Content Tagging for CRM Free-Text Data

Management Summary

HMS developed an automated content tagging solution for unstructured CRM free-text data for a Company in the Pharmaceutical and Healthcare Sector. The LLM-based solution classifies the content according to predefined categories and supplements the model’s logic with rules for classes that are difficult to distinguish. The structured tags support search, grouping, and downstream analyses.

Project summary

#lifesciences

Companies in the pharmaceutical healthcare sector

Industry: Pharmaceuticals and Healthcare
Project Start: 09/2023
Focus: Generative & Agentic AI
Use Case: Document & Content Intelligence

Project Objective

Classify CRM free-text entries and make them available as structured data for analysis

Key Features

LLM-based classification according to predefined content classes
Supplementary rule-based logic for categories with similar subject matter
Production-ready integration and processing in Palantir Foundry

Tech Stack

python langchain palantir foundry gpt models

The combination of LLM-based classification and rule-based logic was crucial for reliably distinguishing between content classes that are similar in subject matter.
Christoph Bergen
CoE Lead for GenAI at HMS

The Starting Point

The client’s CRM system contained large amounts of free-text feedback from customers and sales representatives. The content included information relevant to business analyses and strategic evaluations, but it was not structured in a uniform manner.

The volume and heterogeneity of the texts made systematic evaluation difficult. In addition, there were categories with similar content that were difficult to distinguish using simple rules alone. As a result, the information could only be incorporated to a limited extent into downstream analyses and decision-making processes.

Fünf HMS-Mitarbeitende aus dem Content-Tagging-Projekt im Austausch an einem Besprechungstisch.

The HMS Solution

Automated Content Tagging with LLMs

HMS developed a system that automatically processes unstructured CRM text and assigns it to predefined content classes.

Multidimensional Classification

Large language models classify texts along several domain-specific dimensions. The classification results are used to generate structured tags for further processing.

Combining LLMs with Rule-Based Logic

HMS supplemented the model-based classification with rule-based logic. This logic improves classification for heterogeneous content and classes that are difficult to distinguish based on content.

Implementation with Azure OpenAI and LangChain

The application was developed using Azure OpenAI, GPT models, LangChain, and Python. HMS optimized tagging performance and derived technical best practices from the implementation.

Production Integration with Palantir Foundry

HMS integrated the system into Palantir Foundry and designed the processing to scale.In project testing, the solution outperformed previously used approaches based on NLP Transformer models in terms of tagging performance.

Customer Benefits

Turning CRM Free Text into Structured Data for Analysis

Automated content tagging turns previously difficult-to-analyze information into structured data for downstream use.

Structured

Turning Free Text into Structured Data

Free-form text is assigned to predefined content classes and is then available as structured data.

Searchable

Find Specific Content

The tags generated make it easier to search, filter, and group CRM content by topic.

Multidimensional

Classify Content Across Multiple Dimensions

Texts can be classified according toseveral domain-specific dimensions and used for subsequent analyses.

Actionable

Supporting Data-Driven Decisions

Information from customer and sales feedback can be incorporated more systematically into evaluations and strategic decisions.

The solution thus supports both exploratory data queries and in-depth analyses of key metrics from management reports.

Our Strengths, Your Benefits

Combining LLMs and rule-based classification

HMS was responsible for the development and deployment of the content-tagging system. In this project, we combined LLM engineering with rule-based logic, performance optimization, and integration with Palantir Foundry.

What You Can Expect from HMS

In automated content tagging, model control, classification logic, and data processing need to work together effectively. To achieve this, HMS combines large language models with domain-specific rules and traditional software engineering.

Development of the content tagging system through to production deployment
Rule-based control for heterogeneous documents and similar content classes
Performance optimization and design for scalable processing

Christoph Bergen

CoE Lead GenAI

Would you like to carry out a similar project?

Talk to HMS about how your data, AI, or software initiative can be assessed from a business perspective, implemented technically, and integrated into your existing system landscape.

Discuss the project

Next Steps and Further Study

Generative & Agentic AI

Learn how generative and agentic AI help companies efficiently leverage knowledge, automate processes, and productively deploy intelligent applications.

Learn More About Generative & Agentic AI