Most of what happens between a customer asking and a customer paying is waiting. An enquiry that sits until Monday. A quote that goes out on Thursday. A follow-up that never happens. A review that is never asked for.
Agentic AI automation is the work of building systems that act inside those gaps, within limits a person set, and hand over the moment a person is needed — removing the waits without removing the judgment. This page explains what that means, what we build, and what we will not do.
Walk one customer through a business from the first contact to the money arriving, and count the places where nothing is happening because nobody is free to do it. Those places are not the same in every business, but they are nearly always the same kind of place.
| The wait | What it costs |
|---|---|
| The enquiry | A call at twenty to seven, a form on a Saturday, a message during a job. The customer is usually asking more than one business, and the first to answer properly is usually the one that gets the work. |
| The follow-up | The second and fifth contact that a good salesperson would make and a busy one does not. A quote that was never chased is a sale that was never lost, because it was never counted. |
| The booking | The back-and-forth to find a time. Every exchange loses somebody, and the one who leaves is the one who was least sure. |
| The reminder | The no-show and the late cancellation. A slot you paid to fill, empty, with nobody offered it. |
| The hand-off | The job written down twice, the address wrong once, the note that lived in a text message. Rework, and the time of the person who does the rework. |
| The invoice | Sent late, chased never. Cash arrives when it arrives, and the business borrows to cover the gap. |
| The review request | Asked never, or asked a month later when the customer has moved on. The evidence that fills your third surface — the reviews, mentions and accounts an AI reads — is not there. |
| The report | The owner reading three dashboards on a Sunday night. Decisions made on last month's numbers, or on a feeling. |
Every one of these waits is small. Together they are the reason a business with plenty of demand still feels short of it, and the reason growth means hiring before it means earning. They are also, nearly all of them, the same shape: something arrived, and a decision that a person could make in a minute did not get made for a day.
This page is written for businesses that take enquiries and book work: home services, healthcare practices, professional services, and online sellers. The systems are built for businesses locally, regionally, nationally and internationally; the waits are the same everywhere.
Scripted automation follows a rule. If a form arrives, send this message. It is reliable, and it is only as good as the rule, and the first enquiry the rule did not anticipate is the one it handles wrong.
An agent reads the situation. It takes the message, the record, the calendar and the instructions it was given, decides what to do within limits a person set, does it, and hands over when the situation is outside those limits. The difference is not that it is clever. The difference is that it can handle the case nobody wrote a rule for, and it knows when it should not.
That second half is the whole discipline. An agent that handles everything is an agent that will eventually handle something badly, in your name, to a customer. The limits, the hand-over and the record of what it did are not features added to the automation. They are the automation.
It is not a person, and it does not pretend to be one. It is not a replacement for the judgment that runs the business — which of these customers matters, which job to take, what to charge. And it is not a product installed in an afternoon. The afternoon version exists, and it is the one that answers the phone wrong on the first Saturday, because nobody mapped the flows, wrote the limits or read the transcripts.
We deploy automation where it changes what a business earns, and that is at three points in the life of a customer, plus the operations that connect them.
Acquisition
The demand you generate is answered fast and consistently, at any hour, on every channel. Enquiries are answered, qualified and booked while the customer is still deciding, and the record is written before anybody has to remember it.
Retention
The revenue you have already paid to win stops leaking. Reminders, confirmations, rescheduling, the follow-up after the job, the prompt when the customer is due back, and the answer to the question that comes up every week.
Advocacy
The customers who are happy say so somewhere it counts. Every customer is asked for a review once, at the right moment, in the way the rules allow, and the case study, the testimonial and the referral are asked for the same way.
Operations
The hand-offs between the tools you already run. Enquiry to record to job to invoice to payment to review request, without a person retyping any of it, and the week's numbers read and reported without a person compiling them.
The third point is the one most automation firms do not build, and the one we would build first. Advocacy fills the evidence an AI reads when it decides whether to recommend you. Being recommended is the front of the system; the automation is the back. We do not sell them as separate things because they are not.
Each of these is a system, not a feature: a set of conversations, actions and limits, built on the tools you already use and run against your real traffic before it is trusted. Most businesses need three of them. Few need all of them.
Every enquiry — call, text, form, chat, email — is answered within the minute, at any hour. The reply asks the questions you would ask, answers the ones the customer has from your own information, and writes the whole exchange into your records. When the customer wants a person, a person is told, with the context.
Calls answered when the office cannot: questions answered from your services, hours and prices as you have written them; appointments booked into the calendar you use; messages taken with the detail that matters; and a transfer to a named person for anything it is not allowed to handle. It says it is an AI when asked.
The second, third and fifth contact after a quote or a visit, spaced the way a good salesperson would space them, in your voice, ending when the customer answers, books or says stop.
The old enquiries that went quiet, the quotes that were never answered, the customers who are due back. Contacted only where they agreed to be contacted, with an offer that is true, and handed to a person the moment they reply.
The appointment confirmed, the reminder sent, the cancellation offered to the next person on the list, the reschedule done without a phone call, and the record updated each time.
Form to record, record to job, job to invoice, invoice to payment, payment to review request. Built between the systems you already run, so nothing is retyped and nothing is lost between them.
An AI worker with one narrow job: sorting the inbox, drafting the quote from the job notes, filing the document, checking the order against the record, preparing the report. It reads only what it was given permission to read, every action is logged and can be undone, and anything it is unsure of goes to a person for approval.
For online sellers: the abandoned cart and checkout recovered, the order and delivery status sent as it changes, the question answered from the store's own information, and the review asked for after delivery, all tied to the store's real events rather than a timer.
Every customer, once, when the outcome is fresh, with one link to the platform your trade uses. Nobody is asked selectively, nobody is offered anything for it, and nobody's words are written for them, because the platforms' rules and a federal rule prohibit all three.1 4
The week's enquiries, response times, bookings, no-shows, unpaid invoices and reviews, in one message on Monday morning, with the exceptions named and the trend against last month.
For a business that is not ready to build: the opportunities scored against what they would earn, what they would need and what could go wrong, in an order, with a first pilot designed and a point at which to stop. Written so that you can stop there.
Working sessions on real tasks with the AI tools you approve: what to use them for, what to check before trusting the output, what never to paste into them, and a short guide for each role to keep.
The order matters, and each stage produces something you keep. A first system is typically live between two and eight weeks after the map, depending on how many tools it crosses and how clean the records are; the map names the weeks. We design, run and stand behind every system. Where a build needs a specialist, we bring one in and direct the work; you deal with us.
We sit with the people who do the work and trace every enquiry, booking, job and invoice from first contact to money in. What you get is a map of the flows with the waits marked and costed, and a list of what is worth automating and what is not. The list is usually shorter than expected, and the top three items carry most of the value.
For each system, written down before anything is built: the conversations it will have, the questions it asks, the information it may use, the actions it may take, the limits it works within, and the moment it hands over and to whom. You read it and sign it off. It is the contract the agent is built to.
On the telephony, messaging, calendar and record-keeping tools you already run, wherever they will carry it, so nothing moves and nobody learns a new system. Where they will not, we add the smallest thing that will, and we tell you what it costs to run.
Against real traffic, in the open. Test calls and replayed enquiries first, then a period in the shadow of the person who does the job now, with every transcript read and every miss fixed. It earns trust before it is given any, and it goes live one flow at a time.
Every conversation logged. The hand-offs and the exceptions reviewed. The answers it gives updated as your services, hours and prices change. A review with you every month of what it handled, what it handed over and why, and what to change.
Every measure on this page taken before the work begins, and again on a schedule, by the same method. A number that moves is the report. A number that does not move is also the report.
Before it is built, we agree one measure from the list on this page and take it. The pilot runs long enough to take it again by the same method. You see both numbers, and you decide whether to go on. You can stop there, and what was built stays yours.
The question every owner asks first is what the AI is allowed to do on its own. The answer is written into every design, and it starts from one rule.
Nothing that commits money, a date or a promise goes out without a person named to confirm it.
Answering a question from your own information, asking the questions you would ask, booking an open slot, sending a confirmation, writing to the record: automatic. A quote, a discount, a refund, a change to a booked job, a complaint, a customer asking for something outside what it was designed to do: it stops, and a named person decides.
The hand-over carries the whole context — what was asked, what was said, what it found in the record and what it was about to do — so the person picks the conversation up mid-sentence rather than from the beginning. Nothing is worse for a customer than explaining themselves twice, and a hand-over that makes them do it has failed.
The AI says it is an AI when it is asked, and where a rule requires it, before it is asked. It does not use a person's name as its own, and it does not claim to be in the office. Customers do not mind talking to an AI that is useful and honest about it. They mind being fooled, and they remember who fooled them.
Consent is a limit like any other. Calls and texts go only to people who agreed to receive them, inside the hours the rules allow, from a number registered for the purpose, with a way to stop that works the first time.2 3 The registrations, the consent records and the opt-outs are part of the build, not an afterthought to it.
An automation that lives in somebody else's account is a rental, and the owner finds out the day the vendor is bought, closes or changes its terms. So the ownership is written down.
The workflows, the conversation designs, the instructions the AI works from and the limits it works within are yours, in accounts you control.
The phone numbers, the messaging registrations and the consent records are registered to your business, not to us.
Every transcript, every record and every measurement is yours. None of it is used to train anything, and none of it is shared with another client.
The map, the designs, the owner's instructions, and a plain list of what to watch and what to do when something looks wrong.
If the engagement ends, it keeps running. The costs of the platforms it runs on are yours, at cost, and were visible to you from the first month.
The market sells automation with numbers nobody can check. We take these before the work begins, and again on a schedule, same method, and report whichever way they move.
| The measure | What it tells you |
|---|---|
| Response time | How long a customer waits for the first useful reply, by channel and by hour, including the hours the office is closed. |
| Answer rate | What share of calls are answered at all, and what share of those reach a booking, a message or a person. |
| Booking rate | Of the customers who could be booked, how many were, and how many exchanges it took. |
| Follow-up coverage | How many quotes and visits received every follow-up the design calls for, and how many stopped short. |
| Hand-off rate and reasons | How often the AI passed a conversation to a person, and why. A rising rate is not a failure; a rising rate for a reason we could design out is. |
| No-show rate | Booked slots that went unfilled, before and after reminders and offers to the waiting list. |
| Time to invoice, time to paid | The days from job done to invoice sent, and from invoice sent to money in. |
| Review velocity | Whether reviews arrive on a steady schedule, and what share of customers asked left one. |
| Hours returned | The time your people no longer spend on the work the systems handle, estimated from the map and checked against the record. |
| Cost per handled conversation | What each answered call, reply or booking costs to run, all platforms included, against what it cost before. |
None of these figures appears on this page yet, because none has been measured by us across enough businesses to be stated as a rule. When one has been measured twice by the same method, it will be here, with its method beside it.
The market for this work is full of numbers in headlines. Here is what we will not say, and why.
Nobody can guarantee a booking rate, a response time or a saving before measuring you.
The figure depends on your enquiry volume, your hours, your tools and how your people work today. We measure first, and we report what moved.
We will not promise a date before the map.
The ranges on this page are typical, and they are on the page because they are typical. The map is what tells us how many flows there are, how many tools they cross and how clean the records are, and it is what turns a range into a plan with weeks in it. We hold to the plan, not to the range.
We do not quote a price before the map.
What this costs depends on how many systems are worth building, what they must connect to and what they cost to run. A number given before that is a guess dressed as a quote, and the assessment exists so that nobody has to guess.
We will not build an agent that pretends to be a person.
It says it is an AI when asked, and where a rule requires it, before. A customer who is fooled once does not come back, and the rules are moving in one direction.
We will not skip consent.
No call or text to anyone who did not agree to receive it, none outside the hours the rules allow, and none from a number that is not registered for the purpose. If a list cannot show consent, it is not a list we will use.
We will not automate what we are not qualified to run.
Bookkeeping and tax, clinical scheduling that touches health records, legal intake that touches privilege, and hiring decisions each carry obligations that belong to a licensed or specialist practice. We say so, and we will point you to who does it properly.
Automation firms build the machine and leave. Marketing firms generate the demand and do not see what happens to it. The enquiry the visibility work produces is the enquiry the automation answers, and the review the automation asks for is the evidence the visibility work needs. They are one system, and a firm that builds only half of it cannot measure either half honestly.
Doing both under one roof means the questions the AI answers on the phone are the questions we already know your buyers ask, the records it writes are the ones your listings and your site are built from, and the review it requests lands where the systems that recommend you will read it. It also means that when a number moves, we know which half moved it.
If you have no enquiry volume to lose — if the work comes through one contract, one referrer or one channel that does the answering for you — there is no wait to remove, and we will say so. If your team already answers within the minute, follows up every quote and asks every customer for a review, the map will show it and we will tell you what, if anything, is left.
If you want the person removed rather than the wait, this is not that. If you want an AI that pretends to be your receptionist by name, or messages sent to people who never agreed to receive them, we are not the firm, and the rules we have cited say why.
Automation in which an AI reads the situation — the message, the record, the calendar — and decides what to do within limits a person set, rather than following a fixed rule. It acts on its own inside those limits and hands over to a person outside them. The limits, the hand-over and the log of what it did are what make it safe to use in a business.
Phone answering is one of the systems we build, and for many businesses it is the first. It is not the whole of the work. The same discipline covers the follow-up after the call, the booking, the reminder, the invoice and the review request, and most of the value sits in the parts a receptionist never touched.
Yes, if they ask, and before they ask where a rule requires it. It does not use a person's name as its own or claim to be in the office. Customers are comfortable with an AI that is useful and honest about what it is; what they do not forgive is being fooled.
It stops and hands over to a named person, with the whole context: what was asked, what was said, what it found in the record. Nothing that commits money, a date or a promise goes out without that person confirming it. The hand-off rate and its reasons are one of the things we measure every month.
Usually not. We build on the telephony, messaging, calendar and record-keeping tools you already run wherever they will carry it. Where something is missing we add the smallest thing that will do the job, and we tell you what it costs to run before it is added.
You do. The workflows, the instructions, the phone numbers, the registrations, the data and the documentation are yours and sit in accounts you control. The systems keep running; the platform costs were always yours, at cost.
A first system is typically live between two and eight weeks after the map: follow-up, reminders and reactivation at the short end, phone answering and desk-work agents at the long end. We will not promise a date before the map, because the map is what tells us how many flows there are, how many tools they cross and how clean the records are. Once it exists, the plan names the weeks and we hold to them.
It is free. You get the map of your flows from first contact to money in, the waits marked, the three systems worth building first and what each would need to connect to, and a written note of what we would not automate and why. You keep it whether or not you go on. The full design of the first system, with its costs and its weeks, is the first paid stage, and it ends in a pilot against one agreed measure that you can stop after.
It replaces waits, not people. The person who was answering the phone between jobs gets the jobs back; the person chasing invoices gets the afternoons back. Businesses that use it well hire later, and for different work, than they otherwise would have. The one thing it must not do is make a decision that was somebody's job to make.