Select Page
customer logo
Customer
Success Story

Transforming AI Reliability by Building Knowledge Graph-Powered Customer Support

Avalara overcame the limitations of vector-based RAG by implementing a DOM GraphRAG proof of concept model. They used their existing DITA structured content to achieve 100% precision in content mapping. This established a foundation for reliable, mission-critical AI applications in tax and financial services.

The Client

Leading provider of tax technology software solutions serving businesses worldwide with automated tax compliance, calculation, and reporting services Avalara's vector-based RAG model lacked the accuracy needed for mission-critical tax applications, encountering a "Precision Paradox" where improved accuracy led to user dissatisfaction with errors

The Challenge

Avalara's vector-based RAG model lacked the accuracy needed for mission-critical tax applications, encountering a "Precision Paradox" where improved accuracy led to user dissatisfaction with errors

The Solution

Avalara implemented a DOM GraphRAG proof of concept model using GraphDB to leverage their existing DITA structured content, creating a reliable knowledge graph foundation for trustworthy AI solutions.

Technical capabilities

  • Converted DITA schema into RDF ontology for 100% precision content mapping to knowledge graph
  • Automated DITA-to-RDF conversion preserving content relationships and hierarchical structure

Business outcomes

  • Established deterministic, fact-based AI responses overcoming vector RAG precision limitations
  • Created "Content as a Service" architecture supporting multiple applications beyond initial chatbot

The Challenge

Avalara aimed to enhance customer support by providing more accurate, relevant, and reliable content. They wanted to improve customer satisfaction, support case deflection, customer self-service, time-to-value for new customers, customer retention, and personalization.

Their internal AI/ML teams initially explored vector-based RAG models to power in-app and support chatbots. However, they quickly encountered the “Precision Paradox”. As model accuracy improved, user expectations rose. This made customers less tolerant of errors and irrelevant results, leading to greater dissatisfaction.

The engineering teams realized that vector-only RAG models were precision-limited and could not provide the reliability required for mission-critical tax and financial applications.

The Solution

Graphwise collaborated with Avalara’s team to build a DOM (Document Object Model) GraphRAG model using GraphDB. It capitalized on their existing DITA (Darwin Information Typing Architecture) structured content as the foundation for a reliable knowledge graph. 

Converting the DITA schema (DTD) into an RDF ontology served as “ground truth” to automatically map structured content into a knowledge graph with 100% precision. The DITA Open Toolkit was used to automatically transform DITA topics into RDF, preserving the content relationships and the hierarchical structure for granular retrieval.

The solution enriched content with domain-specific taxonomies, improving Named Entity Recognition and reasoning capabilities. The knowledge graph became the definitive source for content retrieval using SPARQL queries, delivering precise, deterministic answers while overcoming probabilistic limitations of vector-based retrieval.

The project’s success stemmed from collaboration between content, engineering, and knowledge management teams from the start, ensuring critical content management requirements were built into the solution.

The Impact

“Critical content management capabilities are virtually absent from most vector-based RAG models. One can argue that the lack of vital content management capabilities in current models is tantamount to professional content management malpractice, exposing a company’s reputation, if not liability.”

Michael Iantosca, Senior Director, Content Platforms & Knowledge Engineering

The successful DOM GraphRAG proof of concept demonstrated a clear path to reliable, trustworthy AI solutions for mission-critical applications. Key achievements included:

  • Improved accuracy and reliability through fact-based, explainable answers, eliminating precision limitations of vector-only models.
  • Enhanced customer support metrics including better customer satisfaction, retention, and faster time-to-value. 
  • Architectural transformation from document-centric systems to a “Content as a Service” model with APIs supporting multiple applications beyond the initial chatbot.
  • Foundation for future growth enabling incremental addition of domain-specific knowledge and integration with other data sources over time.

The DOM Graph RAG proof of concept proved that human-curated, structured content is crucial for building trustworthy AI solutions. The knowledge graph powered intelligent content improved business metrics and established a foundation for continuous improvement. 

Table of Contents

Details

Solution: GraphDB, GraphRAG
Contact us

Facing Similar Challenges?

Struggling with unreliable AI outputs, precision limitations in vector-based systems, or need for mission-critical accuracy in your AI applications?

Whether you're dealing with probabilistic AI models that can't meet reliability standards, fragmented content management workflows, or regulatory requirements demanding explainable AI, Graphwise can help you:

  • Transform structured content into reliable knowledge graphs with 100% precision mapping
  • Build deterministic AI solutions that provide explainable, fact-based answers
  • Create scalable "Content as a Service" architectures supporting multiple applications
  • Establish enterprise-grade content management with version control and compliance support

Let’s talk