.png)
.png)
6 min
The Challenge: Data trapped in unstructured reports
The Risk & Supervision Department needed to assess the ESG exposure of their corporate clients to ensure compliance and anticipate future credit risks. However, the critical data — GHG emissions (Scopes 1, 2, 3), energy consumption, and HR indicators — was buried in dozens of massive PDF annual reports (80 to 400 pages each).
A standard search approach was insufficient. The team needed to distinguish between past performance and future targets, and to standardize data across different corporate reporting styles.
The Solution: A dedicated CSRD ontology
The core of the solution was the generation of a custom CSRD Ontology derived directly from the financial reports. This semantic layer served as the "decoder ring" for the unstructured text, transforming raw paragraphs into structured, analyzable data.
How the ontology structured the chaos:
- Standardizing Metrics: The ontology defined rigid classes for specific indicators (e.g., distinguishing Scope 1 direct emissions from Scope 3 indirect emissions), ensuring that extracted numbers were strictly categorized.
- Timeframe Disambiguation: By structuring temporal properties, the ontology allowed the system to separate a 2023 Result from a 2030 Target, a critical distinction for risk modeling that keyword search often misses.
- Contextualizing Risks: The ontology mapped relationships between quantitative metrics and qualitative descriptions, linking a specific risk (e.g., "Regulatory Change") to its mitigation strategy and financial weighting.
The result: An ontology-powered GraphRAG
With the data now structured by the ontology, Lettria deployed a GraphRAG (Graph-based Retrieval Augmented Generation) agent.
- High-Precision Extraction: Because the AI referenced the ontology’s structure, it could accurately extract quantitative metrics from the database.
- Trend Analysis: The semantic links enabled the team to query qualitative changes over time, facilitating a deeper understanding of client trajectories.
- Outcome: The ontology successfully turned static PDFs into a dynamic knowledge graph, enabling the bank to automate the collection, reliability checks, and consolidation of ESG data.
Frequently Asked Questions
Yes. Lettria’s platform including Perseus is API-first, so we support over 50 native connectors and workflow automation tools (like Power Automate, web hooks etc,). We provide the speedy embedding of document intelligence into current compliance, audit, and risk management systems without disrupting existing processes or requiring extensive IT overhaul.
It dramatically reduces time spent on manual document parsing and risk identification by automating ontology building and semantic reasoning across large document sets. It can process an entire RFP answer in a few seconds, highlighting all compliant and non-compliant sections against one or multiple regulations, guidelines, or policies. This helps you quickly identify risks and ensure full compliance without manual review delays.
Lettria focuses on document intelligence for compliance, one of the hardest and most complex untapped challenges in the field. To tackle this, Lettria uses a unique graph-based text-to-graph generation model that is 30% more accurate and runs 400x faster than popular LLMs for parsing complex, multimodal compliance documents. It preserves document layout features like tables and diagrams as well as semantic relationships, enabling precise extraction and understanding of compliance content.
.png)
.png)
.png)
