Building Trusted AI for Compliance with Connected Entity Data
How Entity Resolution Cuts False Positives by 80%+ and Makes Compliance Decisions Defensible
Compliance teams face growing challenges as data volumes explode. False positives bury real threats in noise while sophisticated actors exploit gaps in fragmented systems. The solution? Start with clean, connected, explainable entities, then layer AI and policy on top.
Watch this expert-driven panel from the SENZING® 2025 User Conference to learn how entity resolution, paired with high-quality reference data, transforms noisy alerts into reliable signals. “It’s all about speed and how quickly you can resolve these records,” says moderator Dr. Gurpinder Dhillon. “There’s a lot of risk, especially with bad actors trying to get through the cracks.”
Panelists Gurshish Dang (Verisk), Friedrich Lindenberg (OpenSanctions), and Pete James (WireScreen) delivered a roadmap: build a transparent identity layer, test rigorously, and measure results. The proof? Verisk achieved 80%+ suppression of known false positives with no loss of true risk detection.
About the Panelists
Gurshish Dang – Head of Enterprise Data Management at Verisk: connects enterprise data and runs a common referential search platform.
Friedrich Lindenberg – Founder at OpenSanctions: builds an open, publicly scrutinized risk/reference dataset from global watchlists.
Pete James – Director of Training and Data Insights at WireScreen: helps teams leverage Chinese corporate intelligence using native names and social codes.
Dr. Gurpinder Dhillon (Moderator) – Head of Data & AI Ecosystem at Senzing (formerly at Dun & Bradstreet).
The Foundation: Connected Entity Data Comes First
Senzing CEO Jeff Jonas opened the panel with a compelling example: during a previous Senzing Summit keynote, an audience member downloaded OpenSanctions data and fully ingested it before the talk ended, moving from download to searchable and resolved entity graph in minutes. This speed matters when alerts hit in real time. But Jonas emphasized that fast ingestion delivers value only when the resulting entities are explainable and defensible to regulators.
Gurshish Dang reinforces this with Verisk’s 2-billion-record referential dataset, powering enterprise-wide search with 600-millisecond query latency. “We’re using Senzing to solve an online problem,” he explains, noting that underwriting demands both interactive speed and transparent, explainable reasoning and decisions.
Three Types of Messy Data
Friedrich Lindenberg identifies three data problems:
- The internal mess: Fragmented internal systems where entities appear differently across databases
- The geopolitical mess: Rapidly evolving sanctions and ownership structures
- The watchlist mess: Government watchlists built by underfunded teams with limited capacity for precision
“Government watchlists are built by underfunded bureaucracies,” Lindenberg explains. “We’re building a reference dataset that withstands public scrutiny.” Without enrichment, an entry as vague as “Abu Hassan” could match 50 million people, rendering a screening useless. These data quality challenges become even more acute when compliance crosses borders and language barriers.
Cross Border Screening: Language and Regional Identifiers
Pete James tackles cross-border screening and stresses: “An English-language name is not a unique identifier for Chinese corporate entities.” Screening based solely on transliteration misses criminals and flags innocents.
His guidance: anchor on native names and social credit codes, then correlate firmographics so datasets “talk to each other.” This prevents transliteration drift and false matches on common or transliterated names while stabilizing matching as companies and datasets evolve.
Testing and Validating Results Against Trusted Reference Sets
James also recommends continuous test loops: run “data in, data out” cycles against trusted reference sets to validate improvements before production. He calls for vendor transparency: be explicit where LLMs assist versus where LLMs must never drive red-flag decisions.
Gurshish Dang reports that Verisk’s repeated Senzing proof-of-concept (POC) testing consistently suppressed 80%+ of known false positives, improving investigator throughput and accelerating case resolution. Read the full Verisk case study to see the complete story.
Watch the full Big Data London session, Trusted AI Starts Here, to learn how real time entity resolution can power AI you can trust.
Adapting to the New Rules: The 50% Ownership Challenge
Compliance requirements don’t stand still. Pete James highlights a recent regulatory shift: the Bureau of Industry and Security (BIS) extended the “50% rule” – long established for OFAC sanctions – to entity-listed companies, with particular focus on China and corporate evasion practices. This fundamentally changes how compliance teams are required to approach screening.
Teams can no longer rely solely on direct watchlist matches. They must now map corporate ownership structures to identify entities that are 50% owned by sanctioned parties, whether through wholly owned subsidiaries or complex joint ventures that may obscure sanctioned ties. This requires connected entity data that links ownership, control, and beneficial interests across corporate structures, which is exactly the kind of relationship mapping that Senzing entity resolution enables.
Getting Started with Entity Resolution for Compliance
The panel offers a pragmatic playbook for getting started:
- Define risk indicators for your business
- Audit data sources and understand gaps
- Account for evolving ownership rules when mapping control
- Start small with high-impact use cases
- Test in safe environments and iterate
- Measure continuously with a tune-test-measure loop
“Start small… go after your high-impact use cases and find some successes,” Dang advises. “That helps you get credibility. Then you can scale up.”
The Stakes in Practice
Dhillon shares a sobering example: A startup extended credit to what appeared to be a legitimate business, complete with a website and LinkedIn profiles. One week later, the company’s website vanished, phones were disconnected, and online profiles were deleted.
This underscores why entity resolution is critical. Bad actors increasingly use seemingly legitimate digital footprints to exploit gaps in data connectivity. The only defense: unified entity data that connects records across systems and languages, anchors on reliable identifiers, and provides audit trails that explain every match. When done right, you can cut false positives without degrading detection sensitivity, so real risk still gets through.
AI Builds on Data Not Vice Versa
Friedrich Lindenberg cautions against the current GenAI hype cycle, noting that much of today’s enthusiasm will fade without leaving substantial value behind. AI proves valuable for specific tasks – cleaning names, classifying industries – but only when built on connected, explainable entity data.
Gurshish Dang adds that this foundation must operate as a reliable service: “We need the same expectations of data as we have from software.” The consensus: trusted AI-powered compliance starts with clean entities, transparent resolution, continuous testing, and quality data that turns alerts into actionable intelligence.
Video Highlights
00:00:03 – Data to decisions in minutes
OpenSanctions was fully ingested during a keynote, proving entity resolution can deliver operational value in minutes.
00:05:20 – Verisk’s 2B-record platform
How Verisk delivers 600ms queries for real-time underwriting decisions.
00:07:45 – Why watchlists create noise
Government lists contain ambiguous entries (such as “Abu Hassan”) that require enrichment and cross-referencing to become useful signals.
00:10:35 – Cross-border screening challenges
Why English transliterations fail for Chinese entities and how native names plus social credit codes prevent false matches.
00:12:35 – Practical playbook for getting started
Define your risk indicators, audit data sources, and factor in evolving ownership rules like the 50% guidance before testing.
00:20:52 – 80%+ false positive reduction measured
Verisk’s testing results show that explainable entity resolution and continuous validation drive improvements in accuracy.
00:25:10 – Real-world fraud: the disappearing company
How fraudsters create legitimate-looking business identities and why entity resolution is essential for detection and prevention.
Ready to Build Defensible Compliance?
See how Senzing entity resolution can help your organization reduce false positives, accelerate investigations, and make compliance decisions defensible. Watch the full panel discussion above, or schedule a demo to explore how Senzing AI for entity resolution can improve your compliance workflows.
