Your First 30 Days with an AI Receptionist: What Actually Happens

What setup involves, what the PMS connection needs, how the first week of calls is reviewed, and when to widen hours and channels.

Deciding that a clinic needs an AI receptionist is the well-documented part. The signs are set out in 7 signs it is time for an AI receptionist, and most owners recognise their own clinic in two or three of them. The part almost nobody documents is what happens after the sign-up form: what the clinic has to provide, what connects to what, and how long it takes before the phone can be trusted to software.

This guide walks through the first 30 days in order, based on how BookedSolid onboards clinics onto its own platform. The shape is similar with any AI receptionist that connects to a practice management system: setup, a connection, a test, a controlled go-live, then a review loop. The specific settings, dial codes and screens described here are BookedSolid's, and each step has its own guide in the help centre.

One expectation worth setting early: the first month is a staged handover rather than a single switch-over, and the clinic controls how fast each stage moves.

What Setting Up an AI Receptionist Involves (Days One and Two)

Setting up an AI receptionist takes a clinic a few hours of its own time, spread over the first day or two. The work divides into four jobs, and none of them needs technical staff.

1. Connect the practice management system. The connection itself takes a few minutes. On Cliniko, for example, the clinic generates an API key from its own account and enters it at sign-up; the Cliniko AI receptionist guide covers the mechanics in depth. The step worth checking first is the diary settings, because the AI books through the same availability rules as the online booking page. Each practitioner needs online bookings switched on, and each location needs them displayed, or the AI will have no slots to offer. BookedSolid connects to Cliniko, Nookal, Splose, PracticeHub, coreplus, PracSuite, Semble and ClinicOffice; the integrations page holds the current list.
2. Load what the AI needs to know. The knowledge base takes three kinds of content: the clinic's website, PDF documents such as price lists and procedure information, and short text notes for the things no website mentions, like parking, directions and what to bring. An honest caveat here: the knowledge base holds information, not instructions. Rules of the "if a caller asks this, do that" kind belong in the call-handling settings, not in a note.
3. Set the escalation rules. This is the decision that matters most on day one: what happens when a caller needs a person. The options are an instant transfer to a staffed line, an email carrying a summary of the request and the caller's details for a callback, or a mix, with transfers during clinic hours and email outside them. Requests for a named practitioner can route to that practitioner's own inbox.
4. Choose the voice and the details it collects. Accent and speaking speed are settings, and so are the details gathered from a new patient: name, contact information, how they found the clinic. Both take minutes and both can change later.

While the clinic works through this, BookedSolid provisions a dedicated AI phone number, typically ready within 24 hours. Nothing patient-facing has changed yet. The clinic's own number still rings at the desk exactly as it always has.

Test Calls Before Any Patient Hears It

The AI number is live before the clinic's real line ever touches it, which means the team can ring it directly and try to trip it up. Good test calls are the awkward ones: booking "some time after four, but not Tuesday", asking a question the website does not answer, giving a wrong date of birth, and asking outright for a person.

A booking request should end with the appointment read back correctly and sitting in the practice management system diary within moments. A request for a person should end in whichever escalation route was configured: the test phone transferring, or the summary email arriving. A question the AI cannot answer gets flagged, and the account admin is prompted to fill the gap with a short text note.

An afternoon of this moves most of the teething out of patients' earshot. Whatever surfaces in test calls would otherwise have surfaced in week one with a real caller on the line.

Going Live Is a Call Forwarding Decision

Going live changes nothing about the number patients dial. The clinic keeps its published number and diverts it to the AI line. On a mobile or VoIP system that is the call forwarding or divert setting, on a standard landline a short dial code, and on a PBX the usual forwarding or SIP routing. The divert is set from the clinic's side and reversible in under a minute, which removes most of the risk from the decision.

The real decision is when the divert applies, and there are three honest options:

• Out of hours only. The desk answers during opening hours as it always has; the AI takes evenings, early mornings and weekends. About 29.5% of booking activity happens outside business hours, so this recovers real bookings while the team watches results with no change to its working day. It is the lowest-risk way to start.
• Overflow. Most phone systems can divert on busy or no answer, which sends the AI only the calls the desk cannot take: the lunchtime rush, the second caller during a check-in. Reception keeps answering first; nothing rings out.
• Full cover. The divert runs around the clock and the AI answers first at all hours, with staff handling whatever it escalates. Clinics tend to arrive here in week two or three rather than on day one, once the call reviews have earned the confidence.

Week One: Reviewing the Calls

The first week's job is review. Escalations arrive as they happen, each with a summary of the request and the caller's details, and conversations on messaging channels sit in the dashboard where staff can step in mid-thread. Whenever the AI meets a question it cannot answer, the gap is flagged and the account admin is prompted to supply the missing information as a text note.

The habit that works is ten minutes each morning: read yesterday's conversations, answer the flagged gaps, and check that bookings landed on the right practitioners and appointment types. Week one is where a clinic's unwritten knowledge surfaces, the Saturday clinic that runs from a different location, the practitioner who does not treat under-16s, the deposit policy nobody ever wrote down. Every flag becomes a one-line note, and the same question is answered correctly from then on.

Expect gaps in week one; filling them is how a generic assistant becomes the clinic's own. The flag count falls quickly, because patient questions repeat far more than they vary.

Weeks Two to Four: Widening Hours, Then Channels

The pattern that works is widening one dimension at a time. A clinic that started out of hours brings the divert forward into lunchtimes and busy periods, then moves to full cover when the morning review stops turning up surprises.

Channels come next. The same assistant that answers the phone can cover SMS, WhatsApp, email and a web chat widget on the clinic's site, each switched on separately with its own setup guide in the help centre. Every channel feeds the same diary and the same escalation rules, so a WhatsApp message and a 9pm call end the same way: a booking in the practice management system or a summary in a staff inbox.

Outbound work is the final widening, and it is worth sequencing deliberately: switch on automated reminders and confirmations once the diary write-back has been watched for a week or two, because outbound messages act on the same appointment data the AI has been writing. Automated reminders cut non-attendance by around 29% of the baseline rate, so for many clinics this is the second gain of the month after the answered calls themselves. Follow-ups, waiting list fills, deposit collection through Stripe and invoice chasing sit behind the same switches when the clinic is ready.

The First 30 Days at a Glance

What to Tell Patients, and What to Tell the Team

Patients need less preparation than most owners expect. The evidence says acceptance is driven by access rather than by any enthusiasm for AI. Comfort with speaking to a non-human call centre rose from 19% to 35% in a year, acceptance is highest for booking and routine queries, and the fuller picture is in do patients like AI phone systems. A short line on the website and in the on-hold or voicemail message covers the announcement, and the reassurance that actually matters is built into the escalation rules: any caller who wants a person can ask for one and get one where available.

The team needs a more specific message, because "AI receptionist" sounds like a replacement to the people currently answering the phone. The accurate framing is narrower: it takes the calls staff physically cannot take, the ones that ring out during a check-in and the ones that arrive at 9pm, and it hands over the conversations that need a person, summarised, with the caller's details attached. The Physio Lounge, a seven-site group, restructured from five admin staff to three after moving evening and weekend calls to software; the people who remain deal with the patients in front of them. Two practical assignments make the first month smoother: name the person who owns the ten-minute morning review, and agree who answers escalation emails so that no handover sits unclaimed.

What Good Looks Like by Day 30

By the end of the first month the clinic has data where it previously had a hunch. Three numbers carry the verdict:

• Calls answered. The share of inbound calls that now get answered, compared with the unanswered and abandoned calls the phone log showed before. This is the number that motivated the purchase, and it should move immediately.
• Bookings made. Appointments the AI wrote into the diary, and in particular the ones booked outside opening hours, which are bookings the clinic previously had no way to take.
• Conversations that needed a person. The escalation share, and whether the right conversations escalated. A falling flag count and a stable escalation pattern are the signs of a system that has learned the clinic.

For a sense of what an established deployment looks like, Health Sciences University runs two teaching clinics where around 70% of 2,000 monthly calls once went unanswered. Today around 60% of calls are handled with no human involvement, and more than 700 appointment requests a month go through the system. A first month does not need those numbers. It needs the three figures above moving in the right direction on the clinic's own volume.

Frequently Asked Questions

How long does it take to set up an AI receptionist?

Most clinics are live within 48 hours of signing up. The clinic's own share of the work is a few hours: connecting the practice management system, loading the knowledge base, setting escalation rules and making test calls. The dedicated AI phone number is typically provisioned within 24 hours, and going live is a call forwarding change on the clinic's existing line.

Does a clinic need to change its phone number for an AI receptionist?

The clinic keeps its existing number. A dedicated AI number sits behind the scenes, and the clinic diverts its published number to it, around the clock or only outside opening hours. Patients dial the same number they always have, and the divert can be removed from the clinic's own phone settings at any time.

What does an AI receptionist need from the practice management system?

It needs a live connection with permission to read availability and write bookings, which on most systems means an API key generated from the clinic's own account. Because the AI books through the same rules as online bookings, practitioners and locations need online booking enabled, and the availability offered to callers mirrors what the booking page shows. BookedSolid connects to Cliniko, Nookal, Splose, PracticeHub, coreplus, PracSuite, Semble and ClinicOffice.

Can an AI receptionist handle only out-of-hours calls?

Out-of-hours-only cover is a standard configuration and a sensible first step. The team answers during opening hours as usual, the AI takes evenings, mornings and weekends, and the divert widens whenever the clinic chooses. With around 29.5% of booking activity happening outside business hours, out-of-hours cover on its own recovers bookings the clinic was previously losing overnight.

What should a clinic tell patients about an AI receptionist?

Patients mainly need to know that the phone gets answered and that a person is reachable on request. Comfort with non-human call handling rose from 19% to 35% in a year in Bain and Company's survey, with acceptance highest for booking and routine queries. A short note on the website and a line in the on-hold or voicemail message cover the announcement, and the escalation rules guarantee the route to a person.

When does a clinic start paying for an AI receptionist?

BookedSolid runs a 7-day free trial with full platform access, no setup fees and no contract, so the first days of live calls happen before the first payment. Pricing after the trial is practitioner-based and published on the pricing page, and a clinic that decides against it removes the call divert and walks away.