Earlier this year, CBS News reporters set out to investigate 89 licensed hospice companies. The task was simpler than it seemed at first glance: state records tied these companies to a single, 32,000-square-foot Los Angeles building that also housed a salon, a law office and a modeling agency. The building’s owner said only 12 were still there. The gap between the business registry and on-the-ground reality was clearly a problem.
When I saw that story, I did what any curious technologist in my position would do; I started thinking about finding those ghost businesses automatically using modern data analysis techniques. I turned our tools loose on it.
Trulioo’s business data covers more than 100 million U.S. business records, and its value here is in how those records connect. Linking each company to its directors, addresses and owners turns a registry into a map. My team quickly found clusters and registration spikes that appeared to confirm what federal investigators had spent years documenting manually. Many of these providers have been billing Medicare for years and collecting federal tax dollars.
The CBS News story is not an isolated case, unique to Southern California hospice care. We need to understand it as a window into a fraudulent architecture that operates in many different forms across industries and states and around the world. While most of the conversation about synthetic fraud has centered on creating fake people, fake businesses are the quieter half of the same problem, and likely much more impactful from a financial fraud standpoint.
Catching ghost businesses
By the time federal authorities finished connecting the dots on California’s hospice crisis, they had charged 15 people across nine interconnected investigations and estimated $3.5 billion in Medicare fraud from Los Angeles County alone. One licensed physician was linked to 63 separate hospice facilities, billing Medicare more than $35 million in a single year. A couple with prior tax evasion convictions operated a hospice under their daughter’s name.
Every one of those entities was a registered business. Each had a listed director. Each had an address. By every measure a standard Know Your Business check evaluates, the companies were legitimate. And standard verification, which evaluates entities one at a time, is exactly the wrong tool for catching fraud that exists in the relationships between businesses.
The California hospice case is also instructive because it shows the deliberate nature of this particular fraud architecture. Ghost businesses, which are paper companies that exist to bill for services that may never be delivered, don’t advertise their fakeness. They register properly, build websites, list phone numbers in public directories, and size themselves strategically. They cluster geographically to exploit the gap between what regulators check and what they miss.
The pattern gives them away. In Los Angeles County, the number of hospice providers grew by more than 1,500% between 2010 and the state’s 2022 audit, which was six times the national average relative to the county’s elderly population. That spike was visible in the registration data, along with the clustering, the shared director relationships and the address density. None of those signals would show up in a single entity check.
When my team layered those signals against our data, the fraud patterns lit up the screens. Postal codes in Van Nuys, North Hollywood and Glendale emerged as hubs in the entity graph. In one zoomed subnetwork, three directors shared ownership across two hospices with nearly identical names. Invisible in any individual lookup. Obvious at the network level.
Uncovering shells within shells
Bad actors building fraudulent businesses don’t usually stop at one fake company. These criminals build entire networks of shell corporations whose beneficial owners are other shell corporations, registered across multiple jurisdictions, with legitimate-looking surface entities and illicit activity buried underneath.
This means the hardest task in detecting business fraud is finding the layered ownership structure specifically designed to defeat verification. We look beyond the owner of a single fraudulent business and consider whether that person also owns ten others clustered in the same postal code, registered in the same six-month window.
Finding those connections requires graph database technology, which lets us trace connections despite layers of obfuscation. When pulling on suspected shell company threads, we often find the same entity at the center of dozens of businesses that appear entirely unrelated from the outside.
This is the problem ultimate beneficial ownership discovery exists to solve: a registry that lists the owner-of-record. Tracing who actually stands behind a company, through layers of other companies, is what separates a legitimate business from a front.
This same pattern shows up on social media and online marketplaces. Shoppers can see this anytime they find a dozen listings for the same product, clearly from the same manufacturer, just with slightly different logos.
The challenge is that some of these listings are legitimate resellers, right alongside those that aren’t. The line between a fake seller and a ghost business is thinner than most people assume, and the underlying network structure is often identical.
Addressing the forward-looking problem
Most business verification is backward-looking. You check what a company has done, what its compliance record shows and what its registration history looks like. The blind spot is that very few people ask what it’s likely to do next. That ability to look ahead is predictive and possible with modern data techniques including machine learning and AI.
The hospice fraud rings operating in Los Angeles ran long after their entities passed initial registration checks. They kept billing taxpayer money as ownership shifted, new shells formed and directors spread to additional clusters.
Catching a ghost business at onboarding is one checkpoint. These operations stayed hidden because nobody looked again once registration cleared. Risk has to be assessed before you spend on verification, at onboarding, when you set a risk threshold, and continuously afterward. That means holding years of historical business information and tracking businesses in real time, with thousands going inactive each day while thousands of new ones appear. Observed over time and projected forward, that data has enabled us to build what amounts to an early warning system for business fraud. Each flag also has to hold up when a regulator asks why, with an auditable line from the decision back to the data behind it.
Emerging fraud patterns are still hard to catch before clusters become statistically visible, and attackers are using AI to cut the cost of building ghost businesses. That means more of them, registered more convincingly, across more jurisdictions. Defenders can use the same technology, but only if the underlying data is connected and current.
Most organizations still check one business at a time. The fraud they’re up against is built as a network, and detection needs to be built the same way.