All posts
Getting Started

Three Layers of Control: How to Customize Your AI Receptionist's Personality

Every AI receptionist sounds generic on day one. Yours doesn't have to. Three editable layers — Personality, Procedures, Knowledge — let you shape how the agent sounds, what it does, and what it knows. Edit in the dashboard. Changes take effect on the next call.

FTFrontdeQ Team
5 minutes read

Generic AI voice agents all sound the same. That's fine for a pizza chain. It's wrong for a salon. Your front-desk person knows your salon isn't a pizza chain — they know your regulars by name, know which stylist takes walk-ins on Tuesdays, know that Mrs. Patterson always asks about parking before she commits to a time.

FrontdeQ is designed around the same premise. The AI you get on day one isn't the AI you'll have on day thirty — because you shape it across three layers, each one editable from your dashboard, each one taking effect on the next call.

The Three Layers

When a call comes in, your AI's prompt is built from six sources stacked on top of each other. Three of those layers are yours to edit:

  1. Personality — the tone, warmth, style of speaking. The how.
  2. Procedures — the rules for what to do in different call situations. The what.
  3. Salon Knowledge — the salon-specific facts. The what it knows.

The other three (live Square data, current session context, client memories) are built automatically per call. You don't touch them. But the three above are yours.

Layer 1 — Personality

The personality layer is short. A handful of paragraphs defining voice, tone, and communication style. It sets the how of every sentence the AI produces.

A high-end Miami salon:

You are Marissa, the receptionist at Luxe Midtown. You speak with warmth and polish. You never rush the caller. You remember that this salon's clients expect to be treated like the regulars they are, even if it's their first call.

A neighborhood barbershop:

You're the front desk at Mike's. Keep it short, keep it friendly, no unnecessary small talk. Match the caller's energy. Always confirm the barber before booking.

Both accurate. Both completely different. The personality layer is where that difference lives.

Layer 2 — Procedures

Procedures define the call flows. How to handle a booking. How to handle a cancellation. What to do when the salon is closed. When to escalate to a human. When to take a voicemail instead.

This layer is more structured — it's essentially a set of numbered steps for each call type:

## Cancel Flow
1. Identify the caller by phone number.
2. List their upcoming bookings.
3. Confirm the specific appointment.
4. Get an explicit "yes" before cancelling.
5. Call cancel_booking with the booking ID.
6. If that fails (merchant policy, past window), use take_message
   with urgency = urgent.
7. Confirm warmly: "You'll get a confirmation from Square."

You don't write this from scratch. FrontdeQ ships with a default procedures layer that covers booking, rescheduling, cancellation, pricing questions, FAQs, escalations, and voicemail handoffs. What you customize is the specifics for your salon. "Only take walk-ins with Sofia on Thursdays." "Always offer the new deep-conditioning service to color clients." "If they ask about our old location, tell them we moved to Midtown in February."

Layer 3 — Salon Knowledge

This layer is the most concrete. It's the facts your AI needs to know that aren't in Square.

  • Parking. "Free lot behind the building, entrance on Oak Street. Two-hour limit, chalked by the city."
  • Policies. "24-hour cancellation notice required. Late arrivals (15+ minutes) may result in shortened service."
  • Staff context. "Sofia is senior stylist — color and extensions. Cristina specializes in curly hair. Maria takes all walk-ins on Tuesdays."
  • Product questions. "We use Olaplex for bond treatments. Kerastase for hair care. Both sold in-salon and online."
  • Services beyond Square. "We do weddings — packages start at $400, includes hair and makeup for the bride, available by appointment only."
  • Seasonal notes. "Closed Memorial Day, July 4, Thanksgiving, Christmas. Limited hours Christmas Eve and New Year's Eve."

Whatever a great receptionist learns in their first month — the parking spot, the allergy-friendly products, the stylist who's great with first-timers, the answer to "do you do wedding hair?" — goes here. The AI pulls from this layer whenever a caller asks something that isn't a simple booking.

How the Layers Combine

Every inbound call assembles a fresh prompt from all six sources: the three you edit (personality, procedures, knowledge), plus live Square data, the current time/status, and any memories from the caller's past calls.

You never see the combined prompt unless you want to. You edit each layer independently in the dashboard — Settings → Integrations → Identity Stack. Changes are versioned, so you can see what you changed and when, and roll back if something didn't land well.

Versioning and Rollback

Every time you save a layer, a new version is created. The dashboard shows the history:

Personality layer
  v4 — "Softened greeting for first-time callers" — 3 days ago — you
  v3 — "Added warmth with frustrated callers" — 1 week ago — system
                                                             (nightly QA)
  v2 — "Initial customization for Luxe" — 2 weeks ago — you
  v1 — "Default" — 3 weeks ago — system

If version 4 made things worse, roll back to v3 with one click. The live agent updates within seconds.

Don't Start from Zero

The default ships are good enough to take calls on day one. Leave them alone and the AI still answers the phone, books appointments, handles FAQs. Your customization is additive — you're tuning, not building from scratch.

Start with the knowledge layer. It's the most concrete and the easiest win. Add your parking instructions, your late-cancellation policy, your stylist specialties. That alone covers 80% of the edge cases your salon has that a generic AI won't handle.

Come back to the personality layer once you've heard the AI on a few real calls. You'll notice it's too formal, or too chatty, or too quick. Tune it.

Leave the procedures layer alone until you hit a specific flow you want to change. The defaults are tested; editing them without a reason usually makes things worse.

The Loop

As your AI takes more calls, the nightly QA system proposes specific edits to each layer. It shows you: "Across 6 calls, callers asked about Saturday hours and the agent hedged. Add hours guidance to the Knowledge layer?" One click to accept. The layer updates, a new version is saved, and the agent uses the new knowledge on every call after.

The three layers are the foundation. The self-improvement loop is what takes them from "ok" on day one to "sounds like our best staff" on day thirty.

Learn more: /docs/identity-stack.