Problem
Hilton Grand Vacations (HGV) is a timeshare company with 720k+ members and 200+ managed resorts. Its inventory is yielded across revenue, occupancy, liability, owner experience and guest experience at once. Operators compared reports with the same time last year and with booking pace, and passed spreadsheets around. Six ML models existed but were not heavily used. Internal customers said they needed decisions, not more data points.
Example from the article: an operator must place 90,000 room-nights at one resort. No single what-if answers that, because thousands of moves interact.
Solution in brief
HGV shipped a chatbot mainly to log what operators ask, then built the recurring questions into scheduled jobs and defragmentation alerts. It then used the demand forecasts in a simulation engine that runs many guardrailed what-if moves overnight. Operators see a top-10 list and approve, reject or snooze each item.
Questions this case answers
- How to turn demand forecasts into a ranked set of decisions that operators can approve when the space of joint moves is too large to enumerate
- How to use business rules to prune a huge what-if simulation search space so that it finishes overnight
- How to turn demand forecasts into inventory decisions that operators actually use
- How to simulate thousands of what-if inventory moves overnight under thousands of business rules
Tools and methods
- Question-logging chatbot — harvests operator intent and the real decisions people make
- Create-your-own dashboards — an LLM writes code just in time to answer a data question
- What-if simulation — projected occupancy, revenue, budget, burn and liability risk for one move
- Mass overnight Monte Carlo simulation — the same engine run for every resort, yield group and date window
- Agent fleet — Manager, Convo, Resort, Yield and Revenue agents with typed tools and a critic agent
- Databricks platform — Delta Lake, Unity Catalog, MLflow, Lakehouse Monitoring, Model Serving
Google or a one-time email code. No password.
