AI Unlocks Revenue-Driving Product Catalog Intelligence

Why This Matters

In only a few weeks, Aspenware had a working LLM-based proof-of-concept that fully supports real-time pricing logic, personalization, and performance analytics. All without custom logic and vendor lock-in.

The Opportunity to Unlock Value

Aspenware is a tech platform serving over fifty ski resorts. Across the resort ecommerce ecosystem, product information — especially for dynamically-priced lift tickets — is highly inconsistent. Each storefront encodes ticket metadata differently, which can inhibit pricing engine capabilities, real-time analytics, and downstream automation. But manual cleanup of different ticket data and differing taxonomies is labor-intensive and error-prone. 

Seeing an opportunity to unlock new business value with AI, Aspenware engineering and product leadership partnered with 8th Light to build a flexible and accurate data engine. By deploying a custom large language model ("LLM") pipeline, Aspenware is beginning to transform unstructured data into standardized, human-readable product intelligence across all partner resorts. 

Looking ahead, a Retrieval-Augmented Generation ("RAG") model could be used to strengthen the pipeline by grounding outputs in resort-specific data and knowledge for greater accuracy.

Our Shared Objective

Build an LLM-powered pipeline to turn variable product metadata into a unified, enriched format, making it easier to understand and laying the groundwork for new pricing optimizations. Reduce admin overhead in managing complex custom product catalogs that differ from resort to resort.

Solution Approach and Early Results

Using AI to Make Data More Legible

The pipeline is designed to process data within dozens of customized storefronts, focusing initially on lift tickets. The goal? Normalize metadata to a shared schema, extract number of days logic, and enable the calculation of meaningful attributes like per-day price without manual intervention.

The Chosen LLM Solution

To compliment the required "primary LLM" that is tasked with driving the data pipeline, a secondary "judge" LLM was used to validate data structures. This approach increased output reliability, and reduced the potential for down-stream errors. The solution was made up of three components:

• Metadata transformation through the primary LLM using GPT-4o
• Structural validation with a secondary judge LLM using GPT-o3-mini
• A generalized architecture that accommodates varied product formats and types

Engineering Best Practices

Continuing on the theme of quality control, 8th Light followed best practices that have been proven in solution development over the past 20 years. Specifically, data cleansing and sanity checks designed to improve the stability of the system as volume and complexity increases. 

• Pre-processing, where the input data is sanitized and normalized using conventional (non-AI) logic before reaching the LLM. This minimizes token usage and ensures that the AI models will focus only on the relevant parts.
• Fallback logic that verifies that the AI response is valid JSON and matches the expected schema. This includes structured parsing, try/catch fallbacks, and format correction routines.

Aspenware Benefits Realized

While it is still early days in full value realization, early results are showing real promise for the organization. In particular:

• Faster admin insights on what's being sold
• Standardized pricing breakdowns for frontends without custom logic
• Better customer-facing experiences for return visitors

From Kickoff to Concept, Together

8th Light delivered a full end-to-end experience, from early requirements gathering to a working proof of concept. The team worked across evolving priorities and branches, integrating contributions from multiple developers while keeping the codebase cohesive through consistent refactoring.

Throughout the engagement, the team used AI-assisted workflows to support ideation, prototyping, and validation. These tools helped accelerate iteration, improve alignment on business outcomes, and shape a solution that reflected Aspenware’s specific needs. The result was a tailored experience that brought new velocity to product exploration.

Delivery included enriched output samples, a functional frontend demo, and inline documentation to support smooth handoff across both engineering and product teams.

Expected results

This project enables the calculation at scale of meaningful optimizations which are known to have positive commercial potential when deployed as merchandising tactics. Per-day pricing is one example of such attributes. Prior A/B tests have revealed that clearly surfacing per-day pricing increases purchasing behavior — especially among return users to the ecommerce experience. These return users are 25% of traffic but 59% of revenue for Aspenware. 

Estimates predict that offering per-day pricing consistently across resorts may yield the following results:

• System-wide: estimated 3% uplift in ticket product revenue
• Return users: estimated 3–5% uplift in return users, with higher upside possible up to +10%

Reaching Peak Potential

This AI-powered proof-of-concept is a foundational step in transforming Aspenware’s ecommerce operations across all resort partners. By normalizing product data at ingestion, Aspenware stands to unlock high-value pricing features, reduce ops overhead, and set the stage for next-gen personalization and automation. Future enhancements may include live integration, tools to refine and test enrichment logic, and observability tools to ensure confidence at scale.

Looking to reach your own peak potential?  Schedule a call with us to discuss how 8th Light can help your organization bring life to the best ideas.