Skip to main content
This guide gets the Hiveku AI Voice Agent answering your business phone — greeting callers, understanding intents, capturing details, booking meetings, and transferring to humans when appropriate. Plan for 30-60 minutes including testing and refinement.

​
Before you start

Have ready:
  • At least one DID configured. See Buy and Configure a Phone Number
  • A clear greeting script (or willingness to draft one in the builder)
  • Decisions about which intents you want handled and where they should route
  • Optional but useful: FAQ content, pricing, or product specs to upload as the agent’s knowledge base

​
Why the AI agent vs an IVR

The two aren’t exclusive — many businesses use the AI agent during business hours and a simpler IVR after-hours, or vice versa.

​
Step 1: Pick a voice persona

1

Open the AI Agent tab

/dashboard/communications/ > AI Agent.
2

Browse personas

Multiple voice options — male/female, US/UK/AU accents, neutral or warm tone. Each has an inline preview button.
3

Pick one

Match your brand. Casual D2C → warmer, friendlier. Enterprise B2B → measured, professional. You can change later.

​
Step 2: Write the greeting

The greeting is the first thing callers hear. Short and friendly, with an open question that invites the caller to say what they need.
Avoid menu-style greetings (“press 1 for sales”) — the AI is conversational, not a phone tree.
1

Write the greeting in the builder

Type your script. Preview the audio.
2

Iterate

Listen 3-5 times. Adjust word choice and timing until it sounds natural.

​
Step 3: Define intents

Intents are the categories of caller need the agent should recognize. Most businesses have 3-6 intents.
For each intent, decide what the agent does:
Get the caller’s name, company, email, and reason. Transfer to a ring group or queue with a verbal hand-off summary. Use for: sales, support.

​
Step 4: Build the knowledge base

The KB is what the agent uses to answer factual questions inline. Without it, the agent will say “let me transfer you” for any specific question.
1

Open the KB tab in the AI Agent settings

Knowledge Base section.
2

Add sources

Three input modes:
  • URL — Hiveku crawls the page (e.g., your pricing page, FAQ, support docs)
  • Upload — PDFs, Markdown, plain text
  • Direct entry — Paste content directly
3

Tag entries

Tag entries to specific intents. “Pricing FAQ” tagged with sales so the agent only references it on sales calls.
4

Test KB lookups

Use the inline tester: “What’s the price of the Pro plan?” The agent shows which KB entry it would cite.
Start with your most-asked questions — the top 10 questions you get on calls today. That alone will deflect 60-80% of calls from human handoff.

​
Step 5: Set escalation rules

Decide when the agent should transfer to a human.

Caller asks for a person

“Can I speak to someone?” — always transfer.

Frustration detected

Profanity, raised tone, repeated complaints — transfer.

Stuck on a step

3 reprompts on the same input (email, name) — transfer.

Out-of-scope intent

Caller’s need doesn’t match any handler — transfer.

Always transfer intent

Specific intents (e.g., billing dispute) always go to a human.

Configurable timeout

If the agent takes longer than N minutes total, transfer (rare safety net).
Pick a transfer destination — ring group, queue, or specific extension. The receiving agent gets a brief verbal summary of what the AI gathered: “I have Sarah from Acme on the line, she’s interested in the Pro plan, here she is.”

​
Step 6: Connect calendar (for in-call booking)

If you want the agent to book meetings during a call, connect Calendar and pick which meeting type it can book.
1

In AI Agent settings, find 'Booking'

Toggle on.
2

Pick a meeting type

Usually the most common inbound meeting — “Demo (30 min)” or “Intro Call (15 min)”.
3

Set host routing

Single host (always books with one person), team round-robin, or specific-host based on intent.

​
Step 7: Point a DID at the agent

1

Open the DID's detail page

/dashboard/communications/ > Numbers > pick the DID.
2

Set inbound destination to AI Agent

Replace whatever was there.
3

Set time conditions if needed

Common pattern: AI agent during business hours, simpler “leave a message” IVR after-hours. Or AI agent 24/7.

​
Step 8: Test thoroughly

This is where most of the value of setup time is spent. The first time you call, you’ll hear three things you want to fix.
1

Call from your own phone

Run through 3-5 scenarios:
  • “I’m interested in your product” — sales path
  • “I have a billing question” — billing path
  • “What are your hours?” — KB path
  • “Can I talk to a person?” — escape hatch
  • Long silence — does it reprompt?
2

Listen to transcripts

Open each call in /dashboard/communications/calls/. Read the transcript. Note where the agent answered well and where it stumbled.
3

Refine

For factual misses → improve KB entry phrasing. For tonal issues → change persona or adjust prompt instructions. For routing issues → update intent definitions.
4

Test edge cases

Background noise, slow connections, unusual accents, very fast speech, very slow speech.

​
Step 9: Soft launch

Before pointing your main number at the agent, soft-launch on a secondary DID.
1

Buy a second DID

Use it for outbound or one specific marketing campaign so you can monitor inbound carefully.
2

Run for 1-2 weeks

Listen to every call, refine the KB and prompts.
3

Monitor metrics

Containment rate (calls handled fully by AI), transfer rate, customer satisfaction (NPS via post-call SMS).
4

Cut over your main number

Once metrics look good — typically 70%+ containment with positive NPS — point your main DID at the agent.

​
Ongoing refinement

Plan to spend 30-60 minutes/week reviewing flagged calls and refining the agent for the first month. After that, weekly review of outliers is enough.

​
Cost monitoring

The AI agent uses LLM tokens and TTS/STT. Usage and cost are visible in the Billing tab. A typical 3-minute conversation costs roughly the same as a few sales emails. See Monitor Usage.

​
Troubleshooting

Try a different persona — voice quality varies meaningfully. Also keep the agent’s response prompts short; long replies sound more synthetic.
Make sure escalation rules include “if caller asks for a person” and that the transfer destination is configured. Test the escape hatch yourself.
Check the call transcript for the cited KB source. Update the source, or add a more direct phrasing. The KB is what grounds factual answers.
Voice-activity detection may be too aggressive in noisy environments. Lower the noise sensitivity in advanced settings.

​
Next steps

Voice agent reference

Full feature reference.

Compliance

Recording controls and retention for AI-handled calls.