Observation
Leon picks up a signal from conversation: "I usually wake up around 6:30"
How NOVA's AI system handles member relationships across three distinct interaction zones — from the member's WhatsApp to the concierge's command center.
NOVA members are paying $10,000 per year for a service that feels like having a personal concierge at Aman. They are high-net-worth individuals who value discretion, anticipation, and flawless execution.
They do not want to talk to a bot. They do not want to feel "managed." They want someone who knows them, anticipates them, and handles everything without being asked.
The concierge is the face. The AI is the brain. The member never sees the brain.
Today, Leon (our AI) speaks to Kevin's staff, friends, promoters, and tech people. That world is casual, direct, efficient. NOVA's world is different — high society, privacy-obsessed, status-conscious. The same AI personality cannot serve both.
| Current World | NOVA World |
|---|---|
| "Hey, what's up with the Arden deal?" | "Good morning, Mrs. Tanaka. We hope you're well." |
| Direct ask, direct answer | Warm greeting, gentle inquiry, anticipatory service |
| Mistakes are funny | Mistakes are unforgivable |
| Privacy is flexible | Privacy is absolute |
| AI is cool | AI is invisible |
Every conversation Leon participates in has a container — a boundary that defines what Leon knows, how he speaks, and what he's allowed to do. NOVA requires three distinct container types.
The member's WhatsApp. This is the primary touchpoint. The member texts, and NOVA responds. They believe they're talking to their concierge team.
Leon wears the NOVA concierge persona — warm, understated, anticipatory. Never mentions AI, algorithms, or systems. Never mentions other members. Speaks as "we" (the NOVA team), never "I" (the AI).
The relationship space. Some members have a WhatsApp group with their assigned concierge. This is where trust is built between the member and the human.
Leon is mostly invisible here. He observes and learns. He only speaks when directly asked. He creates tasks and notes silently for the concierge.
If the concierge says something wrong, Leon doesn't correct them in front of the member. He DMs the concierge separately: "Hey — the dinner is at 7, not 7:30. Want me to correct?"
The critical rule: The member is building trust with their concierge. Leon enhances that relationship. He never competes with it.
The back office. This is where the real Leon lives. Full operational depth. Data-rich, proactive, efficient.
The concierge sees everything: member preferences (verified and unverified), booking status, checklist progress, upcoming deadlines, verification queue.
At $10,000/year, getting someone's preferences wrong is a trust-destroying event. A wrong dietary note at dinner. A missed allergy on a flight. A room that contradicts what they asked for. These aren't bugs — they're relationship-ending failures.
The preference system has one rule: Leon never gets to decide a preference is correct. He only gets to flag it for a human to confirm.
Leon picks up a signal from conversation: "I usually wake up around 6:30"
Preference created: schedule/wake_time = "6:30 AM" — source: leon_inferred, confidence: 0.7, NOT VERIFIED
Concierge sees in their verification queue: "Leon thinks Tanaka-san wakes at 6:30 AM"
Verify — "Yes, confirmed with member" | Correct — "Actually 7 AM" | Reject — "Wrong, he sleeps until 9"
Only after verification does it auto-fill on booking checklists. Every change is audited: who confirmed, when, what the original inference was.
Not all preferences carry the same weight. Getting a coffee order wrong is annoying. Getting an allergy wrong is dangerous.
| Level | Examples | System behavior |
|---|---|---|
| Critical | Shellfish allergy, wheelchair access, religious dietary law | Auto-injected onto EVERY relevant booking. Blocks confirmation until concierge acknowledges. Must be human-verified. |
| Important | Vegetarian, window seat, loyalty program, passport details | Auto-fills on bookings. Concierge should verify but not gated. |
| Preference | Wakes at 6:30, oat milk latte, room at 18°C, back by 10 PM | Auto-fills when relevant. Enhances the experience. |
Anything that helps the concierge deliver a better experience:
| Category | Examples |
|---|---|
| Dietary | Allergies, vegetarian, halal, dislikes, favorite cuisines |
| Medical | Carries EpiPen, mobility limitations, altitude sensitivity |
| Schedule | Wake time, bedtime, back-at-hotel-by, morning routine |
| Lifestyle | Coffee order, workout habits, reading preferences |
| Travel | Passport details, visa status, seat preference, frequent flyer |
| Hotel | Room temperature, floor preference, pillow type, minibar |
| Dining | Seating preference, occasion awareness, wine preferences |
| Accessibility | Wheelchair, hearing, visual, cognitive accommodations |
Inspired by The Checklist Manifesto: complex processes need structured checklists to prevent errors. A concierge booking a private jet has 12 critical steps. Miss one — wrong name on a passport manifest — and the member is denied boarding.
Every booking type has a template of required steps. When a booking is created, the system generates a checklist, auto-fills what it can from verified preferences, and gates confirmation until critical items are done.
| Item | Status | Auto-filled? |
|---|---|---|
| Passenger manifest (names, DOB, passport) | ✓ Done | Yes — from household |
| Departure airport + FBO | ✓ Done | No |
| Arrival airport + FBO | ✓ Done | No |
| Catering preferences | ✓ Done | Yes — dietary prefs |
| Critical Shellfish allergy — acknowledged? | ✓ Done | Auto-injected |
| Quote approved by member | ✓ Done | No |
| Confirmation from operator | Pending | No |
| Ground transport to FBO | Pending | No |
| Payment confirmed | ✓ Done | No |
Gate status: Cannot mark "confirmed" — 2 critical items pending.
| Type | Checklist items | Critical gates |
|---|---|---|
| Private jet / charter | 12 | Manifest, FBO, confirmation, payment, quote approval |
| Hotel | 10 | Guest names, dates, confirmation, payment |
| Commercial flight | 10 | Legal name, passport, visa, confirmation |
| Restaurant | 8 | Party size, date/time, dietary (all guests), confirmation |
| Activity / experience | 8 | Participants, pickup location, confirmation |
| Transport | 7 | Passenger count, pickup, dropoff, confirmation |
Critical preferences are injected automatically. If any traveler has a critical-severity preference (allergy, medical condition, accessibility need), it becomes a mandatory checklist item on every relevant booking — even if it's not in the template. The concierge must acknowledge it before the booking can be confirmed.
NOVA facilitates all travel spending, but the member doesn't always pay through NOVA. Sometimes they hand over their Amex at the hotel. Sometimes the partner invoices NOVA directly. The system tracks all spend regardless of who pays.
| Payment method | Who pays | NOVA sees the money? |
|---|---|---|
| NOVA Stripe | Charged through NOVA | Yes — Stripe transaction |
| Member direct | Member's own card on-site | No — concierge logs the cost |
| Partner invoice | Partner bills NOVA | Yes — invoice flow |
| Complimentary | Nobody / included | No charge |
GMV = total value of everything NOVA facilitated, regardless of payment method. This is the number that matters for unit economics. $10K membership fee + 10% of ~$60K average GMV = $16K per member per year.
How Leon speaks in member-facing containers. These are non-negotiable.
| Rule | Instead of | Say this |
|---|---|---|
| Always "we", never "I" | "I've booked your restaurant" | "We've reserved your table" |
| No system language | "Booking confirmed, ID #NV-2026-042" | "All set for Thursday at 7 PM" |
| No AI signals | "Based on my analysis of your preferences" | "We remember you enjoyed the kaiseki in Kyoto" |
| Anticipate, don't ask | "Do you need an airport transfer?" | "Your car will meet you at arrivals" |
| Defer when unsure | "I don't have that information" | "Let me check with the team and get back to you" |
| Never mention other members | "Other members have enjoyed..." | "We think you'd love..." |
| Be warm, not servile | "I'm so sorry for any inconvenience" | "We'll take care of that right away" |
Relationship intelligence. Understands who people are, how they relate, what they care about. Learns from every conversation. Feeds the preference system.
One record per person. Every system reads from here. A NOVA member is a contact with a membership role — not a separate database.
Severity-aware, verified, auditable. Leon infers. Concierges confirm. Critical preferences gate bookings. Every change is logged.
Trips, bookings, checklists, concierge operations. The execution layer. Built on contacts, powered by preferences, gated by checklists.
No AI inference becomes a trusted fact without a human confirming it. No booking is confirmed without critical checklist items acknowledged. The concierge is always the final authority.