From Owner on Call to an AI That Books, Dispatches, and Gets Paid

The client is an owner-led wellness studio with a demanding standard of service. Treatments happen in the guest's room, delivered by independent therapists, and a boutique hotel partner books on behalf of its guests. The studio's reputation rests on discretion and polish, so we have left out anything that would identify it.

When we met, the business was growing from a solo practice into a team. The work that made that growth possible was all landing on one person.

The situation

A single booking touched a lot of hands. A guest or the hotel front desk would call or text. Someone had to confirm the details, find a therapist who was free, wait for a reply, confirm with the guest, collect a deposit, tell the hotel, and after the visit settle the balance and work out what each therapist was owed.

• The owner was the switchboard. A part-time assistant covered long days on the line, and anything unusual still came back to the owner.
• An answering service had already been tried. It could answer calls. It could not answer texts, which is how most guests and the hotel actually reached the studio, and it could not dispatch anyone.
• The scheduling tool fought the process. It handled guests booking themselves well, but bookings made by staff needed a workaround just to take a deposit.
• Money was chased by hand. Deposits, balances, tips, and therapist payouts lived in different places and were reconciled after the fact.

Phase one: set up the operation

We did not start with AI. We started by writing down how a booking should move from the first message to the therapist's payout, and making the decisions that had never been made explicitly.

• The process, step by step. A handbook and walkthrough for every stage: intake, the dispatch round, confirmation, payment, the visit, close out, and payouts.
• The policies behind it. When payment is taken, what a late cancellation costs, how long a therapist waits for a guest who does not answer, how much time sits between bookings, and when the day's last treatment has to start.
• The compliance groundwork. Privacy policy, texting terms, and the consent language needed to register the studio for business texting with the carriers.
• The numbers. A plain financial model: cost per booking, profit at different weekly volumes, taxes, and what a hotel commission would mean for pricing.

One decision shaped everything after it: guests pay in full after a therapist accepts and before anyone is dispatched. Nobody travels to an unpaid booking, and no deposit has to be refunded because no therapist could be found.

Phase two: an agent that runs the line

With the process settled, we built a custom agent to run it. It answers the studio's phone and text line at any hour and works from one record per booking, from the first message to the payout.

On the phone and by text

The agent takes the whole request: length and style of treatment, any enhancements, the time, name and room, the best mobile, and how the guest found the studio. It knows the menu, the prices, and the hours, including which treatment lengths still fit before the evening cutoff. When the hotel's front desk calls on a guest's behalf, it recognizes that, takes the guest's own mobile, and skips the questions meant for guests.

Dispatch

Each booking is offered to therapists by text in the owner's chosen order, with a short window to accept before it moves on. The first yes wins, even from someone who passed a few minutes earlier. A therapist who cannot make the time but can make another one simply replies with it; the agent asks the guest, without naming the therapist, and holds that time as a backup while it keeps looking.

Payment and the visit

Once a therapist accepts, the guest gets one short link: a health form, the studio's safety agreement and waiver, and payment in full. Reminders go out every fifteen minutes, and a booking left unpaid is released. The therapist is told the moment payment lands, and the hotel gets a paid and confirmed text. The therapist checks in and out by text, the guest receives a receipt with the option to add a tip that goes straight to the therapist, and a private review request follows an hour later. Low ratings go to the owner to follow up.

The owner's view

The owner has a dashboard with every booking, call transcripts and outcomes, reviews, weekly reports, therapist payout statements, and year-to-date totals for contractor tax forms. Therapists get their own phone view with only their bookings, their availability, and their tips. Urgent issues reach the owner at any hour; routine updates wait for waking hours.

The guardrails

A service built on discretion needs an agent that knows its limits. These were designed in from the start, and most are enforced in code rather than left to the AI's judgment.

• Prices come only from the menu. Any figure that is not on the price list is stopped before the caller hears it.
• No promises it cannot keep. The agent never says a booking is confirmed until it is, and never promises that a person will call back.
• Health details stay private. Form answers are encrypted and visible only to the assigned therapist and the owner. The AI only knows whether the form is done.
• No cards by text. Card numbers sent in a message are removed before they are stored or seen, and the guest is pointed to the secure payment page.
• Transcripts, not recordings. Calls are transcribed for review, the greeting says so, and no audio is kept.
• A clean line on conduct. Anything outside a professional service is declined plainly and reported to the owner. A second attempt ends the call.
• Therapists are protected too. Guests do not learn a therapist's name before arrival.

Built to run, not to demo

A booking line that goes down loses real business, so the last stretch of the work was about reliability.

• A test suite of real calls. Every problem found on a live call became a scripted scenario. No change to the phone agent ships until all of them pass three runs in a row.
• Fast on the phone. In testing, the median time to the agent's first spoken word was under a second.
• A watchdog. Every few minutes the system checks for anything that would leave a guest or a therapist waiting and alerts us before anyone notices.
• A managed database with hourly backups. Moved over during a planned window with every record verified before and after.
• Race conditions handled. Two therapists saying yes in the same second cannot both get the booking, and a duplicate payment notice cannot trigger two sets of texts.

The timeline

What we learned

• The voice was the easy part. As we wrote in our comparison of AI agents and answering services, the value is in the decision logic, the integrations, and the guardrails.
• Decide the policy before writing the code. Paying in full before dispatch removed a whole category of refunds and awkward conversations. No amount of software would have.
• Different callers need different paths. A hotel front desk booking for a guest is not the same conversation as a guest booking for themselves, and treating them the same produced wrong phone numbers until it had its own path.
• Small details carry the brand. We shortened the payment links guests receive by text and moved them to the studio's own web address, because a long, generic link looked wrong for a luxury service.

The pattern behind this engagement is one we see often: a business that has outgrown the owner being the operating system. It is the founder bottleneck, and here it happened to ring and buzz all day.