04 // SERVICE
Votel AI Agent Setup
A Votel AI agent that answers accurately, books appointments, and knows when to hand over.
Turning on Votel's AI is quick. Making it safe to point at a paying client's inbox is the actual work. We write the conversation, build the knowledge it answers from, connect it to the calendar so it can book rather than promise, set hard escalation rules for anything it should not touch, then test it against real transcripts before it speaks to anyone. After launch we keep reading what it said and correcting it, because an agent left alone drifts.
Conversation design
Before any configuration, we decide what the agent is for and, more importantly, what it must never attempt.
- The jobs it owns: qualify, answer, book, follow up
- Tone and voice matched to your client's brand, not a default assistant
- Opening message and channel-appropriate message length
- Explicit refusal behaviour for anything priced, clinical, legal, or contractual
- What it says when it does not know, instead of inventing an answer
Knowledge setup
An agent is only as good as what it is allowed to read. Most bad AI deployments are a knowledge problem wearing a model problem's clothes.
- Source content gathered from the client's site, price list, and policies
- Rewritten into answerable form rather than dumped in raw
- Service, pricing, and eligibility questions covered explicitly
- Hours, locations, and contact routing kept in one place so they stay correct
- A defined owner and cadence for keeping it current
Booking and escalation
The two moments that decide whether the agent is an asset or a liability: when it closes, and when it gets out of the way.
- Live calendar availability so it books real slots, not intentions
- Qualification questions asked before a slot is offered
- Confirmations, reminders, and reschedule handling
- Escalation triggers: complaints, refunds, clinical or legal questions, repeat confusion
- Handover to a named inbox, a VA, or the client's own staff, with the transcript attached
Testing before it goes live
We try to break it before a customer does. This is the step most setups skip and later regret.
- Replay against real historical enquiries where available
- Edge cases: pricing pressure, off-topic questions, hostile messages, prompt injection attempts
- Confirming it escalates when it should rather than guessing
- Booking flow tested end to end into the real calendar
- Sign-off from you and the client before it is switched on
Ongoing maintenance
Optional, and the difference between an agent that improves and one that quietly gets worse. Priced separately as an ongoing engagement.
- Regular transcript review, sampled and read by a human
- Knowledge corrections when the client changes pricing or services
- Tuning escalation thresholds against what actually came through
- A short written report on what it handled and what it passed on
- Adjusting the build when Votel ships changes to the AI features
How we run it
- 01
Scope
A short call, then a written scope and quote. Nothing starts until you've approved it.
- 02
Deliver
The work happens inside Votel, white-labeled to your brand, with a direct line to whoever's doing it.
- 03
Hand off & support
Your client sees your agency. We stay on for ongoing support whenever you want it.
What this looks like in practice
A three-location clinic that was losing after-hours enquiries
Treatment and pricing questions arrived at every hour across three locations, and the front desk triaged them between walk-ins. Anything that came in after closing sat until morning, by which point the person had usually booked with whoever replied first.
- 01Scoped the agent to three jobs only: answer common treatment and pricing-range questions, qualify, and book a consult.
- 02Built its knowledge from the clinic's own service list, published price ranges, and location hours, rather than letting it read the whole site.
- 03Connected each location's consult calendar so it offered real slots and wrote the booking straight into the right diary.
- 04Set hard escalation: anything clinical, any complaint, and any request for a custom quote goes to a named human inbox with the transcript attached.
- 05Tested it against a month of previous enquiries, including the awkward ones, and fixed the answers it got wrong before switching it on.
After-hours enquiries get an accurate answer and a booked consult instead of waiting for morning. Clinical questions still reach a person, which is the point. The clinic sees one inbox and its own brand; it never sees Votel or us.
What you get at the end
- A configured, tested Votel AI agent live on the channels you chose
- The knowledge base it answers from, documented and editable
- A written escalation policy naming who receives what
- A test log showing what was tried before launch and how it behaved
- A short guide for the client on reading transcripts and flagging bad answers
What this doesn't cover
Said up front so it isn't a surprise on the scoping call.
- Promising a fixed accuracy or booking rate; we tune against real transcripts instead
- Letting the agent quote custom pricing or give clinical, legal, or financial advice
- Writing your client's underlying pricing or service policies
- Ongoing transcript review, unless you add the maintenance engagement
Questions buyers ask
- What is the difference between Votel's AI Assistant, AI Worker, and AI Agent?
- Broadly: the Assistant helps your team inside the platform, the Worker runs background tasks across records, and the Agent is the one that talks to your client's customers on a channel. This service is about the customer-facing Agent, since that is the one with reputational risk attached. We will tell you on the call if what you actually need is one of the other two.
- Will it make things up?
- Any language model can. The job is to make that unlikely and harmless when it happens: a tight scope, knowledge written to answer the questions people actually ask, explicit instructions to say it does not know, and escalation rules that route the risky categories to a human. We test for it before launch and keep sampling transcripts after. Nobody honest will promise you zero.
- Can it book appointments directly into the calendar?
- Yes. It reads live availability and writes the booking into the connected calendar, then handles confirmations and reminders. We qualify before offering a slot so the calendar does not fill with people who were never going to show.
- What happens when it cannot handle something?
- It hands over, rather than improvising. You decide the triggers and the destination during setup: a shared inbox, one of our VAs, or the client's own staff. The full transcript goes with it so the person picking it up is not starting cold.
- Do we need the whole platform set up first?
- You need a working Votel instance with the calendar and channels connected, because the agent books into them. If that is not in place yet, Votel Agency Setup comes first and this gets built on top of it. If it is, we can start here.
- Who keeps it accurate when the client changes their prices?
- Somebody has to, and it should be agreed before launch rather than after. Either you own the knowledge base and we hand you the guide, or you add the maintenance engagement and we review transcripts and apply corrections on a set cadence.
More on how engagements are quoted is on the pricing page, or bring the question to a scoping call.
Other services
Need Votel AI Agent Setup?
Tell us about your sub-accounts and what's already in place, and we'll scope it on a call.