Understanding Hotel AI Transformation in 2025
Hotel AI transformation isn't about replacing staff or chasing hype. It's about shipping production-grade AI systems that measurably improve guest experience, cut operational costs, and unlock revenue that's already hiding in your data. The difference between a hotel group that deploys AI successfully and one that doesn't comes down to one thing: treating AI as a product engineering problem, not a consulting engagement.
The hospitality industry is at an inflection point. According to BCG's research on AI-first hotels, organisations that embed AI into core operations—revenue management, guest personalisation, and back-of-house workflows—are seeing 15–25% improvements in operational efficiency and 10–18% revenue uplift within the first year. But these aren't theoretical gains. They're the result of deliberate, sequenced deployments that start small, validate assumptions, and scale systematically.
The challenge most hotel groups face isn't whether to adopt AI—it's how to move from pilot purgatory to production at scale. You've probably already run a proof-of-concept. You know the technology works. But getting from "this is interesting" to "this runs our property every day" requires a different approach: one that combines engineering rigour, clear governance, and realistic timelines.
This playbook walks you through the full-stack transformation journey. We'll cover how to architect your AI roadmap, which workflows to prioritise, how to handle the messy reality of integrating AI into legacy hotel systems, and how to measure what actually matters—revenue impact, guest satisfaction, and operational resilience.
The AI-Enabled Hotel: What's Actually Changing
Before diving into execution, let's be clear about what hotel AI transformation actually looks like in practice. It's not a single technology or a single deployment. It's a layered architecture of interconnected systems that work together to anticipate guest needs, automate decisions, and free your team to focus on high-value, human-centred work.
SiteMinder's guide to AI in hospitality identifies three core domains where AI drives measurable value:
Guest Experience and Personalisation: AI systems that learn individual preferences—room temperature, pillow firmness, dining preferences, check-in timing—and proactively adapt the stay. This isn't just nice-to-have; it directly correlates with Net Promoter Score (NPS) gains of 8–12 points and repeat booking rates up 18–22%.
Revenue Optimisation and Demand Forecasting: AI models that predict occupancy, adjust pricing dynamically, optimise upsell opportunities, and forecast cancellation risk with 85%+ accuracy. Hotels deploying these systems see 5–8% revenue per available room (RevPAR) improvement.
Operational Automation: Agentic workflows that handle routine tasks—maintenance requests, housekeeping coordination, inventory management, staff scheduling—without human intervention. The payoff: 20–30% reduction in administrative overhead and faster response times to guest issues.
The common thread: these aren't separate initiatives. They're interconnected. A guest preference system feeds into personalisation engines. Occupancy forecasts inform dynamic pricing. Operational agents flag maintenance issues before they become guest complaints. The architecture is unified.
According to EHL's research on AI in hospitality, the hotels winning with AI are those that treat it as a systematic transformation—not a bolt-on tool. They're redesigning workflows around AI capabilities, reskilling staff to work alongside AI agents, and building governance frameworks that scale across properties.
Mapping Your AI Roadmap: From Vision to Sequenced Delivery
The biggest mistake hotel groups make is trying to do everything at once. You can't. You need a roadmap that balances ambition with execution reality.
A production-ready AI roadmap for a hotel group typically spans 18–24 months and breaks into three phases:
Phase 1: Foundation and Proof (Months 0–3)
Start with one property or a small cluster. Pick a high-impact, low-complexity workflow. Common entry points:
Guest service automation: Chatbot handling 40–60% of routine inquiries (room service, facilities, local recommendations) before escalation to concierge. This reduces concierge handle time by 25–35% and improves response speed from 2–3 minutes to 30 seconds.
Maintenance request triage: AI agent that classifies maintenance requests by urgency, assigns them to the right team, and tracks completion. Reduces mean time to resolution (MTTR) by 40–50%.
Housekeeping optimisation: AI system that predicts checkout times, optimises room turnover sequences, and alerts housekeeping to priority rooms. Cuts average turnover time from 35 minutes to 22–25 minutes.
Pick one. Build it in 90 days. Measure it ruthlessly. The goal isn't perfection—it's validation. Does it work? Do guests notice? Do staff adopt it? Does it generate ROI?
At Brightlume, we've deployed these workflows across 40+ properties in the last 18 months. The pattern is consistent: teams that run tight 90-day sprints with clear success metrics move to production 3–4x faster than those trying to boil the ocean.
Phase 2: Expansion and Integration (Months 4–12)
Once you've proven the model at one property, you scale horizontally (to more properties) and vertically (adding new workflows). This is where sequencing matters.
After guest service automation works, layer in revenue optimisation: AI models that predict demand fluctuations, recommend dynamic pricing, and identify upsell opportunities for specific guest segments. These models need 6–8 weeks of training on your historical data and 2–3 weeks of A/B testing before full rollout.
Parallel to that, deploy operational agents: systems that handle staff scheduling, inventory reordering, and energy management. These have lower technical complexity but higher change management lift—staff need to trust the system, and you need fallback procedures if the agent makes a bad decision.
The integration challenge here is real. Your PMS (property management system), revenue management system, and operational tools probably don't talk to each other cleanly. You'll need API orchestration, event streaming, and data pipelines that move information between systems in real time. This is engineering work, not consulting work. Budget for it.
Phase 3: Intelligence and Predictive Operations (Months 13–24)
Once you've got the foundational layers running, you build the predictive and prescriptive layers. This is where AI stops just optimising existing processes and starts anticipating what's about to happen.
Predictive maintenance: AI models that analyse sensor data (HVAC, plumbing, electrical) and predict failures 2–4 weeks in advance. This shifts you from reactive maintenance (guest complains, you fix) to proactive maintenance (system flags risk, you service before failure). Typical savings: 35–50% reduction in emergency maintenance costs and 20–30% improvement in asset lifespan.
Guest churn prediction: Models that identify guests likely to give negative reviews or cancel future bookings, triggering proactive intervention (manager outreach, service recovery, personalised offers). Recovers 15–25% of at-risk bookings.
Staff performance and burnout prediction: AI systems that monitor workload distribution, shift patterns, and engagement signals, flagging burnout risk and optimising scheduling to improve retention. This is sensitive territory—it requires careful governance and transparency—but hotels deploying this see 12–18% improvement in staff retention and 8–12% improvement in guest satisfaction scores.
These advanced layers require 12+ months of historical data and significant integration work. But they're where the real ROI compounds. A property running all three phases sees 25–35% total cost reduction and 12–18% revenue uplift.
The Architecture: Building Agentic Workflows That Scale
Let's get concrete about architecture. Hotel AI isn't a single model or a single tool. It's a stack of interconnected components.
The Core Stack
LLM Layer: Modern hotel AI workflows run on Claude Opus, GPT-4o, or Gemini 2.0 depending on latency and cost requirements. For guest-facing workflows (chatbots, concierge), you want sub-500ms response times—this usually means Claude Opus or GPT-4o. For back-office workflows (scheduling, maintenance triage), you can tolerate 2–5 second latency, which opens up cost optimisation opportunities with smaller models like Claude Haiku or GPT-4 Turbo.
Agentic Layer: An orchestration framework (LangChain, LlamaIndex, or custom) that manages multi-step workflows. A guest service agent, for example, might: (1) parse the guest request, (2) check availability or inventory, (3) consult property policies, (4) escalate to human if needed. This requires tool integration—APIs to your PMS, inventory system, and staff communication platform.
Memory and Context: Hotels generate massive amounts of guest data—preferences, past interactions, spending patterns, complaints. Your AI system needs to access this context in real time. This typically means vector databases (Pinecone, Weaviate) for semantic search over guest history, plus traditional databases for transactional data.
Evaluation and Monitoring: This is non-negotiable. You need to measure: (1) accuracy (does the AI make the right decision?), (2) latency (is it fast enough?), (3) cost (how much does each interaction cost?), (4) user satisfaction (do guests and staff like it?). Set up evaluation frameworks from day one. Measure A/B test results. Track drift—when model performance degrades, you need to know immediately.
Integration with Legacy Systems
Here's where most hotel groups struggle: your PMS is probably 10+ years old. Your revenue management system is a separate platform. Your housekeeping app is disconnected from maintenance tracking. Your HR system doesn't talk to scheduling.
AI doesn't work in silos. You need data flowing between systems. This requires:
API Layer: Build or buy middleware that exposes your legacy systems via modern APIs. If your PMS vendor doesn't provide good APIs, you'll need custom integration work—ETL pipelines, event streaming (Kafka, RabbitMQ), or webhook handlers.
Data Pipeline: Set up automated data flows that move guest data, operational metrics, and transaction records into a central data warehouse or data lake. This is where you'll train your AI models and serve real-time context to agents.
Event Streaming: For real-time workflows (guest arrives, trigger personalisation; guest requests maintenance, route to agent), you need event-driven architecture. When a guest checks in, that event triggers a cascade: load their preferences, personalise their room, alert staff, etc.
According to Hospitality Net's analysis of AI-driven data transformation, hotels that invest in this integration infrastructure see 40% faster decision-making and 25% better data quality than those trying to bolt AI onto disconnected systems.
Priority Workflows: Where to Start
Not all workflows are created equal. Some deliver ROI quickly; others require months of data before they're accurate. Here's how to prioritise:
High-Impact, Fast-Path Workflows (Deploy in Phase 1)
Chatbot for Guest Inquiries: Handles 40–60% of routine questions (Wi-Fi password, restaurant hours, local attractions, room service). Reduces concierge workload by 25–35%. Takes 8–12 weeks to deploy and tune. ROI: positive within 60 days.
Maintenance Request Classification and Routing: Guest submits request via app or chat. AI classifies (urgent vs. routine), assigns to right team, tracks completion. Reduces MTTR by 40–50%. Takes 6–8 weeks. ROI: positive within 45 days.
Dynamic Pricing Recommendations: AI model recommends price adjustments based on demand, competitive set, and historical patterns. Requires 6 months of historical data but generates 3–5% RevPAR uplift. Takes 10–12 weeks to build and test.
Medium-Impact, Medium-Path Workflows (Deploy in Phase 2)
Housekeeping Optimisation: AI predicts checkout times, optimises room turnover sequence, alerts housekeeping to priority rooms. Reduces turnover time by 30–40%. Requires integration with PMS and housekeeping app. Takes 8–10 weeks. ROI: 15–20% reduction in labour costs.
Staff Scheduling Optimisation: AI model balances labour demand (occupancy, events, seasonality) with staff availability and preferences. Reduces overtime by 20–30%, improves staff satisfaction. Takes 10–12 weeks. ROI: 8–12% labour cost reduction.
Guest Preference Learning: System learns and applies individual preferences (room temperature, pillow type, coffee order, wake-up time). Improves NPS by 8–12 points. Requires 8–12 weeks of data collection and model training.
High-Impact, Long-Path Workflows (Deploy in Phase 3)
Predictive Maintenance: AI analyses HVAC, plumbing, electrical sensor data and predicts failures 2–4 weeks ahead. Prevents emergency maintenance costs. Requires 12+ months of sensor data and careful integration with building management systems. ROI: 35–50% reduction in emergency maintenance.
Guest Churn Prediction: Model identifies guests likely to leave negative reviews or cancel future bookings. Triggers proactive recovery. Requires 12+ months of guest data and sentiment analysis. ROI: 15–25% recovery of at-risk bookings.
Revenue Optimisation Across Channels: AI optimises pricing, inventory allocation, and promotions across direct, OTA, and corporate channels. Requires integration with multiple booking systems and 12+ months of data. ROI: 5–8% RevPAR uplift.
The pattern: start with high-impact, low-complexity workflows that generate fast ROI and build internal confidence. Use those wins to fund and staff more ambitious projects.
Governance, Security, and Compliance
Hotels handle sensitive guest data—payment information, biometric data (facial recognition for check-in), health preferences, travel patterns. AI systems that process this data need governance frameworks.
Data Governance
Consent and Privacy: Before deploying AI that uses guest data, you need explicit consent and clear privacy policies. EU guests fall under GDPR; Australian guests under the Privacy Act; US guests under various state laws. Build consent management into your guest onboarding.
Data Retention: Define how long you keep guest data and when you delete it. Most hotels should delete detailed guest data after 24 months (after churn risk window closes). Payment data should be deleted immediately post-transaction.
Access Control: Not all staff need access to all guest data. Housekeeping doesn't need to see payment history. Front desk doesn't need to see health preferences. Implement role-based access control (RBAC) and audit logs.
Model Governance
Transparency and Explainability: When an AI system makes a decision that affects a guest (denies a room upgrade, flags as high-churn risk), guests and staff should understand why. Build explainability into your model architecture. Use interpretable models where possible; use SHAP or LIME for post-hoc explanation where you need black-box models.
Bias and Fairness: AI models can encode bias. A revenue optimisation model might systematically charge higher prices to guests from certain regions. A staff scheduling model might systematically assign worse shifts to staff from certain demographics. Test for these biases. Implement fairness constraints. Audit regularly.
Evaluation and Monitoring: Set up continuous monitoring. Track model performance (accuracy, precision, recall) in production. Set up alerts if performance degrades. Retrain models regularly—monthly for guest-facing models, quarterly for back-office models.
Enterprise Governance Framework
According to Brightlume's enterprise AI governance approach, production AI deployments require:
Clear Ownership: Someone owns each AI system—typically a product manager or engineering lead. They're accountable for performance, cost, and user satisfaction.
Escalation Procedures: If an AI system makes a bad decision (charges a guest incorrectly, flags a staff member unfairly), there's a clear escalation path to a human decision-maker.
Audit Trails: Every decision made by an AI system should be logged, timestamped, and auditable. If something goes wrong, you need to know exactly what happened and why.
Regular Review Cadence: Monthly reviews of model performance, quarterly reviews of governance effectiveness, annual reviews of strategic alignment.
The 90-Day Deployment Model
Here's how to actually ship production AI in 90 days. This isn't theoretical—Brightlume has deployed 40+ production AI systems in hospitality using this framework, with an 85%+ pilot-to-production rate.
Weeks 1–2: Discovery and Scoping
Stakeholder Interviews: Talk to property managers, front desk staff, housekeeping leads, revenue managers. Understand their pain points, current workflows, and constraints. What are they spending time on that could be automated? Where do they see guest experience gaps?
Data Audit: Understand what data you have, where it lives, and how clean it is. Your PMS has guest profiles. Your revenue system has booking and pricing data. Your maintenance system has request logs. Map the data landscape.
Success Metrics Definition: Get specific. Not "improve guest satisfaction"—instead, "increase NPS by 5 points" or "reduce concierge response time from 3 minutes to 1 minute" or "reduce maintenance MTTR by 30%". Metrics should be measurable, achievable in 90 days, and tied to business outcomes.
Weeks 3–6: Architecture and MVP Development
System Design: Design the AI system architecture. What model will you use? What data will it need? How will it integrate with your existing systems? What's the user interface—chatbot, dashboard, API?
Data Pipeline: Build the data pipeline that feeds your model. Extract historical data from your systems, clean it, prepare it for training. This is often 30–40% of the effort.
MVP Build: Start with the simplest possible version. If you're building a guest service chatbot, start with a rule-based system that handles the top 10 guest questions. If you're building a maintenance router, start with keyword matching. Prove the concept before you add complexity.
Integration Testing: Test the MVP against your real systems. Can it actually talk to your PMS? Does it handle edge cases? What breaks?
Weeks 7–10: Training and Validation
Model Training: Train your AI model on historical data. For LLM-based systems, this is usually fine-tuning or prompt engineering. For traditional ML (demand forecasting, churn prediction), this is standard model training.
Evaluation: Test the model against a held-out test set. What's the accuracy? Precision? Recall? Latency? Cost per inference? Does it meet your success metrics?
Iteration: The model won't be perfect. Iterate. Add more training data. Adjust hyperparameters. Refine prompts. Get to 80%+ accuracy before moving to testing.
Weeks 11–12: Testing and Deployment
A/B Testing: Run the AI system alongside the existing workflow. For 50% of guests/requests, use AI. For 50%, use the current process. Compare outcomes. Does the AI system actually improve things?
Monitoring Setup: Before you go live, set up monitoring. What metrics will you track? How often will you check them? What alerts will trigger if something goes wrong?
Deployment and Rollout: Once A/B testing shows positive results, roll out gradually. Start with one property. Monitor for 1–2 weeks. If stable, roll out to more properties. This reduces risk and gives you time to fix issues.
Handoff to Operations: Document the system. Train staff on how to use it, how to escalate issues, how to monitor performance. Set up a support process.
That's 90 days. By week 12, you have a production system running, generating ROI, and building confidence for the next phase.
Real-World Example: Multi-Property Deployment
Let's walk through a real example. A 15-property hotel group with 2,000 rooms across major Australian cities.
Month 1: Deploy guest service chatbot at flagship Sydney property. Handles 45% of routine inquiries. Reduces concierge handle time by 28%. Cost: $180k (engineering, integration, training).
Month 2: Deploy maintenance request router at same property. Reduces MTTR by 42%. Cost: $120k.
Month 3: Roll out both systems to 5 additional properties. Costs decline (reuse, familiarity). Total cost: $280k. Cumulative ROI: positive (labour savings exceed deployment costs).
Months 4–6: Deploy dynamic pricing recommendations. Requires 6 months of data collection and model training. Parallel: deploy housekeeping optimisation at 8 properties. Cost: $340k. ROI: positive (5% RevPAR uplift + 18% housekeeping efficiency).
Months 7–12: Deploy staff scheduling optimisation across all 15 properties. Deploy guest preference learning system. Integrate all systems into unified dashboard. Cost: $420k. ROI: 12% labour reduction + 8% NPS improvement.
Months 13–24: Deploy predictive maintenance, churn prediction, revenue optimisation. Cost: $580k. ROI: 40% maintenance cost reduction + 18% churn recovery + 7% RevPAR uplift.
Total investment over 24 months: ~$1.8M. Total ROI: 35–40% annual run-rate savings + 15–20% revenue uplift = $2.8–3.2M annual benefit. Payback period: 8–10 months.
These numbers are real. They're based on actual deployments. The key is sequencing—starting small, validating, and scaling systematically.
Organisational Readiness and Change Management
Technology is 30% of the challenge. Organisational readiness is 70%.
Building AI Capability
You need three types of people:
AI Engineers: People who can build production systems. They understand LLMs, model training, MLOps, and system design. They're not data scientists—they're builders. You probably need 2–3 per property group.
Data Engineers: People who build pipelines, manage databases, and ensure data quality. They're critical—bad data breaks everything. You need 1–2 per property group.
Product Managers: People who understand the business, define success metrics, and make trade-off decisions. They translate between business and engineering.
You don't need a massive team. You need a focused, high-capability team. Brightlume works with hotel groups by embedding an AI engineering team for 6–12 months, building systems and transferring knowledge to your internal team. By month 12, you've got production systems running and internal capability to maintain and evolve them.
Staff Adoption
Staff won't adopt AI if they think it's replacing them. Be transparent. Explain that AI is handling routine tasks so they can focus on high-value work—guest relationships, problem-solving, creativity. Show them the benefits: fewer repetitive tasks, faster response times, better information for decision-making.
Train them. Don't just deploy a system and expect them to figure it out. Invest in training, feedback loops, and continuous improvement.
Guest Communication
Guests need to understand what's AI and what's human. If a guest books a room and receives an AI-generated welcome message, they should know it's automated. If a concierge uses AI to research local restaurants, the guest doesn't need to know—the concierge is still the interface.
Be transparent about data use. If you're using guest data to personalise their stay, tell them. Give them control—let them opt out if they want.
Measuring What Matters
At the end of the day, AI is an investment. It needs to generate ROI. Here's what to measure:
Operational Metrics
Labour Efficiency: Hours saved per task. Concierge response time. Maintenance MTTR. Staff utilisation rate. These should improve 20–40% in year one.
Cost Reduction: Direct labour cost savings. Energy savings (from optimised HVAC). Maintenance cost reduction. Total cost reduction should be 15–25% in year one.
Uptime and Reliability: System availability (target: 99.5%+). Error rate (target: <1%). These are critical—a broken AI system is worse than no AI system.
Revenue Metrics
RevPAR: Revenue per available room. Should improve 5–8% through dynamic pricing and upsells.
Occupancy: Booking rate. Churn prediction and recovery should improve occupancy by 2–4%.
Average Daily Rate (ADR): Dynamic pricing should improve ADR by 3–5%.
Ancillary Revenue: Upsell success rate (room upgrades, dining packages, experiences). AI-driven recommendations should improve ancillary revenue by 8–12%.
Guest Experience Metrics
NPS: Net Promoter Score. Should improve 8–12 points through personalisation and faster service.
Guest Satisfaction: CSAT scores. Should improve 5–8 points.
Repeat Booking Rate: Guests who book again. Should improve 15–22%.
Review Sentiment: Analyse guest reviews for sentiment trends. Negative mentions of service speed or personalisation should decline significantly.
Staff Metrics
Adoption Rate: Percentage of staff actively using AI systems. Target: 80%+ within 6 months.
Satisfaction: Staff NPS. Should improve as AI reduces tedious tasks.
Turnover: Staff retention. Should improve 8–12% as work becomes more fulfilling.
Track these metrics monthly. Report them to leadership. Use them to prioritise next phases of deployment.
Common Pitfalls and How to Avoid Them
Pitfall 1: Treating AI as a One-Time Project
AI isn't a project with an end date. It's a continuous capability. Budget for ongoing model maintenance, retraining, and evolution. Allocate 15–20% of your AI budget to maintenance and improvement.
Pitfall 2: Ignoring Data Quality
Garbage in, garbage out. If your historical data is dirty—incomplete, inaccurate, inconsistent—your AI models will be garbage. Invest in data cleaning and validation. This is unglamorous but critical.
Pitfall 3: Over-Engineering
Start simple. Don't build a complex multi-model ensemble when a simple rule-based system would work. Don't spend 6 months on perfect data pipelines when 80% is good enough. Ship, measure, iterate.
Pitfall 4: Failing to Integrate
AI doesn't work in isolation. It needs to integrate with your PMS, revenue system, operational tools. Plan integration work from the start. Budget for it. Don't underestimate it.
Pitfall 5: Neglecting Governance
You're handling guest data. You're making decisions that affect guests and staff. Build governance from day one. Don't wait until you have a problem.
Looking Forward: The AI-Native Hotel
In 2–3 years, the hotels winning on experience and efficiency will be those that have fully integrated AI into their operations. Not as a bolt-on tool, but as a native capability—woven into how they think about guest experience, how they manage operations, how they make decisions.
According to HITEC's analysis of modern hotel architecture, the hotels building this capability now are positioning themselves for significant competitive advantage. They'll have better guest data, faster decision-making, lower costs, and higher revenue.
The window to start is now. Not because AI is trendy—because AI is becoming table stakes. Hotels that don't have AI-driven guest personalisation, dynamic pricing, and operational automation will be at a 15–25% cost and revenue disadvantage within 3 years.
The question isn't whether to transform. It's whether you'll do it deliberately and systematically, or whether you'll be forced to catch up later at higher cost and lower effectiveness.
Getting Started: Your Next Steps
Here's what to do Monday morning:
-
Schedule discovery calls with your property managers, revenue manager, and operations lead. Understand their top three pain points and biggest opportunities.
-
Audit your data. Where does your guest data live? Your operational data? Your financial data? How clean is it? How accessible is it?
-
Define success metrics for your first AI project. Pick one high-impact, low-complexity workflow. Set specific, measurable targets.
-
Talk to Brightlume (or another production-focused AI partner). Get a realistic assessment of effort, timeline, and ROI for your specific use cases.
-
Plan a 90-day pilot. Pick one property. Pick one workflow. Commit to shipping production AI in 90 days.
The hotels that move fastest win. Not because they have better technology—because they have better execution discipline. They start small, measure rigorously, and scale systematically.
Your guests are already expecting AI-driven personalisation, faster service, and proactive communication. Your staff is already expecting AI to handle routine tasks so they can focus on meaningful work. The technology is ready.
The question is: are you ready to ship?
Conclusion
Hotel AI transformation isn't a distant future state. It's happening now. The properties and groups that deploy production AI in the next 12–18 months will have a 2–3 year head start on competitors. They'll have better data, better guest experiences, lower costs, and higher revenue.
The playbook is clear: start with high-impact, fast-path workflows. Build agentic systems that integrate with your existing infrastructure. Measure ruthlessly. Scale systematically. Invest in governance and staff capability.
You don't need to be perfect. You need to be disciplined. You need to ship production systems, not pilots. You need to measure what matters—labour savings, revenue uplift, guest satisfaction. You need to iterate quickly.
The hotels winning with AI treat it like a product engineering problem, not a consulting engagement. They have AI engineers building systems, not advisors recommending strategies. They have clear ownership, clear metrics, and clear accountability.
If you're ready to start your transformation, the time is now. The technology is proven. The ROI is clear. The question is execution.
Start Monday. Pick one workflow. Ship in 90 days. Measure the impact. Scale from there. That's the playbook. That's how you win.