ATLANTIS UNIVERSITY × FONTAINEBLEAU MIAMI BEACH
Responsible AI for Hotel Leaders
Boost productivity · Practical hospitality applications · Safe & ethical adoption
Presented by Prof. Arthur Lyons & Prof. Rodolfo Capdevilla
with Sam Glynn, Director of Admissions & Development · Atlantis University · July 22, 2026
Who we are
Atlantis University
A Miami-based university built around career-ready, practical education — pairing academic insight with real-world professional experience.
- School of Business — strategy, marketing & leadership
- Master's in Artificial Intelligence — applied, industry-facing
- Today's workshop delivered by AU faculty
"Intentionally different — your credible path to your full potential."
Your Atlantis contact: Sam Glynn, Director of Admissions & Development
Your presenters
Two faculty, one practical lens on AI
AL
Arthur Lyons
Faculty Chair & Professor · School of Business
A business strategist and marketing executive with decades of experience helping organizations grow through effective marketing and customer engagement — combining real-world expertise with academic insight to deliver practical, actionable strategy.
Education: MBA, University of Miami School of Business · Graduate Studies, Georgetown University · Presenter, Direct Marketing Institute & the American Marketing Association (AMA).
RC
Rodolfo Capdevilla
Faculty Lead · MS in Artificial Intelligence
Originally from Colombia. PhD in Physics from the University of Notre Dame, with years spent analyzing data from particle colliders — finding vanishingly small signals inside enormous, noisy datasets — now applied to practical, responsible AI.
Education: PhD, Physics, University of Notre Dame · Faculty Lead for the Master's in Artificial Intelligence at Atlantis University.
"Teaching computers to find needles in the world's biggest haystacks — the particle colliders."
Today's roadmap
Five parts — one idea to carry all session
The throughline: AI is a machine that learns — make it work for you, without handing over sensitive data.
1
What AI actually is
A machine that learns · demos
2
Large language models
How ChatGPT works — and fails
3
AI in action at Fontainebleau
Six opportunities + live demo
4
Putting AI to work
Prompting + your hands-on turn
5
Using AI safely
Real cases, guardrails, next steps
Part 1
What AI Actually Is
A machine that learns from examples — not magic, not rules.
The core idea
AI is a machine that learns
AI is a bunch of nerds teaching computers to do tasks humans do — by learning from thousands of examples, not by being programmed rule-by-rule.
↑
Techniques, Tools, Training, …
The core idea
AI is a machine that learns
"We do not want to look inside the model!"
The core idea
AI is a machine that learns
↑
Techniques, Tools, Training, …
Takeaway: AI is excellent at narrow tasks it has learned — but it is not infallible reasoning.
Demo 1 · Computer vision
A computer that learned to see
Demo
YOLO detects and labels objects in a live video feed in real time — it was never told the rules for a "person" or a "bag," it learned them from examples.
▶
Live detection demo (video backup ready)
Where this helps at Fontainebleau
- Inventory & stock counts from a camera feed
- Security & anomaly detection (unattended items, crowding)
- Housekeeping & room-readiness / occupancy signals
- Spotting spills or hazards on the floor faster
Demo 2 · Speech AI
A computer that translates — live
Demo
I speak Spanish; the computer transcribes and translates it to English in real time — a model that learned language from millions of examples.
ES "Buenas tardes, ¿en qué puedo ayudarle?"
↓
EN "Good afternoon, how may I help you?"
Where this helps at Fontainebleau
- Multilingual guest service across a diverse Miami clientele
- Staff training & accessibility for a multilingual workforce
- Live captioning for events in the ballroom
- Faster, clearer front-desk & concierge interactions
Part 2
Large Language Models
The ChatGPT / Copilot family — powerful, and worth understanding.
How it works
LLMs are glorified auto-completion engines
A large language model breaks text into "tokens" and predicts the most likely next one, over and over. That's the whole trick — repeated at massive scale.
Theguestinroom1204wouldlike…
← each box is a "token"; the model guesses the next one
Brilliant at
- Drafting emails, replies & summaries
- Rewriting tone, translating, brainstorming
- Turning messy notes into clean structure
But remember
- It doesn't "know" facts — it produces plausible text
- It is stochastic: same prompt, different answers
- Confidence ≠ correctness
The catch
It can be confidently wrong
Because it generates plausible text rather than looking up truth, an LLM can invent details — a "hallucination" — and state them with total confidence.
Why it matters in a hotel
- Invented policies, rates or availability mislead guests
- Inconsistent answers confuse staff across shifts
- A wrong "official" answer can become real liability
- Fabricated citations or figures in a report
The rigor habit (a physicist's reflex)
- Never ship a high-stakes output unverified
- Cross-check with a second model as an "auditor"
- Keep a human in the loop for anything guest-facing or legal
- Ground answers in your own approved documents
Part 3
AI in Action at Fontainebleau
Six opportunities across the resort — presented by Prof. Arthur Lyons.
The opportunity
Six ways AI creates value at Fontainebleau
Personalization at scale · operational intelligence · revenue optimization — applied to six concrete parts of the business.
Opportunity 1
Guest-Facing Operations
An AI concierge agent, available 24/7 across every channel.
Opportunity 2
Revenue Optimization
Dynamic pricing beyond room rates, across every outlet.
Opportunity 3
Personalization & Loyalty
Predicting guest needs before they're voiced.
Opportunity 4
Operational Excellence
Housekeeping & maintenance, coordinated by AI.
Opportunity 5
Marketing & Brand
Real-time sentiment & reputation monitoring.
Opportunity 6
Sales & Event Planning
AI-assisted proposals and lead prioritization.
Opportunity 1 · Guest-facing operations
AI-powered guest concierge agent
24/7 multichannel availability
The AI concierge operates around the clock via SMS, WhatsApp, app, and in-room tablets to assist guests anytime.
Enhanced guest service efficiency
It manages routine requests quickly, escalating complex issues to staff, improving response speed and consistency.
Personalized recommendations and revenue
The AI suggests upgrades and experiences based on guest context, creating new revenue opportunities for the resort.
Supporting human staff
By filtering routine interactions, the AI enables concierges to focus on personalized, high-value guest relationships.
Opportunity 2 · Revenue optimization with AI
Beyond room rates
AI-driven pricing across services
AI extends dynamic pricing beyond rooms to spas, cabanas, pools, nightlife, and premium amenities in the resort.
Maximizing guest spend
AI analyzes demand, trends, weather, and guest segments to optimize total guest spend across all outlets.
Managerial benefits
Managers benefit from moving to responsive pricing strategies that boost revenue in peak times and improve utilization off-peak.
Brand-aligned pricing rules
AI-driven pricing follows brand guidelines to maintain luxury positioning and ensure guest satisfaction.
Revenue impact example
Weather-driven demand shifts
AI responds to weather changes
AI detects weather conditions and adjusts pricing and promotions automatically to optimize guest experience and revenue.
Proactive guest engagement
AI offers alternative indoor experiences like spa and wellness classes to guests, enhancing satisfaction during bad weather.
Revenue protection strategy
AI-driven pricing and bundling maintains occupancy and protects revenue that might be lost due to cancellations or dissatisfaction.
Improved manager efficiency
AI reduces last-minute manual management efforts, enabling staff to focus on service excellence despite external changes.
Opportunity 3 · Personalization and guest loyalty
Predictive guest personalization with AI
AI anticipates guest needs
AI analyzes historical and real-time data to predict and fulfill guest preferences before requests are made.
Individualized guest experiences
Personalization moves beyond segmentation to tailor wellness, dining, and service packages uniquely for each guest.
Managerial benefits and privacy
AI-driven personalization enhances satisfaction and loyalty without added workload while respecting guest privacy and trust.
Personalization example
Pre-arrival upsell strategy
AI-based guest profiling
AI analyzes guest data to identify high-value customers with specific interests like spa and dining.
Personalized pre-arrival offers
Guests receive customized bundled spa and dining offers before arrival, enhancing relevance and appeal.
Revenue and engagement benefits
This strategy increases conversion rates and reduces manual outreach through automated, targeted messaging.
Opportunity 4 · Operational excellence
Operations AI agent for housekeeping and maintenance
Predictive task coordination
AI forecasts checkout times and prioritizes room cleaning for efficient housekeeping management.
Maintenance issue detection
AI flags maintenance problems early to prevent service failures and minimize disruptions.
Optimized staff routing
AI optimizes staff movement, reducing unnecessary travel and overtime in housekeeping and maintenance.
Managerial decision support
AI supports supervisors by recommending actions while maintaining human oversight and accountability.
Operations example
Predicting late checkouts
Predictive AI for checkouts
AI analyzes guest data to predict late checkouts and improves operational decision-making.
Optimized housekeeping scheduling
Housekeeping schedules adjust automatically to match predicted room availability, reducing delays.
Benefits for staff and guests
Staff get clearer priorities and manageable workloads, while guests experience timely room readiness.
Proactive management impact
AI enables supervisors to focus on coaching and quality assurance, improving satisfaction and margins.
Opportunity 5 · Marketing and brand preference
AI sentiment and reputation monitoring
Real-time sentiment analysis
AI continuously analyzes guest feedback to detect trends and issues immediately, enhancing responsiveness.
Operational decision support
Managers use AI insights to address service problems quickly, improving guest satisfaction and operations.
Brand protection and improvement
AI-driven reputation monitoring supports brand equity and fosters continuous improvement based on guest feedback.
Opportunity 6 · Sales and event planning
AI-powered sales and event planning assistant
Customized proposal generation
AI generates tailored proposals quickly using real-time availability, pricing, and historical data, enhancing response speed.
Lead prioritization and revenue optimization
AI estimates booking probability and recommends upsells, helping managers focus on high-value leads and increase event revenue.
Manager control and brand alignment
Sales managers retain control over final proposals, ensuring client expectations and brand standards are met.
AI in action · wrap-up
Key takeaways
AI as hospitality enabler
AI enhances hospitality services without replacing the essential human touch in guest experiences.
Focus on operational challenges
Successful AI initiatives address specific operational or revenue challenges rather than adopting technology for novelty.
Pilot programs and scaling
Pilot programs help test AI tools, measure outcomes, and build confidence before wider implementation.
Collaborative leadership
Cross-department collaboration ensures alignment and governance for successful AI adoption.
Live demo
Training simulation demo
A working hotel dashboard and AI assistant, built with the same everyday tools you already own — Microsoft 365, Copilot, and data from systems like Opera.
Part 3 · presented by Prof. Arthur Lyons
Part 4
Putting AI to Work
The productivity you ranked #1 — with prompts you can use tomorrow.
Prompt engineering
Better prompts = better, safer results
Role+Task+Context+Constraints+Examples+Output format
WEAK PROMPT
"Write an apology email to a guest."
Generic, off-brand, may invent compensation you don't offer.
Prompt engineering
Better prompts = better, safer results
Role+Task+Context+Constraints+Examples+Output format
WEAK PROMPT
"Write an apology email to a guest."
Generic, off-brand, may invent compensation you don't offer.
STRONG PROMPT
"You are a Fontainebleau front-desk manager. Draft a warm 120-word service-recovery email to a guest kept awake by noise. Offer only spa credit or late checkout. Professional, apologetic, no admission of legal fault."
Your turn
Hands-on: try it yourself
Exercise 1 · Hands-on
Golden rule: use only fictional or anonymized details — never real guest names, rooms, or payment info.
Pick one task
- Draft a service-recovery email (use the strong-prompt recipe)
- Turn a long policy into a 5-point staff checklist
- Translate a guest notice into 3 languages*
- Summarize sample meeting notes into action items
* Look at the FB website for policies.
Fill-in template
You are a [role].
Write a [task] about [context].
Keep it [constraints: length / tone].
Format as [email / list / table].
Stuck? Grab a ready-made prompt from the handout.
Original content
Generate images you own — and the legal reality
Exercise 2 · Demo
Live: generate a campaign visual
- Prompt an elegant Fontainebleau summer-gala flyer
- Original asset — not scraped from Google Images
- Great for flyers, socials & internal decks
Why generate, not grab?
Random Google Images often carry copyright and infringement risk. Generate your own and you control the asset.
THE COPYRIGHT REALITY
Purely AI-generated works lack human authorship and are not copyrightable. But when you add significant human creative input — detailed prompts, selection, editing, arrangement — you can claim copyright in the human-authored elements. You own the outputs per the tool's terms.
Original content
Generate images you own — and the legal reality
Demo
Text generation template:
You are a [role].
Write a [task] about [context].
Keep it [constraints: length / tone].
Format as [email / list / table].
Stuck? Grab a ready-made prompt from the handout.
→
Image generation template:
You are a [role].
Generate an image of [main subject / scene].
Include these key visual elements: [specific details].
Style & mood: [artistic style, lighting, atmosphere, tone].
Composition & layout: [framing, balance, focal point, space for text].
Color palette: [specific colors or vibe].
Text to include (if any): [event name, date, tagline – make it legible and integrated].
Constraints: [high resolution, professional quality, elegant, original design, no text overflow, suitable for print/digital].
Make it suitable for [use case: flyer, social media post, campaign visual, etc.].
Original content
Generate images you own — and the legal reality
Demo
Ask AI for the prompt
You are a professional luxury hospitality graphic designer.
Generate an image of an elegant promotional flyer for a summer gala event at the Fontainebleau Miami Beach grand ballroom.
Include these key visual elements: magnificent crystal chandeliers, sophisticated guests in evening attire, elegant table settings with floral centerpieces, soft golden lighting, and a sense of celebration and luxury.
Style & mood: photorealistic yet artistic, warm and inviting luxury atmosphere, sophisticated and professional tone, high-end marketing aesthetic.
Composition & layout: centered elegant ballroom scene as the main focal point, clean balanced composition with generous negative space at the top and bottom for text overlay.
Color palette: soft golds, warm creams, deep navy accents, and subtle champagne tones.
Text to include (integrated naturally and legible):
"Fontainebleau Summer Gala"
"July 25, 2026 • Grand Ballroom"
"Experience an unforgettable evening of elegance and celebration"
Constraints: High resolution, clean professional composition, original design with no copyright-infringing elements, elegant and modern, suitable for both print flyers and digital marketing.
Make it suitable for a high-end hotel promotional flyer and social media campaign.
Original content
Generate images you own — and the legal reality
Demo
Grok:
Gemini:
Original content
Generate images you own — and the legal reality
Demo
Grok:
Gemini:
Try Notebook Gemini to copy a template
Part 5
Using AI Safely
A first look — enough to protect you now, with a deep dive to follow.
Real cases
When AI goes wrong — Evaluate the risk
Consequences of not being careful with private info in ChatGPT.
Front Desk / Guest Services
- Classify & route guest requests
- Draft service-recovery replies
- Multilingual phrasing & pre-arrival notes
→
Generic template — Safe!
Operations
- Summarize internal policy
- Create inventory / waste reports
- Meeting notes → action items
→
Is this a private document?
What sensitive information was discussed?
Could it be found by someone else?
“Could bad actors access what I upload?”
· Consumer tools may retain prompts — and can use them to train future models
· A provider breach or a compromised account exposes stored chat history
· Models can memorize and regurgitate unique text they were shown
· So assume anything you paste leaves your control — anonymize, or use governed tools
Interactive
Rank the risk — you decide
Interactive
For each scenario, call it out: Low · Medium · High · Critical — for data, legal, and reputation.
5 = high risk · 1 = low risk
Risk Scale
1
Pasting a guest complaint with name & room into public ChatGPT for a reply
2
Using a public AI to brainstorm internal pricing strategy with real targets & competitor data
3
Using a personal “shadow AI” account to summarize staff performance notes
4
Launching a customer-facing reservations chatbot with no added guardrails or oversight
Interactive
Rank the risk — you decide
Interactive
For each scenario, call it out: Low · Medium · High · Critical — for data, legal, and reputation.
Risk Scale
1
Pasting a guest complaint with name & room into public ChatGPT for a reply
4/5
Clear PII leakage. Data goes to a public model that may retain or train on it. Strong privacy/compliance violation risk.
2
Using a public AI to brainstorm internal pricing strategy with real targets & competitor data
3/5
Competitive/IP risk if strategy leaks or gets absorbed into the model. Less sensitive than personal/guest data. Still serious for business, but lower immediate legal/compliance.
3
Using a personal “shadow AI” account to summarize staff performance notes
4/5
Sensitive employee/HR data. Very low auditability (personal accounts). Potential employment-law and privacy issues if exposed.
4
Launching a customer-facing reservations chatbot with no added guardrails or oversight
5/5
High chance of wrong information reaching guests. Very high blast radius because it’s customer-facing and real-time.
Real cases
When AI goes wrong — Recent cases
Samsung 2023
Engineers pasted confidential source code and meeting notes into public ChatGPT. The company banned generative-AI tools shortly after.
Public tools can leak your data.
BREAKING | BUSINESS · FORBES
Samsung Bans ChatGPT Among Employees After Sensitive Code Leak
By Siladitya Ray, Forbes Staff · May 2, 2023
Real cases
When AI goes wrong — Recent cases
Medium · CUT THE SAAS
Chatbot Case Study: Purchasing a Chevrolet Tahoe for $1
Puran Parsani · 5 min read · Jun 20, 2024
Chevrolet dealer bot 2023
A customer manipulated a dealership’s ChatGPT-powered chatbot into “agreeing” to sell a car for $1 and calling it legally binding.
No guardrails = brand & legal risk.
Real cases
When AI goes wrong — Recent cases
Air Canada 2024
A tribunal held the airline legally liable for wrong information its chatbot gave a grieving customer. “It was part of our website.” Damages awarded.
Your AI’s answers = your liability.
BBCTRAVEL
Airline held liable for its chatbot giving passenger bad advice — what this means for travellers
23 February 2024 · Maria Yagoda, Features correspondent
Real cases
When AI goes wrong — Recent cases
CX FOUNDATION MARCH 24, 2026 · 6 MIN READ
Chipotle's Customer Service AI Agent Goes Off the Rails, Follows Incidents at Amazon & Woolworths
Charlie Mitchell, Director of Content & Market Research
Chipotle chatbot 2026
Chipotle (2026): Customer chatbot was repurposed for coding tasks.
Guardrails failed = vulnerable to threats.
Guardrails that work · 1 of 2
Set the rules — then let tools enforce them
Layer 1 · Policy — rules people follow
A short, written list of AI do's and don'ts every team knows.
- Never paste guest PII, payment, financials or staff data into public tools — and verify outputs before you act
- Advantage: stops the most common leak — copy-paste — at zero cost
Guardrails that work · 1 of 2
Set the rules — then let tools enforce them
Layer 1 · Policy — rules people follow
A short, written list of AI do's and don'ts every team knows.
- Never paste guest PII, payment, financials or staff data into public tools — and verify outputs before you act
- Advantage: stops the most common leak — copy-paste — at zero cost
Layer 2 · Train your own local model
Train or fine-tune a model in-house — the JPMorgan route — on your own data, not the entire internet.
- Advantage: no data ever leaves your premises
- Trade-off: needs serious computing power to reach a good-enough model
Guardrails that work · 1 of 2
Set the rules — then let tools enforce them
Layer 1 · Policy — rules people follow
A short, written list of AI do's and don'ts every team knows.
- Never paste guest PII, payment, financials or staff data into public tools — and verify outputs before you act
- Advantage: stops the most common leak — copy-paste — at zero cost
Layer 2 · Train your own local model
Train or fine-tune a model in-house — the JPMorgan route — on your own data, not the entire internet.
- Advantage: no data ever leaves your premises
- Trade-off: needs serious computing power to reach a good-enough model
Layer 3 · Hosted private models
Data centers host your local model and supply the computing resources.
- Advantage: reasonably powerful models at a reasonable cost
- Your model and your data stay under your control
Guardrails that work · 1 of 2
Set the rules — then let tools enforce them
Layer 1 · Policy — rules people follow
A short, written list of AI do's and don'ts every team knows.
- Never paste guest PII, payment, financials or staff data into public tools — and verify outputs before you act
- Advantage: stops the most common leak — copy-paste — at zero cost
Layer 2 · Train your own local model
Train or fine-tune a model in-house — the JPMorgan route — on your own data, not the entire internet.
- Advantage: no data ever leaves your premises
- Trade-off: needs serious computing power to reach a good-enough model
Layer 3 · Hosted private models
Data centers host your local model and supply the computing resources.
- Advantage: reasonably powerful models at a reasonable cost
- Your model and your data stay under your control
Layer 4 · Prompt guardrails — e.g. NeuralSeek
A service that sits between your staff and ChatGPT / public models, “auditing” every prompt.
- Redacts PII, covers sensitive data, then sends the API request for you
- Advantage: staff keep using the tools they know — safely
Guardrails that work · 2 of 2
The gold standard — AI grounded in your own documents
What “grounded & governed” means
- Grounded: the AI answers only from your approved documents — policies, SOPs, rates — not from the open internet, and every answer cites its source so staff can verify it
- Governed: every question and answer is logged and auditable, PII is redacted automatically, and access follows staff roles
- The payoff: no invented policies, no data wandering off, and an audit trail Legal can review
Proven at scale
- Children's Health — hospital agents run fully on-premises, “no PHI leaves” — the strictest privacy bar, met
- NatWest Cora+ — 11.2M customer conversations in 2024, GDPR-governed and fully replayable — guest-facing AI can be compliant
- Penn State MyResource — ~90,000 students served with PII-safe, cited answers — grounded answers at scale
This is the taste. A dedicated security & implementation deep-dive is the natural next session.
Quick wins
Where each team can start Monday
Front Desk / Guest Services
- Classify & route guest requests
- Draft service-recovery replies
- Multilingual phrasing & pre-arrival notes
Rooms Division / Housekeeping
- Summarize inspection & maintenance reports
- Shift handover notes
- Scheduling & room-readiness ideas
F&B / Operations
- Menu descriptions & upsell scripts
- Summarize inventory / waste reports
- Demand-forecasting support
Revenue / Marketing & Leadership
- Promotional copy & campaign ideas*
- Summarize reports (anonymized)
- Meeting notes → action items
Wrap-up
Key takeaways & next steps
Remember this
- AI is a machine that learns — powerful, not infallible
- Six real opportunities across the resort — start where the pain is
- Strong prompts + human oversight = better, safer output
- Protect data with clear policy & guardrails
DO THIS WEEK
1 Draft a simple AI-use policy
2 Try 2–3 prompt templates from today
3 Pick one department pilot
Atlantis University is your partner next
1-hour follow-up Zoom Q&A · a dedicated security & implementation deep-dive · a custom internal assistant grounded in your own policies
Thank you
Questions & discussion
Prof. Arthur Lyons · Prof. Rodolfo Capdevilla · Sam Glynn
Atlantis University · School of Business & the MS in Artificial Intelligence
