
From Free-Text Party data to AI/BI-ready features
Payment messages make up a significant share of interbank traffic. Yet sender/receiver details often arrive as loosely formatted free text (names, streets, cities) across SWIFT MT, ISO 15022 and ISO 20022. That variability blocks automation and forces manual parsing and enrichment.
PtyParser converts messy party fields into structured, machine-readable data, improving entity identification, sanctions/screening accuracy, and downstream analytics.
Key contributions
- Entity recognition: Detect and label entities (persons, organizations, locations) using spaCy1.
- Address parsing: Break free-text into structured components (house number, street, city, region, postcode, country) with Libpostal2.
- Standard-agnostic: Works across MT and ISO 20022 messages; robust to format variation.
- Simple API: Raw strings in, structured JSON out — easy to embed in ETL and screening workflows.
- Open source: Available as a Python package on PyPI3.
Use cases
Compliance & Screening: Reduce false positives by normalizing names and addresses; support AML4/CTF5 investigations with consistent entity extraction.
- Normalize party data before sanctions screening.
- Surface risky patterns with standardized entities and locations.
Operations & Data Quality: Improve straight-through processing (STP) and reconciliation with structured party fields.
- Automate enrichment of sender/receiver details.
- Reduce manual corrections and exception handling.
Analytics & AI: Generate clean features for BI6 and ML7 — from fraud signals to client profiling and forecasting.
- Create consistent features (entities, geos, postcodes) for downstream models.
- Enable reliable dashboards with standardized party attributes.
Installation
PtyParser is available as an open source Python package8.