How are lenders catching deepfake documents? (Featuring Rapid Finance)

I recently spoke with two long-time partners of ours at Rapid Finance: Patrick Lord and Timothy O’Rear. Their team has spent more than a decade underwriting small-business loans, and they’ve experienced the full evolution of document fraud from simple formatting errors to AI-generated statements that look identical to the real thing.

In the first half of 2025, we saw AI-generated and template-based documents grow by more than 200% across our customer base. Rapid Finance is seeing those same patterns, so this episode focused on what has changed, how they’ve adapted, and what the next stage of fraud detection will look like.

How document fraud has changed since 2019

Patrick and Tim both described a dramatic increase in the scale and sophistication of fraud attempts. Fraudsters are using:

‍

• AI tools to modify balances or identities

• Spoofed phone numbers and email domains

• Look-alike business websites created within minutes

• Template-based documents that mimic bank formats exactly

‍

The result is a much noisier environment. There is more high-quality fraud and more low-effort fraud, and it all arrives in far greater volume. Underwriters now face a mix of polished synthetic documents and obvious fakes, with little visual difference between the two.

Tim put it plainly: the job has turned into an arms race.

Why manual review doesn’t scale

Before adopting automation and fraud detection tools, Rapid Finance reviewed every document manually. Underwriters examined thousands of PDFs each week, checking balances, formatting, and basic inconsistencies. Some nights the team stayed up until midnight to clear the backlog.

That approach worked when fraud was easier to spot. But today, many fraudulent documents are so clean that no amount of visual inspection can reliably catch them. Even when X-ray analysis reveals an edit, Tim pointed out that those edits are often invisible to the human eye. At this point, the limitation is not human skill but human capability.

Manual review also created inconsistency. With dozens of underwriters, decisions often varied from person to person. Patrick stressed that building consistent fraud detection across a large team is nearly impossible without shared tooling.

How Rapid Finance modernized their fraud workflow

Rapid Finance reworked their process around early, automated document screening. This ensures that fraud is identified as soon as a bank statement enters the system, rather than only when an underwriter becomes suspicious.

Automated, top-of-funnel screening

Every document is evaluated before underwriting begins. This allows the team to catch fraud they never would have flagged manually and ensures that underwriters spend their time on cases that are actually viable.

Integrated trust scoring

Rapid Finance uses Inscribe’s trust score inside their internal workflow to guide decisioning. The score helps them determine which documents to trust, which require deeper investigation, and which should not progress further. This has significantly increased consistency across underwriting.

A culture of transparency

Patrick and Tim emphasized how openly their organization discusses fraud. Sales, underwriting, and operations share information in dedicated channels, which helps resolve borderline cases quickly and keeps everyone aligned on emerging patterns.

Better broker communication

Because the team now understands exactly why a document appears suspicious, they can prepare broker partners early if a case may require extra verification. This avoids surprise declines and improves trust on both sides.

Measurable impact on speed and quality

The operational improvements have been substantial:

‍

• Faster turn times across the underwriting process
  

• Fewer late-night review sessions

• Less time wasted on applications that will never fund
  

• Pull-through from final review to funding has increased from about 50 percent to 85–90 percent

• Clear revenue protection by catching fraud cases earlier and avoiding downstream losses

‍

Patrick described it well: catching a single major case that would have slipped through can justify the cost of the system. Everything beyond that is margin protection.

What comes next for fraud detection

We also talked about the future of AI adoption in fraud operations. A few themes stood out.

Finding incremental improvements

After several years of rapid change, the next set of gains will come from steady, incremental improvements. Small workflow optimizations compound quickly for large underwriting teams.

Responsible AI deployment

Patrick is clear that AI should support underwriters, not replace them. Teams need proper controls, oversight, and a human review layer to ensure decisions remain reliable.

Staying ahead of fraud trends

Fraud evolves rapidly. The teams that succeed are the ones that adjust quickly and integrate new tools before fraudsters shift tactics.

Tim summed it up well: the next stage is about finding the inches that make the entire workflow more efficient.

Final thoughts

Rapid Finance is a great example of a lending organization that refuses to stand still. They recognized how quickly document fraud was changing, rethought their workflow from the ground up, and built a system that improves both fraud detection and overall underwriting quality.

The broader takeaway is that fraud detection is now a real-time problem. AI is driving the scale of fraud attempts, but it is also enabling better detection. The teams that adapt quickly and pair automation with strong human judgment will stay ahead.