David Kong, Former CEO of Best Western Hotels, is my podcast guest today. Tune into Spotify & Apple Podcast now!
Why this episode matters: David Kong didn’t just lead Best Western for nearly two decades—he reshaped it. In this conversation, one of the most respected CEOs in hospitality opens up in ways he rarely has before: about leadership, identity, DEI backlash, brand strategy, and the decisions he still thinks about today.
What you’ll hear:
Why David believes “talent is evenly distributed, but opportunity is not”—and how unconscious bias still shapes leadership rooms today
The real story behind Best Western’s transformation: brand segmentation, quality resets, and bold portfolio moves
His candid take on DEI, quotas, and why the backlash was predictable—but not the solution
The decision he couldn’t get approved that still haunts him—and how it could have changed Best Western’s future
Why scale, loyalty programs, and data may determine who wins (and loses) in an AI-driven travel economy
Why listen: This isn’t a legacy victory lap. It’s a masterclass in long-term leadership, reinvention, and telling the truth after the seat is empty. Required listening for owners, operators, and anyone thinking seriously about the future of hospitality.
As more hoteliers get access to AI, it is critical that they are learning how to effectively leverage AI. Today, I am pleased to share with you a quick, tactical guide for GMs on the art of “prompting” with templates that you can copy and paste today.
Eager to hear your feedback.
Cheers,
Sloan
A Tactical Guide on AI Prompting for Hotel General Managers
Why Most Hotel Teams Get Mediocre AI Results
The difference between AI that saves you 10 hours a week and AI that wastes your time comes down to one thing: how you ask. Most hoteliers type a simple question and get a generic answer. Worded in a different way: A simple question into ChatGPT will yield a simple response from ChatGPT. Remember, the AI has no idea you run a 300-key full-service property, that your GOP margins are under pressure, or that you have a union workforce.
This guide gives you frameworks and ready-to-use prompts that actually work for hotel operations. No theory. Just tactics.
The RISEN Framework: Your Go-To Prompt Structure
Taken from my learnings at MIT this semester, the ‘RISEN’ acronym is the most practical framework for hotel operations prompts. It stands for:
R - Role | Tell the AI who it should be. "Act as a hotel revenue manager with 15 years of experience." |
I - Instructions | State clearly what you need. "Create a weekly labor schedule for my front desk." |
S - Steps | Break down what you want covered. "First analyze our occupancy patterns, then allocate shifts." |
E - End Goal | Define success. "The goal is to reduce overtime by 15% while maintaining service levels." |
N - Narrowing | Add constraints. "Maximum 500 words. Use bullet points. Focus on actionable items only." |
RISEN in Action: A Real Hotel Example
WEAK PROMPT: "Draft up responses to these hotel guest reviews." RISEN-STRUCTURED PROMPT: 1. Act as a hotel operations consultant who specializes in guest experience optimization (Role) 2. Analyze the following 50 guest reviews from our 280-room full-service Marriott property and identify the top 5 operational issues hurting our scores. (Instruction) 3. (1) Categorize complaints by department, (2) Identify root causes, (3) Recommend specific fixes with estimated costs. (Steps) 4. I need to present this to ownership next week to justify a $50K capex request. (End Goal) 5. Focus only on issues that appear 3+ times. Format as a table. (Narrowing) |
For More Complex Decisions: Include Chain of Thought
When working through a multi-step problem, add this phrase to the end of your prompt:
"Let's think through this step-by-step." |
This simple addition dramatically improves AI reasoning on complex questions. Use it for pricing decisions, staffing analysis, budget allocation, and any multi-variable problem.
Example: Revenue Management Decision
"We have a citywide convention arriving March 15-18. Our current occupancy forecast is 65% those nights. The convention has requested a room block of 80 rooms at $189/night (our rack is $249). Our variable cost per occupied room is $42. Competitors are selling out. Should I accept this block, negotiate different terms, or hold inventory? Let's think through this step-by-step." |
Remember: AIs like ChatGPT are not here to make decisions on your behalf but to save you time in researching the best decisions to drive the best outcomes for your operations. AI is not here to replace your role but to enhance your ability to make the best decisions.
AI Hallucinations: What They Are and How to Prevent Them
AI "hallucination" is when the model generates information that sounds completely confident but is actually fabricated or incorrect. It's not lying—it simply doesn't know when it's wrong. As a GM, you need to understand this limitation to use AI safely.attributed quotes to people who never said them.
⚠ REAL-WORLD HALLUCINATION EXAMPLES
In 2023, an attorney submitted a legal brief citing court cases that ChatGPT had completely invented. He was fined and sanctioned.
AI has fabricated statistics, invented citations, created fake URLs, and attributed quotes to people who never said them.
The AI will present these fabrications with the same confident tone as accurate information.
Why Hallucinations Happen
AI doesn't "know" facts—it predicts the most likely next words based on patterns. When asked something outside its training data, or when the question is ambiguous, it generates plausible-sounding but potentially false content. It's optimized for coherence, not truth.
Anti-Hallucination Prompt Techniques
Add these phrases to your prompts to reduce the risk of fabricated information:
ADD TO YOUR PROMPTS:
"Do not hallucinate. Do not make up facts or statistics."
"If you don't know the answer or aren't certain, say 'I don't know' or 'I'm not sure.'"
"Only provide information you are confident about. Do not invent sources or citations."
"If I ask for statistics, only provide them if they come from your training data. Otherwise, tell me you don't have that specific data."
The "What Did I Miss?" Technique
After getting a response, ask the AI to check its own work. This catches blind spots and errors:
FOLLOW-UP PROMPTS AFTER GETTING A RESPONSE:
"What did I miss? Are there important considerations I haven't thought about?"
"What could go wrong with this plan that we haven't discussed?"
"Are there any assumptions in your response that I should verify?"
"What information would you need to give me a more confident answer?"
"Are there any assumptions in your response that I should verify?"
"What information would you need to give me a more confident answer?"
High-Risk vs. Low-Risk AI Tasks
✓ LOWER RISK (AI is helpful) | ✗ HIGHER RISK (Verify everything) |
Drafting emails and responses Brainstorming ideas Structuring presentations Analyzing data YOU provide Creating training outlines Writing job descriptions | Specific statistics or metrics Legal or compliance information Brand standards or policies Competitor pricing (use your data) Citations or sources Recent events or news |
Golden Rule: Treat every AI response as a first draft that requires your expert review. The AI handles the heavy lifting; you provide the judgment and fact-checking.
Ready-to-Use Prompts by Function
Below are a few templates you can use today. Replace the bracketed items with your specifics.
Revenue Management
Comp Set Analysis: "Act as a revenue manager. Analyze these rate screenshots from [competitor names]. Our current BAR is $[X]. Based on demand signals and competitive positioning, recommend whether we should raise, lower, or hold rates for [specific dates]. Do not make up competitor rates—only analyze what I've provided. Explain your reasoning."
Forecasting: "Given this historical data [paste or upload], our current pace is [X rooms] for [date range]. Major events in market: [list]. Build a 14-day occupancy and ADR forecast with confidence ranges. If you need more data to be accurate, tell me what's missing."
Displacement Analysis: "A group wants 60 rooms at $159/night for [dates]. Our forecasted transient demand is [X] rooms at $[Y] ADR. Calculate the revenue displacement and recommend accept/reject with terms. Show your math step-by-step."
Operations & Labor
Labor Scheduling: "Create a housekeeping schedule for next week. We have [X] rooms, forecasted occupancy is [paste daily numbers], productivity standard is [X] rooms/housekeeper, and we have [X] full-time and [X] part-time staff available. Minimize overtime while ensuring coverage. What did I miss in this analysis?"
SOP Development: "Write a step-by-step SOP for [specific task, e.g., handling a guest room move request]. Include who is responsible, timing expectations, system entries required, and escalation paths. Format for a front desk training manual."
Cost Analysis: "Our laundry costs are $[X]/occupied room vs. brand standard of $[Y]. Analyze potential causes and recommend 3 specific cost reduction strategies with projected savings. Do not invent industry benchmarks—use only what I've provided."
Guest Experience & Recovery
Review Analysis: "Analyze these [X] guest reviews. Identify recurring themes, sentiment trends, and specific operational failures. Present as a table with frequency counts and recommended fixes prioritized by impact."
Service Recovery: "A Marriott Bonvoy Titanium member had these issues during their stay: [list]. Draft a recovery letter that acknowledges specifics, explains what we're doing to fix it, and offers appropriate compensation. Match our brand voice."
Review Response: "Write a response to this 2-star TripAdvisor review that addresses the guest's concerns professionally, shows we take feedback seriously, and invites them to return. Keep under 150 words."
Team Communication
Performance Feedback: "Help me write a performance conversation outline for a [position] who excels at [strengths] but needs to improve [areas]. I want to be direct but supportive. Include specific examples I should cite."
Meeting Agenda: "Create a focused 30-minute department head meeting agenda. Topics: [list]. Include time allocations, who presents each item, and required pre-read materials."
Policy Communication: "Draft an all-staff memo announcing [policy change]. Explain the why behind it, what changes for them specifically, timeline, and who to contact with questions. Keep it under 300 words."
Financial & Ownership Communication
Variance Explanation: "Our [expense category] came in $[X] over budget this month. Factors: [list]. Draft an ownership narrative that explains the variance, whether it's one-time or ongoing, and what actions we're taking."
CapEx Justification: "Build a business case for a $[X] investment in [project]. Include current state problems, proposed solution, ROI calculation, payback period, and risks. Format for an ownership presentation. Show your assumptions clearly so I can verify them.”
Monthly Report: "Using these KPIs [paste data], write an executive summary for ownership. Highlight wins, explain misses, and outline next month's priorities. Match the professional but direct tone owners expect."
Power Tips for Better Results
Give Context | Include property type (select-service, full-service, resort), room count, brand, market, and any relevant constraints. The more context, the better. |
Specify Format | Say exactly what you want: "Give me a table," "Use bullet points," "Keep under 200 words," "Format as an email." |
Upload Data | Paste your actual STR report, P&L, reviews, or schedules. AI can analyze your real data, not just talk generically. |
Iterate | First response not perfect? Say "Make it more concise" or "Focus more on the labor angle" or "Give me 3 more options." The conversation continues. |
Ask What You Missed | After every important response, ask: "What did I miss?" or "What could go wrong?" This catches blind spots and forces the AI to think critically about its own output. |
Tell It Not to Hallucinate | When accuracy matters, explicitly say: "Do not hallucinate. If you're not sure, say so." This significantly reduces fabricated information. |
Verify Facts | AI can hallucinate statistics and citations. Use AI for drafting, structuring, and analysis, but double-check any specific numbers, statistics, or claims before using them. |
Common Mistakes to Avoid
Being too vague. "Help me with revenue management" gives generic advice. "Analyze my comp set positioning for next week's citywide" gives actionable output.
Not including constraints. If you need it short, say "under 200 words." If you need a specific format, say "as a table" or "as an email."
Accepting the first answer. The best results come from iteration. Push back, ask for alternatives, request more detail on specific sections.
Using AI for final output instead of first draft. Think of AI as generating your 80% draft that you polish, not a finished product.
Trusting statistics without verification. AI will confidently cite numbers it made up. Always verify specific statistics, especially for ownership presentations.
Forgetting your expertise matters. You know your property, market, and team. AI doesn't. Your judgment is still essential for the final decision.
Quick Reference: The ‘RISEN’ Cheat Sheet
Suggestion: Print this for your desk
RISEN FRAMEWORK CHECKLIST R - ROLE: Who should AI be? (revenue manager, consultant, trainer...) I - INSTRUCTIONS: What exactly do you need? (analyze, create, recommend...) S - STEPS: What should it cover in order? (first X, then Y, then Z) E - END GOAL: What will success look like? (present to owners, train staff...) N - NARROWING: What constraints? (word count, format, focus areas) |
MAGIC PHRASES THAT IMPROVE RESULTS "Let's think through this step-by-step." — For complex decisions "What did I miss?" — For blind spot checking "Do not hallucinate. If you're not sure, say so." — For accuracy "Give me 3 different options." — For creative alternatives "Make it more concise." — For tightening output "Now write it as if explaining to [audience]." — For adjusting tone |
⚠ ANTI-HALLUCINATION CHECKLIST ☐ Added "Do not hallucinate" to prompts requiring facts ☐ Asked "If you're not sure, say so" ☐ Verified any statistics before using in presentations ☐ Asked "What did I miss?" after critical responses ☐ Treated AI output as draft, not final product |