Skip to main content
Home How It Works Pricing Case Studies Calculator Blog Book a Call
The Build Method

A voice agent you can actually trust on the phone.

Most voice bots are a demo that never grows up. We build yours like production software — mapped to your calls, bounded by explicit rules, deployed through a tested pipeline, and measured every month.

Anyone can stand up a voice bot in an afternoon. Putting one on the line that answers your real customers — and never makes a promise you can't keep — is a different job.

Four steps, in order: map the calls, build the agent, govern the risk, measure the result. You can scale the depth to your business. Skip a step and you get the brittle demo everyone else ships.

Map. Build. Govern. Measure.

What happens at each step — and exactly what you get.

01

Map

We map every reason your phone rings — and how each call should end.

  • + Call-flow map: intents → desired outcomes
  • + Knowledge base: hours, services, FAQs, policies
  • + Booking rules and calendar targets
  • + Escalation rules: what must reach a human
02

Build

We build and voice an agent that knows your business and sounds like your front desk.

  • + Agent persona, script, and guardrails
  • + Calendar / CRM / scheduler integrations
  • + Your Twilio number connected and tested
  • + Test scenarios run against real call types
03

Govern

We set the limits that make the agent safe to put on your line.

  • + Boundaries: never invents prices or promises
  • + Human-escalation matrix for emergencies
  • + Minimal caller data — the returning-caller record keyed on a one-way hash, deletion on request or within 30 days of the business leaving
  • + Monitoring, and a rollback in minutes if anything drifts
04

Measure

We prove it is working — and keep improving it.

  • + Baselines: answer rate, missed calls, bookings
  • + Targets agreed before launch
  • + Every call graded against explicit criteria — fixes shipped to everyone
  • + Answers checked against your own information, not the agent's confidence
  • + What failed gets reviewed and fixed through the tested pipeline
  • + A dashboard with a summary of every call, searchable — plus every callback and booking, and a recap email (sent after any month the agent booked something) that lists questions callers asked the agent couldn't answer, when there are any
Workflow Topology — Decision Flow Input Process Decision Node Handoff Output Escalation workflow topology
The Stack

Production infrastructure, not a side project.

The reliability of your phone matters more than anything. Here's what's underneath.

Twilio — telephony

Your existing number connects natively to the agent — or we provision a dedicated one for you. Carrier-grade reliability behind millions of business lines, with no change to your phone system. If you ever leave, your number is still yours.

ElevenLabs — voice & conversation

Natural, sub-second speech. Callers can interrupt, change topic, and ask follow-ups — a real conversation, grounded in your knowledge base, not a rigid script.

Agents as code — CI/CD

Your agent's configuration lives in version control. Every change is reviewed and pushed through a tested pipeline before it touches your live line — never hand-edited in a dashboard mid-shift. If something looks wrong, we roll back in minutes.

Disclosure & data care — built in

Every call opens with an AI and recording disclosure. Caller data is minimized. The returning-caller record is keyed on a one-way hash — we never store your callers' numbers in readable form there — and it's scoped to your business alone. Where the business needs to call someone back — a captured lead, a booked appointment — the name and number are kept in readable form for that purpose. They're deleted on request, or within 30 days if the business stops using LinearZeta. Booking performance data holds no caller details at all. Recordings and caller data stay yours.

How it sounds on a call

The first thing most owners ask: will it embarrass me in front of my customers? Here's how every agent is set up.

  • It waits for the caller to finish before it answers, so it doesn’t talk over someone on speakerphone or in the car.
  • It filters out voices in the background — a TV, a passenger, other customers in a busy shop — so it listens to the caller, not the room.
  • The greeting, which says it’s an AI and the call is recorded, plays in full. After that, callers can interrupt it like any conversation.
  • It knows what time it is where your business is and talks in your local time, so it won’t offer “today” after you’ve closed.
  • Prank and time-wasting callers get redirected once or twice and warned once, then the call ends. A caller who is genuinely struggling is offered a person instead.
  • No call runs on forever: every call has a seven-minute ceiling, well past what a typical booking takes.

Live in about two weeks.

From first call to answering yours. See pricing and scope →

Week 1

Map & Build

We map your call flows, gather your FAQs and policies, then build and voice the agent and wire up your calendar.

Week 2

Test & Govern

We test against real call types, set escalation and compliance rules, connect your Twilio number, and dry-run it before go-live.

Ongoing

Measure & Tune

Every call your agent answers is graded against explicit rules, and what falls short gets reviewed and corrected by us. Once a month we walk through it together on a tuning and review call. What you see of it is below.

Included

Measuring it isn't a slide we send you.

You get a login. Everything the agent did is in there, in your own words and your own time zone — not summarized for you by the vendor being measured.

The Calls view of a sample dashboard for a made-up salon: filters for a date range, outcome and a search of what the call was about, an Export this page (CSV) button, and calls 1 to 25 of 131, each with its time, what it was about and an outcome badge such as Booked, Answered or Transferred.
Example dashboard, sample data: a made-up salon's call history.

The number that matters

Appointments booked in the period against the job value you set yourself — so the headline figure is arithmetic you can check, not a model we tuned. Beside it: callbacks still owed, calls handled, and how many arrived after hours or while your line was already busy. Once you've built up about two months of booking history, two more counts join them: new customers (people booking with you for the first time through the agent) and returning customers (people who booked through it before and booked again).

Your whole call history

Filter by date, by outcome, or search what a call was about. Open one to read a summary of what the caller wanted and how it ended — every call, searchable. Export what's on screen. Twelve weeks of trend on the front page.

A callback list, not a spreadsheet

Every caller who needed a human, with their number and what they wanted — and a button to mark them contacted once you've rung back. The callback alert is emailed to you the moment a call ends — and texted, once you opt in.

The controls that are yours

Set your average job value, choose how you're notified, switch on a weekly recap, reconnect your calendar yourself if the connection ever lapses, and manage billing through Stripe's own portal. The agent's own rules stay with us — that's deliberate, and it's what keeps it bounded.

When something goes wrong.

Phones, calendars and software all have bad days. Here's what's built in for the ones we've planned for — what we do, and what your caller hears.

What could go wrong, what we do about it, and what your caller hears
What could go wrong What we do What your caller hears
Your calendar or booking system can't be reached mid-call The agent takes the caller's name, number and what they wanted, and you get a callback alert — emailed, and texted once you opt in. That there's trouble booking right now and someone will call them back to lock in a time. They aren't left hanging.
Someone else takes the slot while your caller is still talking On a Google Calendar or Cal.com line, the agent re-checks your availability instead of booking over them. On other booking systems it takes a message and you get a callback alert. That time just went, and either another opening your calendar actually has or a callback to find one.
The booking step gets sent twice on the same call We check whether this call already booked that slot, and confirm the one booking instead of making a second. One confirmation, for one appointment.
A caller asks for a time they're already booked into On a Google Calendar or Cal.com line, the agent recognizes it as their own appointment, not a clash with someone else's. Their existing appointment read back, and a question: do they want to change it?
Your calendar connection lapses A daily check spots it and sends you a reconnect link — texted if you've opted in to texts, otherwise emailed. You can also reconnect from your dashboard. Until you reconnect, an offer of a callback instead of a booking. Their details still reach you.
A billing problem on your account A billing issue doesn't switch your line off by itself. We flag it and get in touch with you to sort it out. Nothing different. Your agent keeps answering.
A change to your agent doesn't behave the way it should Every change is version-controlled and reviewed before it goes live, so we roll back to the last version that worked in minutes. Once we roll back, the agent that was working before.

What is your phone costing you?

In a few quick questions, estimate the calls — and the revenue — slipping through today.

Estimate Missed-Call Cost →

See it applied to a real business.

Walk through how a local practice replaced its phone tree — call flow, agent, governance, and the numbers.

Read the Example →