The two started asking other people in the industry about it. “We had hit on this unarticulated pain, something everyone had an issue with,” Berlind said. “No one really understood how big of a problem it was. Literally, no one had been talking about this.”

The duo had “thousands of documents that had been submitted to us over the years.” So, they hired an engineering team that anonymized the data and then applied machine learning, a form of artificial intelligence, to better understand the patterns and predict whether a document was real or an edited fake.

To understand the approach they took, it’s necessary to understand the concept of metadata. This means data about the data in a document, database system, or other place. Metadata explains what the data represents. It could be an explanation of a type of data, the name or author of a document, version, or much more.

All documents have at least some metadata embedded in them. Snappt’s engineers used machine learning to look at the metadata in what became 11 million documents over eight years and find ways of detecting when they were modified. “We have over 2 million apartments that use our software, over 100,000 applicants a month,” Berlind says.

To make their system work, they had to insist that all supporting documents had to be downloaded from original sources like employers or banks and then uploaded to Snappt’s systems. If not, then someone could do things like edit a PDF, print it out, scan it in, and then send it in. The company’s systems can detect if a document is a digital copy, even if scanned in.

This has some limitations. If an applicant can’t download PDFs of needed documents for whatever reason, there is no valid electronic record. But the limitations might be necessary. Calling banks, employers, or even the IRS to check the validity of information would be complex and time-consuming.

And the need to check is only getting larger. Over the last 12 months, 10% of applications came with fraudulent documents. It used to be that only 2% of the buildings had problems with fraud — now it’s a full quarter. And there are even organized criminal groups that make things even more complicated. They create fake companies, get included in legitimate payment systems, and then sell sets of documents to would-be fraudsters.

Berlind doesn’t know the percentage of people who might do this to get into an apartment and pay rent but is “one flat tire away from not being able to pay their rent,” or if a gang looks to run a drug or human trafficking operation.

“It’s bad for the property owner and manager and it makes things for the residents as well,” he said.