Smarter, Faster, and Scalable: How Keyrus Leveraged GenAI and UNSPSC for Product Classification

Background

Our client is a cloud-based eProcurement software company, specializing in serving small to mid-sized biotech organizations. Founded by industry veterans, they offer an intuitive procure-to-pay platform tailored to accelerate R&D in the life sciences sector. Following a recent growth equity investment, the company is scaling its operations and expanding its reach across the life sciences industry.

Challenge

The client faced the challenge of classifying a growing and complex product catalog using the UNSPSC taxonomy, a key enabler for procurement, pricing strategies, and data-driven decision-making. Manual classification by suppliers was time-consuming, inconsistent, and covered only a small fraction of the data (3% of the Spends), leading to poor visibility and misaligned reporting. With the rollout of a new Snowflake Data Warehouse, the client needed an intelligent, scalable solution to automate classification, improve accuracy, and ensure consistency across datasets.

Approach

To address the client's need for scalable, accurate, and consistent product classification, we followed a phased and collaborative approach: -Engaged early with stakeholders to align on business goals, taxonomy priorities, and success criteria, focusing on high-impact segments tied to confirmed PO activity. -Built a central Product Dimension as part of the client's new Snowflake Data Warehouse using AWS DMS for ingestion and dbt for data modeling, ensuring a standardized foundation for classification. -Deployed a GenAI classification engine using AWS Lambda to orchestrate calls to OpenAI/ChatGPT, enabling automated UNSPSC predictions across prioritized products. -Integrated a human-in-the-loop validation process, where subject matter experts could review AI-predicted codes. Confirmed classifications were stored in a Snowflake Vector DB for future AI retrieval and model improvement. -Established a confidence-based review funnel, only surfacing GenAI-predicted classifications above a defined threshold for downstream analytics and user interfaces. -Delivered key reporting views using AWS QuickSight to track taxonomy coverage, supplier vs. GenAI comparison, and emerging high-spend categories—empowering business teams to revisit procurement strategies and category definitions. -Collaborated cross-functionally with the procurement and analytics teams to ensure the classification outputs aligned with end-user workflows, procurement software features, and decision-making processes.

Key results

Benefits

The GenAI classification solution delivered significant value to the client by transforming a previously manual, error-prone process into an automated, scalable, and intelligence-driven workflow. By integrating classification directly into the Snowflake DWH and leveraging OpenAI’s LLM capabilities, the client achieved over 20x improvement in coverage compared to manual efforts. The result was a consistent, high-confidence product taxonomy that improved procurement visibility, enhanced user experience in search and categorization, and empowered better strategic sourcing decisions. With validated results stored in a vectorized feedback loop, the system continues to learn and improve, ensuring long-term sustainability and alignment with business growth.

Share this key play