The idea of having an AI agent answer your phone sounds great until you start wondering how it knows what to say. What if someone asks about your pricing? Your service area? Whether you're available on Saturdays? If you're going to train an AI agent for your business, it needs to know your business — and the good news is that doesn't require any technical skills or a developer on speed dial.
Most business owners can go from zero to a fully trained AI phone agent in under 30 minutes. Here's how it works.
Your AI agent is only as good as what it knows
An AI phone agent without context is like a temp on their first day with no briefing. They'll be polite, but they won't be able to answer basic questions about your business. Callers will notice, and the experience will feel generic.
The difference between a mediocre AI agent and one that sounds like it belongs at your business comes down to two things: a knowledge base and a prompt. The knowledge base is the facts — your pricing, services, hours, and policies. The prompt is the personality and behaviour — how the agent introduces itself, what questions it asks, and how it handles different situations.
Get both right, and callers won't believe they're not talking to a real person on your team.
Building your knowledge base
Think of the knowledge base as a cheat sheet for your AI agent. It's a collection of short entries that cover the questions your callers ask most often. You don't need to write a novel — just clear, specific answers to the things that come up again and again.
Here's what to include:
- Services you offer — List each service with a one or two sentence description. "We do full house rewires, switchboard upgrades, and new power point installations" is more useful than "we do all electrical work".
- Pricing guidance — You don't need exact quotes, but ranges help. "A standard call-out is $80–$120 plus GST, and most small jobs are done in under an hour" gives callers enough to decide if they want to book.
- Service area — Be specific. "We cover Christchurch city and the surrounding suburbs including Rangiora, Rolleston, and Lincoln" is better than "Canterbury region".
- Business hours — When you take calls, when you do the work, and whether you charge extra for after-hours or weekends.
- Booking rules — Can the AI book directly into your calendar? Do you prefer to call back and confirm? Is there a minimum notice period?
- Common FAQs — Whatever your callers ask most often. Do you offer free quotes? Do you charge for travel? Are you licensed and insured? What payment methods do you accept?
Writing entries that work well for voice
There's a difference between information that reads well on a website and information that sounds natural when spoken aloud. Your knowledge base entries are going to be used in a phone conversation, so they need to be written for the ear, not the eye.
A few practical tips:
- Keep entries short. One to three sentences per entry. The AI doesn't need to read out a paragraph — it needs a quick, accurate answer to weave into the conversation.
- Be specific. "We can usually get there within 24 hours for non-urgent jobs" is better than "we aim to provide prompt service".
- Use conversational language. Write the way you'd actually talk to a customer. If you'd say "yeah, we do that" on the phone, don't write "this service falls within our scope of offerings".
- Avoid jargon. Your callers might not know what an RCD is, but they know what a safety switch is. Use the words your customers use.
- Include boundaries. It's just as important to note what you don't do. "We don't do gas fitting — you'll need a registered gasfitter for that" prevents the AI from overpromising.
Train an AI agent with the prompt builder
The knowledge base gives your agent facts. The prompt builder gives it behaviour. This is where you tell the AI how to act — and you do it in plain English, not code.
A prompt might include instructions like:
- "Introduce yourself as the answering service for [business name]"
- "Always ask for the caller's name, phone number, and address"
- "If the call sounds urgent — like a water leak or power outage — let the caller know you'll flag it as high priority"
- "If someone asks for a price, give the range from the knowledge base but explain that a final quote depends on the specific job"
- "Don't book appointments directly — let the caller know that [name] will call them back to confirm a time"
That's it. You write the rules in your own words, and the AI follows them. No scripting language, no decision trees, no flowcharts. Just tell it what to do the same way you'd brief a new team member.
What most businesses get wrong
The most common mistake is trying to cover everything. You don't need a knowledge base entry for every possible question — you need entries for the questions that come up on 80% of calls. Start with the basics, listen to a few calls, and add entries for anything the AI couldn't answer well.
The second mistake is being too formal. Your AI agent should sound like your business, not like a corporate call centre. If you're a Kiwi tradie, your agent should sound like a Kiwi tradie's office — friendly, direct, and no-nonsense.
The third is setting it and forgetting it. Your best results come from checking the call summaries in the first week or two, spotting any gaps, and adding a few more knowledge base entries. After that initial tuning, most businesses rarely need to change anything.
From zero to trained in 30 minutes
Here's a realistic timeline for setting up your AI agent:
- Minutes 1–10: Write out your core services, pricing ranges, service area, and hours. You already know this — it's just putting it down in short entries.
- Minutes 10–20: Add your top 5–10 FAQs. Think about the last 20 calls you took and what people asked.
- Minutes 20–30: Write your prompt. Tell the agent how to introduce itself, what to ask, and how to handle common situations.
That's it. You can refine and add to it over time, but most businesses are up and running with a solid AI agent in a single sitting.
Platforms like dareena.ai make this process straightforward with a built-in prompt builder and knowledge base — no technical skills required. You type in plain English, test it with a call, and tweak from there. It's designed for NZ small businesses and trade operators who want their phone answered properly, not a science project.