Find out how Tribe built a vendor classification engine for a global consulting firm that reduced process time from 9-12 days to just 45 minutes.
The Challenge
A global consulting firm renowned for its leadership in corporate turnarounds and complex bankruptcy restructuring needed a better way to analyze large, intricate vendor ecosystems as part of its diligence processes. This task is foundational to first-day motions, 13-week cashflow forecasts, and the broader case management process—but it was historically slow, manual, and error-prone.
The firm’s due diligence team was spending weeks manually classifying thousands of vendors across a complex taxonomy. Each vendor needed to be mapped to the correct category in a multi-level hierarchy based on limited, inconsistent information. This process constrained the team’s ability to scale, introduced human error, and delayed insight delivery. The firm turned to Tribe AI to help automate and accelerate vendor classification without sacrificing accuracy or oversight.
The Solution
Tribe AI collaborated with the firm to design and implement an AI vendor classification engine within its proprietary due diligence platform. This system leverages OpenAI’s GPT-4o and GPT-3.5-turbo models, along with structured output generation and historical classification patterns, to automate and standardize vendor mapping workflows—even when the available data was sparse or messy. By automating this foundational task, the platform enabled faster, broader, and more reliable vendor analysis.
Key Features
The vendor classification engine was built with features tailored to handle both the scale and complexity of the firm’s vendor data:
- A multi-level taxonomy engine capable of mapping vendors across hierarchical industry and category structures.
- Natural language interpretation of messy or incomplete vendor descriptions, enabling classification beyond structured fields.
- Continuous reclassification that automatically updates as new vendor information is ingested or corrected.
- Confidence scoring and auditability, providing transparency into classification certainty and allowing human reviewers to easily trace decisions.
These features collectively empowered the platform to handle high-volume vendor data with both speed and oversight.
How It Works
The vendor classification engine works by orchestrating multiple AI techniques in a seamless pipeline. At a high level, the system ingests raw vendor data and processes it through successive layers to produce a confident classification:
- LLM inference interprets free-text vendor descriptions and generates probabilistic category predictions.
- Deterministic business rules apply hard-coded mappings where known (e.g., specific keywords or industry codes).
- Confidence scoring and thresholding determine whether to auto-classify or flag for human review.
- Feedback loops capture any human corrections, allowing the model to learn and improve over time. This hybrid approach balances the flexibility of AI with the reliability of rule-based logic, ensuring consistent outputs across diverse data inputs.
This hybrid approach balances the flexibility of AI with the reliability of rule-based logic, ensuring consistent outputs across diverse data inputs.
Tech Stack
- Cloud: Azure
- LLM: Azure OpenAI - GPT 4o (with structured outputs)
- Other cloud services: web search to get more vendor data; Azure app services for hosting; MSSQL Server for storing results
- Languages used: FastAPI in Python for backend; Streamlit for frontend
Impact
The AI-powered classification system transformed the firm’s vendor analysis process, eliminating a longstanding manual bottleneck and unlocking new efficiencies across due diligence workflows. Consultants gained back valuable time for interpretation and client advisory while benefiting from faster, broader, and more accurate vendor insights.
Key results included:
- Successfully categorized 18,000+ vendors in a live engagement with 97% accuracy (a higher accuracy level than the past human baselines), validated by subject matter experts.
- Reduced process time from 9-12 days to 45 minutes.
- Used across both the Private Equity (PE) and Turnaround & Restructuring (TRS) service lines; actively deployed on hundreds of thousands of vendors and set as the standard going forward.
The Future
This project has laid the foundation for future AI innovation at the firm. With an automated vendor classification engine in place, the firm is exploring additional applications of AI across procurement, vendor risk management, and supply chain mapping. Future iterations aim to integrate dynamic data feeds, continuously update vendor intelligence, and expand classification models into new industries and regions.
