Session

How Hilton Grand Vacations Optimizes Inventory with AI on Databricks

Overview

ExperienceIn Person
TrackArtificial Intelligence & Agents
IndustryEnterprise Technology, Travel & Hospitality
TechnologiesAI/BI, Unity Catalog, Databricks Agents
Skill LevelIntermediate
DOWNLOAD SESSION SLIDES

Timeshare and hospitality inventory decisions are complex, spanning multiple revenue streams and liabilities, constrained by thousands of contractual and operational rules. This talk explores how HGV’s Inventory AI transformed yield and revenue optimization into an AI-powered, simulation-driven workflow that enforces business guardrails, scores alternative moves, and surfaces real-time revenue impact. Powered by Databricks, the solution builds on our modernization program that migrated thousands of data elements and analytics workloads to Databricks, enabling scalable data engineering and ML operations. We’ll showcase how these components integrate into an AI-first decisioning platform - a conversational interface bridging business users and data science, and the evolution from decision support to semi-autonomous agentic AI for “What If...” simulations that streamline inventory processes. Finally, we’ll share organizational success predictors for enterprise AI adoption.

Session Speakers

Evgeny Bob

/Director, AI & Architecture
Hilton Grand Vacations

Full Summary

How Hilton Grand Vacations Rebuilt Inventory Operations With Agentic AI

Hilton Grand Vacations runs a complex timeshare operation with hundreds of resorts, overlapping ownership rights, and thousands of business rules. The team's multi‑year rebuild shows how agentic systems, not more dashboards, can turn messy inventory data into confident daily decisions. The story tracks a shift from models and BI to decision workflows, culminating in an agent‑driven "mission control" that operators now depend on.

FAQ


The team needed full visibility into questions and intent. Logging every query revealed high‑value operational gaps, including the inventory defragmentation issue that funded the MVP. Generic assistants would not provide the same depth of intent data.

Agents operate within bounded action spaces, use demand models as an oracle to validate moves, and write to scratch pads for context. A human approval queue gates consequential actions, so operators confirm or reject changes before they take effect.

No. The team prunes the search space heavily rather than brute‑forcing permutations. Demand model predictions filter candidate actions so thousands of simulations can complete overnight on practical infrastructure.

A year of data consolidation and cleansing into the Databricks Platform came first, followed by operational dashboards. Only then did the large‑scale simulations and agents deliver consistent value in daily workflows.

Adoption fluctuates and can be superficial. Decision velocity captures the operational change that matters: operators who used to reconcile many reports over days now resolve the same questions in minutes.