Service · AI voice agents

Voice agents that answer the call and write up what happened

Calls outside office hours go to voicemail, or to a competitor. The agent answers, holds a conversation, captures what the caller needs, and puts a qualified record in your CRM before the next working day starts.

The problem

Voicemail is where enquiries go to die.

A call that arrives at seven in the evening, or while both people who could answer it are already on calls, has two outcomes: a voicemail nobody returns quickly enough, or a caller who rings the next firm on the list. Neither shows up in a report, which is why the cost is easy to miss.

Reception is also expensive to scale for peaks. Most inbound-heavy businesses are staffed for the average and quietly lose the spikes.

Fit

Who this is for.

Inbound-heavy service businesses where a first conversation is mostly qualification: lettings and property, clinics and practices, trades and field services, professional services. The questions are known in advance, and the caller mainly needs to be understood and routed.

Not a fit if most callers are existing clients who need one named person, or if the first conversation genuinely needs professional judgement. Those calls should reach a human, and the agent should be told to route them there.

In practice

What gets built.

Answering and qualification

The agent answers, asks your qualifying questions, and holds a conversation rather than reading a menu.

Structured capture

Caller details and intent captured as fields, not as a recording somebody has to sit through.

CRM write and summary

The record is written and the summary reaches the right person, ready for the morning.

Escalation and handover

Rules for the calls that need a person, and a clean transfer when they do.

Deliverables

What you get.

  • A voice agent trained on your services, tone and qualifying questions
  • Structured capture of caller details and intent, not a recording to listen to
  • CRM write and call summary sent to the right person
  • Escalation rules for the calls that genuinely need a human
  • A documented call flow your team can change without a developer
  • Documentation written for handover

Design decision

The handover is the product

Most of the work in a voice build is not the talking. It is deciding what the agent must not attempt, and making the exit clean when it hits that boundary. An agent that improvises past its limits costs more than the calls it saves, so the escalation rules are agreed before the conversation design starts.

Integrations

What we have built against.

Vapi and ElevenLabs for the voice layer, a call summary webhook into your systems, n8n for the surrounding workflow, and CRM write-back where the CRM has an API. Calendar and email delivery where a callback or confirmation is part of the flow.

Delivery

How the build runs.

Discovery

We listen to what callers actually ask and agree what the agent must never attempt.

Call design

Qualifying questions, tone and escalation rules written down before anything is built.

Build

One call flow, with capture, summary and the handover path tested.

Handover

Your team edits questions and rules, and reads the records, without us.

Scope
One call flow, fixed
Typical timeline
2 to 3 weeks
Ongoing
Retainer, optional
Ownership
Yours, documented

Timelines depend on the number of call types and how settled the qualifying questions are. Ongoing work runs as Integration Cover: a monthly block of 10, 20 or 40 engineering hours on a three-month minimum, with overflow agreed in advance at the tier rate.

Oversight

Data handling, and what happens when something is wrong.

  • The agent introduces itself as an assistant rather than implying it is a person
  • Escalation rules decide what it must hand to a human
  • Call records are structured and reviewable, not just audio
  • Caller data goes only where you already send it
  • If the integration fails, the call falls back to your existing route rather than dropping

Proof

Hear it before you talk to us.

This system's public proof is a demonstration scenario rather than a client case study. It runs the full call flow, so you can judge the handling rather than take our word for it.

Questions

What buyers ask.

Will callers know they are not speaking to a person?

The agent introduces itself as an assistant. Pretending otherwise damages trust the moment a caller notices, and callers usually do.

What happens when it cannot answer?

It hands over according to rules you set: transfer to a person, take a message, or book a callback. An agent that cannot hand over cleanly is worse than voicemail, so the handover is designed before the conversation is.

Is it only for out-of-hours calls?

No, though that is the common starting point because the cost of a missed call is clearest there. It can also take overflow during the day when everyone is already on a call.

What does it write into the CRM?

Structured fields rather than a recording: who called, what they wanted, what was agreed, and what happens next. The point is that the next person does not have to listen to anything.

Does it work with our phone system?

It needs a number that can route to it, which most modern systems and providers support. We check this in discovery rather than assuming it, because it is the one thing that can block a build.

Who owns the call flow afterwards?

You do. The flow is documented and editable, and the handover covers changing questions and escalation rules without a developer.

Tell us what a missed call costs you.

Send the call types you get and what currently happens to the ones nobody answers. If your calls mostly need a specific person rather than a first response, we will say so.

Back to all four systems · About the studio