An AI chat widget that books meetings
Most chat widgets collect a name and an email and call it a lead. Ours reads your price list, checks your actual calendar, and puts a meeting on it. Here's what's different, and why that's safe to turn on.
A contact form with a personality
Open most website chat widgets and ask a real question, something like “how much for the medium size, delivered by Friday”, and what comes back is a polite deflection: someone will be in touch, please leave your email. The widget is not actually answering. It is a contact form that types one letter at a time, and the personality is doing the work that a proper answer should be doing.
That’s not a knock on the idea of a chat widget. It’s a knock on what most of them are given to work with, which is usually nothing: no price list, no calendar, no memory of what the visitor already said two messages ago. A generic model with no data behaves exactly like a form, because a form is the safest thing it can fall back to.
What changes with a real price list
The agentlyleads chat agent, covered in full on the chat agent feature page, reads your actual catalogue: the offerings and prices you’ve entered into the CRM, plus any pages you point it at. Ask it about the medium size and it answers with the number you put in, not a plausible-sounding guess. Ask about something you haven’t listed, a size you don’t carry or a service you don’t offer, and it says so and offers to get a person. The rule that makes this safe is that if a fact isn’t written down, the agent doesn’t know it. It fails by admitting it rather than by inventing an answer.
That single rule is most of what separates “an AI on the website” from “an AI on the website you can trust with a price”. A wrong price is not a bad chat experience, it’s a promise your team now has to either honor at a loss or argue about with a visitor who has a transcript.
What changes with a real calendar
When you connect a meeting type, the agent stops saying “someone will reach out to schedule a time” and starts offering actual times, read from the same booking link engine that powers your public scheduling page. That means the times it offers already account for your working hours, your buffers between meetings, your minimum notice period, and your connected Google or Outlook calendar’s existing busy time. A visitor taps a button instead of typing “does Tuesday at 2 work” and waiting for a reply that might not come until the next business day.
Once a time is taken, it’s gone from what the agent can offer next, because it’s the booking itself that blocks the slot, not a separate counter the agent has to remember to update. Two visitors chatting with the widget at the same time cannot both walk away with the same 2pm.
Asking for a human always works
However good the agent gets at prices and scheduling, some visitors just want a person, and that request is checked before the model is even called. Phrases like “can I talk to someone” are caught and handed off immediately, which matters because it means the handoff can’t be talked past by a model that’s mid-answer and inclined to keep going. The conversation moves to your team’s queue with the full transcript attached, so whoever picks it up isn’t starting cold.
The ceilings that make this safe to turn on
Giving a public chat widget access to your calendar and price list sounds like an invitation to bots and time-wasters, so it ships with three limits rather than one. An anonymous visitor, one who hasn’t left an email or phone number, gets 8 replies: enough to answer a real question and ask twice for a way to reach them, not enough to carry on a conversation indefinitely. Any single conversation, anonymous or not, stops at 40 messages. And each site carries a monthly cap you set yourself, with a warning at 80% of it rather than a surprise after the fact.
Those three numbers are the actual guardrail. Not a promise that it will behave, a hard stop that doesn’t care whether it’s a slow visitor, a bored one, or a script testing the widget.
Where the conversation ends up
A chat that leaves an email becomes a contact or a lead automatically, and the transcript sits on that record’s timeline next to their emails and notes rather than in a separate chat log you have to remember exists. A booked meeting shows up as a task on the calendar. By the time you open the record next week, the chat isn’t a thing you have to go dig up, it’s just part of the history.
Common questions
- Will it quote a price I haven't entered?
- No. It only knows the services and prices you have put into your catalogue. Ask about something you haven't listed and it says it doesn't know, then offers to get a person, rather than guessing a number.
- Can a visitor always reach a human?
- Yes. A request for a person is checked before the model is even called, so it can't be argued past or smoothed over by the conversation. It hands off to your team immediately.
- What stops it running up a bill talking to a bot?
- Three ceilings. An anonymous visitor with no email or phone on file gets 8 replies. Any single conversation stops at 40 messages. Each site carries a monthly cap you set yourself.