Virtual AI Receptionist: What It Is, What It Does and What It Costs

A virtual AI receptionist is software that does the first-line work of a front desk: greeting visitors, answering common questions, capturing contact details and booking appointments. It does not sit at a desk or take a lunch break, and it answers instantly at any hour. For a small business where the owner or a single team member also does everything else, that first layer is often what gets neglected, and where enquiries are lost.

This guide explains what a virtual AI receptionist actually does, how it differs from older tools, what to prepare, what it costs and how to keep the human touch. It covers the web-chat version, which is where Crescelo works today; phone answering is a separate channel that is under development at Crescelo, and we say where that matters.

What a virtual AI receptionist does

• Greets and orients: welcomes visitors and states what it can help with.
• Answers questions about services, prices, hours, location and policies from your documents.
• Qualifies enquiries by asking the questions you would ask, such as which service, when and where.
• Books and reschedules appointments into your calendar.
• Captures leads and sends them to your email, spreadsheet, Slack or other tools.
• Escalates to a person when a case needs one, with the conversation attached.
• Records everything, so you can see what customers ask and where the gaps are.

It is a receptionist for the routine layer. The judgement calls and the relationships still belong to your team.

How it differs from the alternatives

How it works underneath

A virtual AI receptionist combines four parts. A knowledge base holds your documents. A retrieval step finds the passages that relate to a question. A language model writes the reply from those passages under instructions to answer only from them. And an action layer books appointments, records leads and notifies people. The action layer is ordinary software, which is why bookings can be exact even though the conversation is flexible. The mechanics are explained in what an AI website assistant is.

Training it on your business

There is no training in the machine-learning sense. You load your content and tell it how to behave. The quality of the result depends on three things.

1. Your service catalog: clear names, prices, durations and what is included. See training an AI receptionist on your service catalog.
2. Your policies: cancellation, deposits, lateness, emergencies, refunds.
3. Your voice: how formal, how friendly, what to avoid saying. See brand voice alignment for an AI receptionist.

Test with the questions customers really ask, in different words and with typos, and fix each failure by improving the source content.

Business hours, holidays and locations

A receptionist that offers a 9 a.m. slot on a public holiday, or quotes the wrong branch's address, damages trust. Set your opening hours and holidays where the assistant can read them, and give each location its own information if you have more than one. See business hours logic for AI chat and multi-location AI receptionists.

Where a virtual receptionist should hand over

The best receptionists know their limits. Build clear rules for handing over to a person: complaints, requests for medical, legal or financial advice, urgent or emergency wording, anything involving a refund or dispute, and any visitor who asks for a person. The conversation should go with the handover so nobody has to repeat themselves. See live human handover flow design and handling emergencies in automated chat.

What it costs

Crescelo's plans start at $49 a month for one assistant with 100 chat sessions, and Growth at $149 adds booking, calendar sync and white-label options. See the pricing page. The important comparison is against what missing an enquiry costs you; see the ROI calculator guide.

Protecting your front desk team

A common worry is that an AI receptionist replaces people. In practice it tends to remove the repetitive load that causes burnout, so staff spend their time on the conversations that need them. Front desks that are constantly interrupted make mistakes and lose patience. Our piece on preventing front desk burnout with AI explores this. Be open with your team about what the assistant will and will not do.

Voice: web chat today, phone later

Customers reach businesses by phone, web chat and messaging apps. Crescelo's assistant works in website chat and, on higher plans, WhatsApp and Telegram. Answering phone calls is under development and shown as planned on the roadmap, so we do not sell it as available. If your customers overwhelmingly call, consider an answering service or phone AI product for the phone today, and use web chat for your website. Our comparison of voice AI and web chat receptionists explains the trade-offs.

Setting one up: a one-afternoon plan

1. Gather your services, prices, hours and policies into short documents.
2. Create the assistant and load the documents.
3. Set the greeting and tone.
4. Add your calendar and services if you want bookings.
5. Test twenty questions, including three the content cannot answer.
6. Set handover rules and notification emails.
7. Publish the snippet on your site.
8. Review the first week of conversations and fix gaps.

You can rehearse steps 2 to 5 with no account in the playground.

Objections you will hear, and honest answers

• 'Customers hate talking to bots.' They hate bots that trap them. They tend to accept an assistant that answers quickly, is clear that it is an AI and lets them reach a person in one step.
• 'It will say something wrong.' It can, which is why grounding, testing and weekly review matter. Treat the knowledge base like a page on your website: accurate, dated and reviewed.
• 'We are too small.' Small businesses gain the most, because the person answering enquiries is usually the owner, and every interruption costs billable time.
• 'It sounds robotic.' Tone is a setting. Give it three sample replies in your own voice and it will follow them.
• 'What if it books the wrong thing?' Confirmations state the service, time and price, and every booking can be changed or cancelled. Review the first fifty bookings.

Common mistakes

• Leaving out-of-date prices in the documents.
• Letting the assistant answer medical, legal or financial questions.
• Not telling visitors it is an AI.
• Hiding the option to talk to a person.
• Measuring success by conversation volume instead of bookings and leads.
• Never reviewing transcripts.

A day in the life: what it handles

To make this concrete, here is how a virtual AI receptionist might spend a normal day for a physiotherapy clinic that is closed on Sundays.

Notice the last row. The receptionist improves every day because someone reads what customers asked. Without that review, the assistant stays as good as it was on day one.

Measuring the result

Pick a few numbers before launch and compare them after a month.

• Enquiries answered within a minute, compared with your previous response time.
• Bookings from the website, and how many were made outside office hours.
• Missed or abandoned enquiries, where visitors asked something and left without an answer.
• Time your team spends on repetitive questions, estimated before and after.
• Handover requests and their causes, which show what to add to the knowledge base or what genuinely needs a person.

If the numbers do not improve, the cause is usually content: missing answers, unclear prices, or no booking option. It is rarely the technology. See how to review your chatbot conversations each week.

Questions to ask before you choose

1. Does it answer only from my documents, and does it say when it does not know?
2. Can it book into my real calendar and prevent double bookings?
3. How do I take over a conversation, and will I be notified?
4. What does it cost at three times my volume?
5. Where is my data stored, and can I delete it?
6. Which channels does it cover today, and which are only planned?

Frequently asked questions

Further reading

• Google Search Central: creating helpful, people-first content: What Google looks for in useful pages.