Transforming Procurement with AI-Powered Spend Classification at Oraczen

Introduction:

In the rapidly evolving procurement industry, organizations face the challenge of managing vast amounts of expense data while ensuring accurate classification for strategic sourcing and cost optimization. Oraczen, a leader in AI-driven procurement solutions, has developed an advanced machine learning (ML) model to address the classification problem in enterprise spend management. This case study explores how Oraczen’s AI-powered spend classification model, leveraging a taxonomy sheet, customer expense data, enterprise rules, and human-in-the-loop (HITL) capabilities, has transformed procurement processes for a top Fortune Global 500 manufacturing client, achieving significant time savings and high accuracy.

The Challenge

The Procurement Classification Problem

The client, a top Fortune Global 500 manufacturer with over $15 billion in annual revenue, struggled with manual spend classification processes. The procurement classification problem presented a significant hurdle, characterized by the need to categorize millions of transactions annually from diverse suppliers across multiple categories into a complex taxonomy. Key challenges included:

Inconsistent Naming and Descriptions:

Suppliers and internal systems often used different names, spellings, or formats for the same items (e.g., “laptop,” “notebook computer,” “LT-HP-Elite”). Manual entry errors or abbreviations further created ambiguity, making automated classification harder.

Free-Text Purchase Descriptions:

Transaction descriptions were often written in unstructured free text with varying formats, missing product codes, or no clear categorization, complicating the classification process.

Multiple Taxonomies:

The company used different classification standards (e.g., UNSPSC, ISIC, and internal schemas), which didn’t always align. Merging datasets with different taxonomies or levels of granularity introduced significant complexity.

Decentralized Procurement:

As a global organization, different business units and regions used disparate systems and standards, leading to inconsistent classifications across the enterprise.

Supplier and Product Diversity:

With thousands of suppliers and millions of products spanning categories like raw materials, services, and maintenance, repair, and operations (MRO), high-resolution classification was difficult to scale manually.

Data Volume and Velocity:

The large volume of purchase transactions flowing continuously required automation to maintain up-to-date and accurate classifications, a task that was challenging without robust models and labeled data.

Complex Taxonomy Management:

The company used a detailed procurement taxonomy with over 200 categories, making manual classification labor-intensive and prone to errors.

Enterprise-Specific Rules:

Certain transactions, such as those with sister companies or specific suppliers (e.g., logistics providers), required custom classification rules, increasing the risk of inconsistency.

Inconsistent Classifications:

Manual processes led to an estimated 20% misclassification rate, impacting sourcing decisions, cost-saving opportunities, and compliance with internal policies.

Impacts of Misclassification:

Errors in classification affected spend visibility, category management, supplier consolidation, cost-saving analysis, and compliance monitoring, leading to inefficiencies and missed opportunities.

Time-Intensive Processes:

The procurement team spent approximately 12,000 hours annually on spend classification, diverting resources from strategic tasks.

Scalability Issues:

The growing volume of transactions, especially during peak seasons, exacerbated the classification problem, making it difficult to maintain accuracy and efficiency.

The client sought an automated solution to overcome these challenges, improve classification accuracy, reduce human effort, and align with their digital transformation goals.

The Solution

Oraczen’s AI-Powered Spend Classification Model

Oraczen developed a sophisticated AI-driven spend classification model tailored to the client’s needs. The model integrates machine learning, a Retrieval-Augmented Generation (RAG) system, large language models (LLMs), and human-in-the-loop oversight to deliver accurate and scalable spend categorization. The key steps in the classification algorithm are as follows:

1. Input Data Integration:

Taxonomy Sheet:

Oraczen ingested the client’s procurement taxonomy, comprising 200+ categories (e.g., raw materials, logistics, IT services).

Customer Expense Data:

Transaction data comprising supplier names, product descriptions, and transaction amounts, was collected from the client’s ERP system.

2. Enterprise Rules Application:

Custom rules were encoded into the model to reflect the client’s policies. For example:

Transactions with sister companies were excluded from procurable expenses.
All transactions from specific logistics suppliers were classified under “Logistics.”
Certain transactions were flagged as non-source able based on predefined criteria.

3. Model Training:

The model was trained on available yearly and quarterly expense data to capture seasonal and temporal patterns.
Supervised ML algorithms, including Random Forest and Gradient Boosting, were used to classify transactions based on historical patterns.

4. Real-Time Classification:

For new transactions, the model first classified expenses into the appropriate taxonomy category using trained ML models.
The model then queried the RAG system to retrieve relevant classification information based on historical data and enterprise rules.
If the information was not available in the RAG system, the model utilized LLM to analyze supplier names and product descriptions, leveraging natural language processing (NLP) to infer the most suitable category for novel or ambiguous transactions.

5. Human-in-the-Loop (HITL) Oversight:

A HITL interface allowed procurement professionals to review and correct misclassified transactions.
o Each classification was accompanied by an accuracy score (0 to 1), indicating the model’s confidence level. Low-confidence classifications (e.g., <0.7) were flagged for human review.

6. Continuous Improvement:

Feedback from human interventions was fed back into the model to retrain and improve accuracy over time.
The model achieved an estimated accuracy rate of 95% after six months of deployment.

Implementation

Oraczen collaborated with the Fortune Global 500 client to integrate the AI model into their existing procurement platform. The implementation process included:

Data Integration:

Extracting and cleaning three years of transaction data from the client’s ERP system, addressing inconsistencies in naming and free-text descriptions.

Rule Configuration

Working with the procurement team to define and encode enterprise-specific classification rules, ensuring alignment across decentralized units.

Training and Testing:

Training the model on historical data and validating it on a holdout dataset, achieving an initial accuracy of 92%.

User Training:

Conducting workshops to train the procurement team on the HITL interface and interpreting accuracy scores.

Deployment:

Rolling out the model in phases, starting with a pilot for the logistics and IT spend categories before scaling to all categories, ensuring compatibility with multiple taxonomies.

Results and Success Metrics

The deployment of Oraczen’s AI-powered spend classification model yielded transformative results for the client’s procurement operations:

Accuracy Improvement:

The model achieved a 95% classification accuracy rate, reducing misclassifications from 20% to less than 5%. This improved the reliability of spend analytics and sourcing strategies.

Time Savings:

Manual classification time was reduced by 85%, from 12,000 hours to 1,800 hours annually. This freed up procurement professionals to focus on strategic tasks like supplier negotiations and sustainability initiatives.

Cost Savings:

Accurate classifications enabled the identification of $30 million in annual cost-saving opportunities through optimized supplier selection and contract negotiations.

Example:

Improved Classification Accuracy Through HITL Feedback

Following the initial deployment of Oraczen’s AI-powered spend classification model, iterative feedback loops were implemented using human-in-the-loop (HITL) corrections and model retraining. As a result, discrepancies between the AI-classified and original category spend percentages were significantly reduced. For example, as shown in the latest evaluation, the AI’s predicted spend shares now differ by less than 1% from the original across all Level 1 taxonomy categories. This demonstrates the model’s ability to self-correct and improve through continued exposure to enterprise-specific data and expert validation—resulting in highly reliable and consistent spend classification over time.

Client Feedback

The success of Oraczen’s solution is reflected in the feedback from key stakeholders at the Fortune Global 500 client:

Chief Procurement Officer:

“Oraczen’s AI-powered spend classification model has been a game-changer for our procurement operations. The 85% reduction in manual classification time has allowed our team to focus on strategic initiatives, while the 95% accuracy ensures we’re making data-driven decisions with confidence.”

Procurement Analytics Lead:

“The human-in-the-loop interface is intuitive and empowers our team to refine classifications seamlessly. The model’s ability to

handle our complex taxonomy and enterprise rules has significantly improved our spend visibility.”

IT Director:

“Integrating Oraczen’s solution into our existing systems was smooth, and the scalability of the model has supported our growing transaction volumes without compromising performance.”

Market Figures and Trends

The procurement industry is undergoing a significant transformation driven by AI adoption. According to a 2024 study by BCG, 60% of enterprise companies plan to invest over $25 million in AI-related projects, with procurement being a key focus area. Key market trends include:

AI Adoption Growth:

Up to 94% of procurement teams now leverage AI tools, particularly for spend analytics and supplier management.

Generative AI (GenAI):

GenAI is increasingly used for tasks like contract analysis and supplier communication summarization, enhancing efficiency.

Sustainability Focus:

AI is being applied to evaluate supplier ESG performance, aligning with corporate sustainability goals

Human-AI Collaboration:

HITL models are gaining traction to balance automation with human expertise, ensuring fairness and compliance.

Oraczen’s solution aligns with these trends by combining ML, RAG, LLMs, and HITL capabilities to deliver a robust and future-ready procurement platform.

Impact on Procurement Business

Oraczen’s AI-powered spend classification model has redefined procurement operations for the Fortune Global 500 client and set a benchmark for the industry:

Operational Efficiency:

Automation of repetitive classification tasks reduced processing times by 70%, enabling faster decision-making.

Data-Driven Decisions:

Accurate spend categorization provided actionable insights into spending patterns, supplier performance, and market trends, empowering strategic sourcing.

Risk Mitigation:

The model’s ability to flag low-confidence classifications and incorporate enterprise rules minimized errors and ensured compliance with internal policies.

Strategic Focus:

By reducing manual workload, procurement teams could prioritize high-value activities like supplier relationship management and innovation.

Competitive Advantage:

The client gained a competitive edge by leveraging AI to optimize costs and respond proactively to market changes, aligning with Industry 4.0 principles.

Conclusion

Oraczen’s AI-powered spend classification model has transformed the procurement operations of a top Fortune Global 500 manufacturing client, delivering 95% accuracy, 85% time savings, and $30 million in cost-saving opportunities. By addressing the multifaceted procurement classification problem—including inconsistent naming, free-text descriptions, multiple taxonomies, and decentralized procurement—Oraczen ensured reliable and scalable spend categorization. The model’s use of a RAG system to first retrieve relevant information, followed by LLM classification for novel cases, optimized the process for efficiency and accuracy. The flowchart illustrates the end-to-end process, highlighting the model’s ability to self-improve, as evidenced by reduced discrepancies in spend percentages across taxonomy categories. The enthusiastic testimonials from the client’s leadership further underscore the solution’s impact on efficiency, decision-making, and strategic focus. As AI adoption continues to accelerate, Oraczen remains at the forefront, empowering procurement teams to navigate the challenges of a dynamic global market.