AI receptionist or button-based bot? If someone can pick from a short menu, buttons are enough. If they write in their own words, change the subject and circle back later, you need a conversation. But neither option turns a message into a confirmed booking by magic
"How much?" can be answered from a current price list. "Is this treatment right for me?" needs a professional, not a confident bot that read the price list

Buttons aren't stupid. They're honest
Address, opening hours, a link to prices, service selection, a callback request: clear paths that are easy to test. A bot like that launches in a few days and behaves predictably. If a question falls outside the menu, the bot should offer a person, not make the customer press "other" all over again
When enquiries are predictable and rules are stable, adding complexity just to display "AI" buys you nothing. Anthropic also recommends starting with the simplest solution that works
You need conversation when the customer breaks the script
Someone asks about a treatment, then an available date, then returns to the price. An AI receptionist can keep the context, ask for missing details and prepare a booking request. But understanding a message is not permission to make promises: prices come from the current source, available slots from the booking system, and unusual or medical questions go to a person
In the clinic Sales Bot project, the bot handles Instagram Direct messages and passes a record to the doctor; a booking is created only after the doctor confirms it. That's how one process is designed, not a forecast for your conversion rate
How it works at the clinic
The clinic's bot answers only from the clinic's own documents: the price list, service descriptions and booking rules. That currently covers about 114 price items and 48 services. What isn't in the documents the bot doesn't invent; it hands it to the owner as a card: who is writing, what they asked and where the conversation stopped. The owner steps in only where a person is needed
The bot was built and tested on real questions from the clinic's inbox, not on made-up examples. That's where you see how people actually write: with typos, voice notes, three questions in one message and a return to the price a day later. A button menu can't handle messages like that; a conversation keeps the thread
The same core connects to Instagram, WhatsApp and a website chat, so the rules for answering and handing over to a person are written once. Building a bot like this takes about 2–3 weeks, including testing on your questions

The crucial part is handing over to a person
Decide in advance what happens when someone asks for a team member, when there is no reliable answer or when a price doesn't match the list. A handoff means keeping the context, letting a person respond and stopping the automated replies from talking over them
Test both setups with anonymised but realistic requests: a routine question, a change of subject, a request for a person and uncertainty about a procedure. Then count prepared enquiries and confirmed bookings separately. A conversation, an enquiry and a visit are three different outcomes; roll them into one number and you have a lovely report about nothing
How to choose between them
Open last month's messages and sort them into two piles. The first holds questions answered in one line: the address, opening hours, the price of a specific service. The second holds conversations where a person clarifies, hesitates, changes the subject or sends a voice note. If nearly everything is in the first pile, start with buttons: they're cheaper and easier to test. If the second pile is noticeable, buttons will send people to a team member so often that there are no savings. Then you need a conversation, with the same handoff rules
If your conversations have outgrown a menu, see how the AI receptionist works. Decide who handles exceptions and confirms bookings first. Pick the model later