GIS International: MRO Product Recognition & Enrichment Engine for Faster Sourcing

The operational sourcing process was extremely time-consuming due to the hurdles they faced, often reducing the ability of strategic sourcing teams to negotiate and optimise vendor contracts effectively:

• Inconsistent data formats: Customers submit Excel files in varying, often unstructured formats, requiring manual restructuring into GIS’s predefined template.
• Time-intensive categorisation: Manual classification into GIS’s internal taxonomy for vendor matching slows down processing significantly.
• Information extraction issues: Employees spent considerable time extracting and cross-referencing important pieces of information from lengthy item descriptions, such as the brand, OEM number, or country. Moreover, sometimes crucial information is missing, requiring employees to conduct internet searches to complete the missing data.
• Vendor selection complexity: Matching items to appropriate vendors involved extensive manual work, as there was no readily available structured way of matching items with the right supplier.

By integrating operational and strategic sourcing through systematically curated, structured data, GIS creates a self-reinforcing flywheel that enhances both operational responsiveness and strategic decision-making.

To address these challenges, GIS partnered with Faktion to develop a robust, AI-powered system that combines classical machine learning (supervised ML) techniques with generative AI (LLM based). This hybrid solution automates the most labour-intensive parts of the sourcing process. The system consists of the following components:

1. Automating Data Extraction

Before classification begins, automated data extraction plays a crucial role in streamlining the process.

• If customer-provided data already contains key metadata—such as brand, OEM number, classes, subclasses, or vendor item numbers—the LLM system extracts it directly.
• This means that no model prediction is needed if the data is already available, reducing processing time and increasing accuracy.
• However, since customer data quality varies, not all required metadata is always present. When data is incomplete, downstream models (AutoCat ML and AutoCat LLM) step in to predict and enrich missing information.

2. Automated Classification System Using Generative AI and Classical Machine Learning

The automated classification system is built as a flywheel, there is a flywheel between operational sourcing, Strategic sourcing and curated central database. Under the hood, there is also a technological flywheel between classical machine learning “GIS AutoCat ML” and Generative AI “GIS AutoCat LLM”:

• GIS AutoCat ML is a machine learning based AI system trained on labeled data to automatically classify items into predefined classes, subclasses, brands, and other relevant categories. Its primary strength lies in providing accurate, reliable, and quantifiable predictions with measurable confidence levels. However, its effectiveness heavily depends on the availability of sufficient labeled training samples per category, meaning performance can degrade when data is sparse or incomplete.
• For items where AutoCat ML's confidence is low, GIS AutoCat LLM assists by providing an alternative suggestion, leveraging agentic AI reasoning and external internet search and cross referencing tools. We also ask AutoCat LLM to reason and explain about its label suggestion. However, unlike AutoCat ML, its predictions are suggestions rather than guaranteed classifications backed by model confidence. In ensemble setup with Auto Cat ML, the pair forms a very strong combination.

The integration of AutoCat ML and AutoCat LLM creates a continuous, self-improving classification flywheel:

• When AutoCat ML encounters classes, brands, labels, or other logistical metadata that fall outside its trained dataset, AutoCat LLM steps in to provide an initial classification suggestion.
• A data drift detection system ensures efficient routing: known items are processed by AutoCat ML for high-confidence classification, while unseen or novel items are directed to AutoCat LLM for interpretation.
• Human experts validate LLM-generated labels, ensuring accuracy and refining ambiguous classifications.
• These validated labels are then fed back into AutoCat ML as new training data, continuously strengthening its predictive accuracy and reducing dependency on LLM-based classifications.

Over time, this iterative cycle increases the automation rate, enhancing classification precision and minimizing the need for human intervention—creating a continuously evolving, AI-driven procurement intelligence system.

GIS International’s AI-driven sourcing flywheel is built around an innovative model, seamlessly connecting operational sourcing with strategic sourcing through a central knowledge base of curated, validated data.

Operational sourcing, enhanced by AI, rapidly addresses customer requests for new or existing C-class items—shortening time-to-market and maximizing efficiency. Each transaction enriches the central knowledge base with validated, structured data on product specifications, vendor capabilities, pricing, and service levels.

This evolving dataset becomes a powerful asset for strategic sourcing teams, who leverage these insights to:

• Consolidate procurement volumes across multiple customers and locations, improving negotiation positions.
• Negotiate smarter vendor contracts, achieve cost reductions, and enhance supply chain reliability.
• Re-engage prospects and previous customers with improved propositions, driven by better negotiated terms, pricing, and conditions.

With each turn of the flywheel, the quality and depth of GIS’s data increases, fueling a virtuous cycle: smarter operational sourcing accelerates customer fulfillment, enriched data enhances strategic negotiations, and optimized contracts continuously refine the overall value proposition.

In short, GIS’s AI-powered sourcing flywheel turns procurement complexity into strategic clarity—boosting efficiency, reducing costs, and driving continuous growth.