Exposing Hidden Connections: How Knowledge Graphs Helped Us Identify PPP Fraud—and Can Help You, Too

When the COVID-19 pandemic struck, the U.S. government moved swiftly to deploy the Paycheck Protection Program (PPP) to support small businesses. Speed was critical—but so was scale. In a matter of weeks, the Small Business Administration (SBA) and its partners issued more than 8 million loans through over 5,000 lenders, totaling almost $800 billion. The volume of data was staggering. And as with any emergency response, cracks in the system began to show.

Fraudsters saw an opportunity. Shell companies, synthetic identities, and hidden ownership networks were used to siphon taxpayer dollars meant for struggling Main Street businesses. The SBA Office of Inspector General (OIG) later reported substantial concerns with data accuracy and transparency. The need was clear: better tools, better data, and a smarter way to see the full picture.

At Redhorse, we wanted to use this as an opportunity to efficiently identify fraud in these types of programs, because while the PPP is over there are many organizations that face similar fraud challenges.

A Deceptively Simple Question

At its core, the PPP challenge boiled down to a seemingly simple question: “Who owns what?” But anyone with experience in corporate registries, financial compliance, or investigative analysis knows the answer is rarely straightforward. Businesses can be owned by other businesses. Individuals may control multiple entities through layered relationships. Ownership can be shared, obscured, or even deliberately hidden.

Answering “who owns what” required a marriage of subject matter expertise and powerful technical architecture. We brought both to the table to tackle this problem.

Building the Technical Backbone

Our team designed and implemented a system that ingested raw PPP loan data and enriched it with external sources: public records, watchlists, conviction databases, and commercial ownership datasets. But we didn’t stop there.

We used high-accuracy entity resolution (via Senzing) to identify and connect duplicate or alias identities. We then employed a native graph database (Neo4j), augmented by Hume’s graph analytics platform, to model the complex, constantly shifting relationships between entities. This allowed analysts to traverse networks of control and ownership to find the different patterns of risk and evaluate each case with intuitive visual tools.

Our custom-built “Accountable” interface sat on top of this infrastructure, offering non-technical users a way to explore the data through configurable dashboards and search experiences tailored to fraud detection use cases.

Human Expertise in the Loop

Technology alone doesn’t solve fraud. It only makes solving it faster, smarter, and more scalable. What made our work successful was the deep partnership between our data scientists and analysts who understood both the policy intent of PPP and the many ways it could be gamed.

We worked hand-in-hand with federal stakeholders to identify what suspicious patterns looked like in practice: multiple loans tied to the same beneficial owner, unusually fast disbursement patterns, geographic clusters of suspicious applicants, or loans linked to individuals with a fraudulent past. By embedding this expert knowledge into rule-based filters and machine learning models, we could surface actionable leads, not just statistical noise.

Delivering Results and a Blueprint for the Future

The result was a living, breathing knowledge graph that adapted as new data arrived and new fraud tactics emerged. It could accelerate investigations, enable continuous program-level analytics, and create a permanent data asset that could be used for oversight and refinement of future funding programs.

And this wasn’t our first time solving problems like this. Redhorse has delivered similar knowledge graph solutions for the Department of Defense, building global-scale networks of scientific collaboration and cross-domain intelligence pipelines. But this project underscored something we’ve long believed: the key to mission success is the combination of agile engineering, powerful data infrastructure, and domain-savvy human insight.

The Bigger Picture

The recent 60 Minutes exposé on fraud in U.S. government programs only scratches the surface. Relief efforts, disaster response, grant programs—any initiative that moves money fast and at scale will face similar threats. The answer isn’t just more compliance paperwork or delayed payments. It’s smarter systems that surface hidden risks before they cause large-scale damage.

We know how to build those systems. And we’re ready to help others do the same.

If your agency is navigating similar complexity—if you need to connect the dots across people, entities, and events to stop fraud before it happens—let’s talk. Email graphs@redhorsecorp.com and we’d be happy to show you what’s possible when good data, strong architecture, and expert judgment come together.