How to build a great AI agent in the Build Your Agent step
A practical guide to filling out Build Your Agent so your BookFlow voice setter converts leads, handles objections, and sounds on-brand.
Why this step matters more than everything else
Your AI voice agent is only as good as the knowledge you give it during Build Your Agent. Short, generic inputs produce short, generic conversations. Specific, honest, detailed inputs produce conversations that feel like your best rep having her best day.
Expect to spend 10–15 minutes on this step. It is worth it — the agent will reuse what you write here on every single call.
The seven inputs that matter most
1. Business name and industry
Obvious but load-bearing. The agent uses them in opening lines ("Hi, this is Sarah calling from \{business name\}...") and to shape which vocabulary is appropriate. If your industry is not in the dropdown, pick **Other** and write a specific label like "Dental implants clinic" or "SaaS data security" rather than "Services."
2. Service description
Answer three questions in 2–4 sentences:
- What do you sell?
- Who do you sell it to?
- How does the buyer usually come to you? (inbound form, Meta Ads, referral, cold outbound)
Example: *We sell all-on-four dental implants to US adults 45+ who have been putting off full-mouth replacement. Most leads come from Meta Ads after seeing a before/after reel. They usually call wanting to know the cost before committing to a consult.*
3. Ideal customer (ICP)
This is the single biggest fit filter. The agent will decide whether to book the meeting or politely nurture the lead based on what you write here. Be specific about budget band, geography, urgency, and disqualifying factors.
Example: *Ideal: US, within 50 miles of one of our four clinics, budget $15k+, medically cleared, ready to start within 90 days. Not a fit: outside service area, looking for cosmetic veneers only, or insurance-only patients (we do not accept Medicaid).*
4. Guardrails
Tell the agent what it must **never** say. This is where you protect yourself from AI going off-script.
Example: *Never quote specific prices beyond the \$15k starting range. Never promise medical outcomes. Never discuss competitor pricing. Never agree to a payment plan without explicit approval from a human.*
5. Services & pricing
Be specific. The agent will quote these numbers verbatim and will not invent numbers outside this list.
Example: *Full-mouth All-on-4 starts at $15,000 per arch ($30,000 both arches). Includes consult, 3D scan, surgery, and provisional teeth. Financing available through CareCredit 0% for 18 months. Premium zirconia upgrade: add $4,000 per arch.*
6. FAQs (minimum 6 pairs)
The more high-quality FAQs you write, the more natural the agent sounds. Think about the top 10 questions you get on real calls and write the answer exactly as your best rep would say it.
7. Objection handling
When a lead says "it's too expensive," "I need to think about it," or "I'm talking to another clinic" — what does your best rep say? Write those scripts. The agent will follow them word-for-word.
Things that trip up first-time users
**Writing too short.** A 15-word ICP produces a 15-word-level qualification conversation. Aim for 40–80 words per field.
**Contradictions.** If guardrails say "never quote prices" but services-and-pricing has full prices, the agent will not know which to follow. Decide which rule wins and remove the other.
**Assuming the agent knows "standard stuff."** It does not. If you do not write "we accept credit card and ACH but not PayPal," the agent will say it accepts PayPal the moment a lead asks. Every specific must be in writing.
**Forgetting the handoff phone.** If the agent cannot answer a question, it offers to transfer the caller to a human. That needs to be a real number that gets answered. Put your best inbound line there.
Iterate weekly
The first version of your agent will be 80% right. After your first 20 real calls, review the transcripts (Dashboard → Recordings), find patterns where the agent stumbled, and update the relevant field. Changes take effect on the next call.
Most teams iterate once a week for the first month, then monthly after that.
Questions?
Hit the support widget in the corner. We will look at your specific Build Your Agent setup and give you pointed feedback.