Skip to content

Use case · Phone

Every call answered. Urgent ones put straight through to a person.

An AI receptionist picks up when your team cannot: after hours, at lunch, on the busy line. It says it is an AI, works out what the caller needs, books or answers what it can, and transfers anything urgent to the person on call with a summary already spoken and texted.

Book a Strategy CallSee the step trace

At a glance

Connects to
  • Phone line
  • Speech providers
  • Model provider on your account
  • Your calendar, via its API
  • Job, ticket or CRM system, via its API
  • On-call roster
Typical rollout
Week 1 → Week 5See the plan

The problem

Where the work gets stuck today

Calls arrive when nobody can pick up: after hours, at lunch, while the team is with a customer. Voicemail is where many callers give up and ring the next business, and the urgent ones, a leak, a breakdown, a patient in pain, wait in the same queue as a question about opening hours.

A traditional answering service helps, but it follows a script without access to your calendar or job system. The next morning someone still has to call everyone back to book, check details and decide what was urgent.

  • Missed calls pile up overnight and over weekends.
  • Urgent calls are spotted only when someone listens to the voicemail.
  • Callbacks start with “sorry we missed you” and a second round of questions.
  • The same simple questions take the front desk away from customers in front of them.

Task catalog

What the system takes on

  • Task 01

    Answer on your own number

    Calls are forwarded when nobody answers, when the line is busy, out of hours, or always, depending on the rule you pick.

  • Task 02

    Disclose and ask consent

    The agent says it is an AI assistant in its first sentence and states recording where it is on, using the greeting configured for each market.

  • Task 03

    Triage urgency

    Your urgency words and a short set of risk questions decide what counts as urgent, and the rules lean towards transferring when in doubt.

  • Task 04

    Book, move or cancel

    Appointments go straight into your calendar with a read-back to the caller before anything is written.

  • Task 05

    Answer approved facts

    Hours, locations, services, published price ranges and directions, from an approved fact list rather than whatever the model remembers.

  • Task 06

    Transfer with a summary

    Urgent calls and anyone who asks for a person are warm-transferred, with a spoken and texted summary for whoever picks up.

  • Task 07

    Take structured messages

    Everything else becomes a callback task with the caller’s details, the reason and the transcript, in the system your team already uses.

Step trace

One run, step by step

Sample after-hours call: the AI agent discloses it is an AI and that the call is recorded, asks an electrical-risk question about a ceiling leak, then warm-transfers the caller to the on-call technician with a texted summary.
Sample data
maintenance-line · after hoursSample data
  1. AI agentThanks for calling the maintenance line. I’m an AI assistant, and this call is recorded so the team can follow up. How can I help?
  2. CallerWater’s coming through my kitchen ceiling from the flat above.
  3. AI agentThat sounds urgent. Is the water anywhere near lights or sockets?
  4. CallerIt’s dripping right next to the ceiling light.
  5. AI agentPlease keep clear of that light and switch it off at the fuse box only if you can do that safely. I’m connecting you to the on-call technician now and passing on what you’ve told me.
  6. SystemWarm transfer to the on-call number. Summary sent by text: flat, leak from above, near electrics.
Read the trace as text (Sample data)
  1. 00:00Phone lineStepAfter-hours call forwarded from the main number to the AI agent.
  2. 00:01AI agentStepGreets, says it is an AI assistant and that the call is recorded, as configured for this market.
  3. 00:09TriageCheckLeak through the ceiling matches an urgency rule; asks the electrical-risk question.
  4. 00:16TriageErrorWater near a light fitting: urgent, transfer now. Safety line read from the approved script.
  5. 00:24TransferApprovalWarm transfer to tonight’s on-call technician; the agent speaks a short summary first.
  6. 00:31SMSDoneSummary texted to the technician and a reference texted to the caller.
  7. 00:32Job systemDoneEmergency job created with the transcript, flat and risk flag for the morning team.

Routing map

Five ways a call can go

Every call starts the same way and ends in one of five branches. Pick a branch to see what triggers it, the path it takes, what the caller hears and what lands with your team.
call-flow · routing mapSample data

Call answered · AI disclosed · recording line for this market

Triggered by Your urgency words plus risk questions, such as water near electrics or a patient in severe pain.

Path

  1. Risk questions
  2. Safety line
  3. Warm transfer to on-call
  4. Next person if no answer

The caller hears

“That sounds urgent. I’m putting you through to the on-call technician now.”

Your team gets

A call already summarised out loud, the summary by text, and an emergency job with the transcript.

One turn of the conversation

The pause a caller notices is the sum of four steps. We budget the whole turn, choose providers to fit it, and test it on your real phone line before launch.

  1. 01

    End of speech

    Deciding the caller has finished, without cutting them off mid-sentence.

  2. 02

    Transcription

    Speech to text, tested with the accents and line quality your callers actually have.

  3. 03

    Response

    The model picks the next step from the call flow and drafts a short reply.

  4. 04

    Speech

    Text to a natural voice, streamed so the agent starts speaking before the sentence is complete.

Systems involved

What it connects to

Your existing tools stay the system of record; the AI reads and writes through their APIs.
Phone line
A Twilio number, or forwarding from your existing line on no-answer, busy or out of hours. Callers keep dialling the number they know.
Speech providers
Transcription and voice chosen per language after tests with your callers’ accents, on accounts in your name.
Model provider on your account
OpenAI, Anthropic Claude or Google Gemini to follow the call flow and word replies.
Your calendar, via its API
Real openings for bookings, plus moves and cancellations, written only after a read-back.
Job, ticket or CRM system, via its API
Where messages, emergency jobs and callback tasks land, with the transcript attached.
On-call roster
Who takes urgent transfers at each hour, with the fallback order when the first person does not answer.

Guardrails

  • The agent says it is an AI assistant in its first sentence, on every call.
  • Recording consent rules differ between countries and, in the US, between states. We configure the greeting and the recording setting per market with your counsel’s sign-off, and keep recording off until consent where that is required.
  • Urgency rules are explicit checks in code, written with you, and lean towards transferring when the answers are unclear.
  • No payments are taken by voice, and the agent never asks a caller to read out card details.
  • If a transfer is not answered, the agent tells the caller, tries the next person on the roster, and texts a reference so nothing is lost.
  • If the agent cannot understand the caller after a second attempt, it offers voicemail with a summary or a person, never a loop.

Human checkpoint

Urgent calls and anyone who asks for a person are warm-transferred to the on-call number for that hour, with a spoken and texted summary. Out of hours with no cover, the agent takes a structured message and gives the callback window you chose. Each morning a team member reviews the flagged calls, such as failed transfers, low-confidence understanding and complaints, and the call flows are adjusted from that review.

What we plan for

Typical targets

Typical targets are planning ranges we use when scoping, each with its basis. They are not results from a client project; your own baseline decides the real number.

  • Typical target

    About 0.8–1.2 seconds per turn

    Pause before the agent starts replying

    Basis: Design budget across end-of-speech detection, transcription, model and speech, checked on test calls over your own phone line before launch.

  • Typical target

    Every one, including failed first transfers

    Urgent test calls that reach a person

    Basis: Urgency rules and the transfer fallback chain are tested with scripted emergency calls, including a no-answer from the first on-call number, before launch.

  • Typical target

    Only calls that need a person

    Callbacks left for the next morning

    Basis: Every answered call ends in a booking, an answer, a transfer or a structured message with a callback task; compared with your missed-call log from the month before launch.

Rollout

Week by week

Typical ranges for a first release; the scoping call fixes the plan for your systems.

  1. Week 1

    Call types from your call log, urgency rules, the on-call roster, and the consent and recording setting for each market.

  2. Week 2

    Call flows and the approved fact list on a test number; voice and transcription tests with your team’s and your callers’ accents.

  3. Week 3

    Calendar and job-system connections, the transfer chain with fallbacks, and text summaries.

  4. Week 4

    Scripted test calls, including emergencies and failed transfers, and turn-latency checks on your real line.

  5. Week 5

    After-hours go-live first, daytime overflow next, and the morning review of flagged calls.

FAQ

Frequently asked questions

Can the AI receptionist pick up only when we miss a call?

Yes. Most teams start with conditional forwarding: the call rings your staff first and goes to the AI receptionist on no answer, when the line is busy, or outside opening hours. You can switch to answering every call later without changing the number customers dial.

How does it decide that a call is urgent?

From rules you write with us: words and situations that are always urgent for your business, plus a few risk questions the agent asks when a call is borderline. The rules lean towards transferring, because a needless transfer costs a minute while a missed emergency costs far more.

What do we have to tell callers about recording?

That depends on where your callers are. Some places require you to inform callers, others require every party to agree. We set the greeting and the recording behaviour per market with your counsel, and where consent is required, recording stays off until the caller agrees.

What happens if the on-call person does not pick up?

The agent tells the caller it is still trying, rings the next person in your fallback order, and if nobody answers it takes a structured message, texts the caller a reference and flags the job for first thing in the morning.

Can callers leave a voicemail instead of talking to the AI?

Yes, at any point. The voicemail is still transcribed and summarised, and it lands as a callback task with the caller’s number, so it does not sit unheard in a mailbox.

Will callers notice a delay before it answers them?

We budget roughly one second from the moment a caller stops talking to the moment the agent starts, and test it on your actual line before launch. If a provider pushes the pause beyond that budget in testing, we change providers rather than ask callers to wait.

Can one receptionist cover several locations or phone numbers?

Yes. Each number gets its own greeting, opening hours, fact list and on-call roster, and calls are logged per location, so a multi-site business can share one setup without callers hearing the wrong address.

Next step

Have a workflow in mind? Let’s scope it together.

Book a free 30-minute strategy call. Bring one process you want to automate and we will talk through scope, integrations and a sensible first step.

Book a Strategy CallGet an AI Plan

Free 30-minute call · no commitment

Send an RFP or security questionnaire