The process

How an agent gets built

01

Start with the person who drops off

the one who can't type, can't read English, or won't call, because that is where people are being lost.

02

Learn what your organisation actually says

your services, your timings, your prices, your rules, so the agent answers the way your own people would, not like a search engine.

03

Decide what it must never say

no advice it isn't qualified to give, no invented prices, no promised outcomes. The list is agreed with you before anything is built.

04

Make it sound like the place it serves

the way people where you operate actually talk, so they answer it the way they'd answer a person.

05

Set the handover

the moment the agent stops and a human takes over: by topic, by question, by tone of voice.

06

Put it where people already are

phone, WhatsApp, or web chat, connected to the systems you already run on.

07

Make every conversation a record

who called, what they wanted, what was agreed, what still needs a person.

08

Test on real calls, then fix it

we listen to what actually happened and tune it. This is the part that decides whether an agent is any good.

What you actually get

Three things a caller will notice.

Not a feature list. These are the differences a person feels on the call, and the reason an agent gets used instead of hung up on.

01 · reachability

It speaks the way your people speak

Someone talks to it the way they'd talk to a person at the desk, in whatever mix of words they normally use, and it keeps up. No form to fill, no English required, no learning how to phrase things for a machine.

02 · trust

It says only what it can prove

AI agents fail by being confidently wrong, and a person then acts on something that was never true. Ours says plainly when it doesn't know, confirms only what actually happened, and passes the caller to a human rather than guessing.

03 · care

It slows down when someone is struggling

An older caller, someone in distress, someone who didn't catch it the first time: the agent gives them room, repeats without impatience, and hands over to a person the moment the conversation needs one.

04 · continuity

Nothing said on a call gets lost

Every conversation ends as a clean record your team can act on: what the person wanted, what was agreed, what still needs a human. Follow-ups stop depending on somebody remembering.

How we think

What you're really hiring.

Tools change every quarter. Judgment is the thing worth paying for, and ours came from twelve agents across six industries going wrong in different ways. Here's what that left us with.

principle 01

The person comes before the software

We start with whoever is on the other end (often anxious, often not typing and often not in English) and design the conversation around them. A dentist's instinct for how people actually talk when something is wrong.

principle 02

Safety is a design input, not a disclaimer

An agent that oversteps is worse than no agent. What it must never say is agreed with you before anything is built, and a human is always one step away.

principle 03

It has to survive a real caller

Anyone can make a demo work. We test against real, messy, interrupted calls, with people talking over it, changing their minds, asking things nobody scripted, because that is what real people do.

principle 04

You own what we build

Every engagement leaves you with something you control: no lock-in, no black box, and an honest answer about whether you should be building this at all.

Where it runs

Wherever your people already are.

Phone calls
WhatsApp
Web chat
Reminders
Follow-up & reactivation
Handover to a person
Start a conversation →