Training Your HonorElevate AI Agent on Your Prices, Services, and Hours

An untrained AI voice agent is a generic chatbot pretending to be a receptionist. A trained AI voice agent is your specific business answering its own phone. The difference is the training methodology. This is how I do it.

Stage 1: Discovery

One 60-minute call. Cameras on. I take notes. You answer the following.

What services do you sell?

Full service catalog. For HVAC: service calls, diagnostic, repair, maintenance plan, system replacement, indoor air quality, ductwork. For dental: new patient exam, hygiene, restorative, cosmetic, ortho, recall. The agent needs to know what you sell and how each service is described in plain customer language. "Crown" not "indirect restoration." "AC tune-up" not "preventive maintenance protocol."

What does each service cost (or what is the pricing logic)?

Base pricing where it exists. Pricing logic where it does not. "Service call is $89, applied to repair. Diagnostic adds $0. Most repairs run $300-$800 depending on part. Full replacements quoted on-site only." That logic gets encoded so the agent can give honest expectations on the call without lying or overcommitting.

What are your hours and service area?

Office hours, dispatch hours, emergency hours. Specific zip codes you serve. Cities you do NOT serve. Buffer zones where you might serve but charge a travel fee. The agent uses this to qualify or politely decline.

What questions do you ask every new caller?

Your existing qualifying script if you have one. Or your mental checklist if you do not. "What zip code? What is the issue? Is the system blowing? How old is it?" Those become the agent's intake flow.

What is an emergency?

Define what triggers immediate escalation vs same-day vs next-day vs scheduled appointment. For HVAC: kids/elderly + no cooling in summer = emergency. For plumbing: active leak or sewage backup = emergency. The agent applies these triggers automatically. The full escalation logic gets documented in What Happens When the AI Cannot Handle the Call.

What is your tone of voice?

This is the part most setup people skip. The agent will sound like whatever you train it to sound like. Friendly but professional? Direct and efficient? Warm and slow? Honest and a little blunt? I capture the answer with sample phrases you actually use. "Thanks for calling, this is Sarah, how can I help" lands different than "Smith Heating, this is Sarah." Both work. Yours has to feel like yours.

Stage 2: Knowledge Base Build

Day 2 and 3. I do the work. You go run your business.

Everything from discovery gets encoded into three places.

1. The system prompt. The agent's core identity. Business name, tone, primary directive ("book the appointment, qualify the caller, escalate edge cases").
2. The knowledge base. Structured data on services, pricing, hours, service area, FAQs, edge cases. The agent retrieves from this in real time.
3. The conversation flow. Decision-tree logic for how the agent routes a call. Greet → identify problem type → qualify → route to booking flow or escalation flow → confirm → close.

I also load the agent with the 20-40 most common FAQs you actually get on calls. "Do you offer financing?" "Are you licensed and insured?" "Do you do free estimates?" "Can I get a price over the phone?" "Do you work on Trane systems?" If you do not have a list, I generate one from your industry baseline and let you correct it.

Stage 3: Prompt and Personality Tuning

Voice selection happens here. You pick the voice from a panel of options. Female warm. Female direct. Male professional. Male relaxed. Some clients want a specific name (Sarah, Mark, Betty, Diana). Some want their actual receptionist's voice cloned (with consent and a 5-minute recording, this is possible). Most go with a stock voice that matches their brand and let me name it.

Personality tuning is the texture. How much filler vs how direct. Whether the agent uses "absolutely" or "of course." Whether it says "no problem" or "you bet." Whether it pauses dramatically on bad news or drives forward. These are calibrated against the tone samples I captured in discovery.

Stage 4: Live Test Calls

Day 4. The fun part.

I call the agent. Thirty scenarios. Some are standard ("I need an AC service appointment"). Some are tricky ("my mom is 87 and the AC is broken, she has a heart condition"). Some are designed to break it ("I want to schedule a unicorn massage for my pet llama"). Some are noise tests (background TV, kids screaming, dog barking).

I score each call on:

• Greeting: picked up fast, sounded natural, on-brand?
• Empathy: acknowledged the situation before qualifying?
• Qualification: asked the right questions in the right order?
• Booking: offered a real slot, captured complete info?
• Confirmation: sent SMS, updated CRM, fired the right notifications?
• Edge cases: escalated correctly when it should have?

Whatever fails, I fix. We re-test until 30/30 passes. Then I send you a recording link with the 10 most representative calls and you score them yourself. Anything you flag, I fix.

Stage 5: Soft Launch and Full Cutover

Day 5-7.

Soft launch routes 10-25% of your inbound calls to the agent. The other 75-90% still hit your normal flow. I watch the first 50 live calls in detail. I look for anything I missed in test scenarios.

If everything looks clean, we go to full cutover on day 7. Your main number routes to the agent. The agent answers everything. Escalations route to your cell. You start sleeping through Tuesday-night calls.

What gets updated after launch

The agent is not a one-and-done deploy. It is a living system.

Week 1 post-launch: I review every single call. Anything weird, I fix. New edge cases get added to the knowledge base. The agent gets better daily.

Week 2-4: Weekly review. Trends emerge (callers asking about a service we forgot to load, pricing question we did not anticipate, an FAQ we missed). All of it gets added.

Ongoing: Monthly check-in. Update for seasonal pricing, new services, modified hours, holiday schedules. You add a Saturday hour? 30 minutes of work, deployed same day. You roll out a new $99 maintenance plan? Same.

What you should NOT do during training

• Do not try to anticipate every edge case. The 80/20 rule. Train on the 80% of calls that follow standard patterns. The other 20% escalate to you and we add them to training week 2. Trying to handle every edge case upfront slows the deploy by a month.
• Do not over-script the agent. Tight scripts make the AI sound robotic. Loose principles ("be warm but efficient, get to booking fast") make it sound human. Train on principles, not literal words.
• Do not skip the test calls. The 30-scenario test is what catches problems before customers do. Skipping it for speed is how you end up with a broken AI in production.
• Do not get precious about voice selection. Pick a voice in five minutes. The tone matters more than the voice. You can change the voice in week 2 if you hate it.

How this compares to DIY voice AI

You can buy generic AI receptionist tools from Vapi, Bland, Retell, Air, Synthflow, and a dozen others. They give you a dashboard, a few prompt fields, and a phone number. Cost runs $99-$499 per month depending on minutes.

The catch: you do the training. You write the system prompt. You build the knowledge base. You configure the calendar integration. You test the conversation flow. You debug when it books a customer in the wrong time zone or quotes a price that does not exist.

For an experienced AI builder this is fine. For a service-business owner who is also a plumber, an HVAC tech, or a dentist, you are now doing two jobs. You will either underbuild the agent (and lose customers) or spend three months learning prompt engineering when you should be running trucks.

HonorElevate's value proposition is operational. I do the training so you do not have to. The platform under the hood is the same enterprise infrastructure the DIY tools sit on top of. The difference is who's holding the wrench.

The bottom line

A trained AI voice agent sounds like your business and books appointments by your rules. An untrained AI voice agent sounds like a chatbot. The training methodology is what bridges the gap. 5 stages, 3-5 days, one operator (me) doing the work while you run your business.

If you want the full agent's-eye view of how a real call plays out post-training, read How a HonorElevate AI Agent Answers a 7:48 PM Emergency Call. If you want to see what happens when the AI hits something it cannot handle, read What Happens When the AI Cannot Handle the Call. If you want to start, book the free audit.