript, so the caller does not start again.

Can it actually book into our calendar?

Yes, against live availability, and it books only what your capacity rules permit — technician skill, drive time, bay availability, room type. A booking that cannot be honoured is worse than a missed call, so those rules are modelled explicitly.

What does it cost to run?

Per minute of conversation, and the number depends on call length and complexity. We size it against your real call volume during the audit, so the comparison against a staffed line is arithmetic rather than a promise.

What if it goes wrong at 3am?

Calls fall back to your existing line automatically if the agent is unavailable. Failure means callers get what they get today, never a dead line.

The other half

Someone still has to take the transfer.

A voice agent set up honestly transfers more calls, not fewer — the ones needing judgement, the upset caller, the deal worth a person. That only works if a trained human is reachable when it happens. We run that line as well, which is the reason we can set transfer rules at the right threshold instead of the flattering one.

Voice Calls