The applicant who looked too thin

Synthetic identity fraud — stitching real and fabricated data into a person who doesn't exist — often incubates inside a thin credit file. So the industry's instinct has been to treat thin files as guilty until proven innocent.

The cost of that reflex is enormous and invisible: real people — new-to-country immigrants, young adults, the credit-invisible — get rejected or frozen because they pattern-match to risk. Every one is a lost customer and a false positive you paid for.

• A threshold-based model can only see the thinness of the file. It can't distinguish a fabricated history from a genuinely new one — so it paints both with the same brush.
• The signal that separates them isn't file size. It's behavior and intent over time: a real new customer behaves like a person building a life; a synthetic behaves like an asset being inflated for a payday.

• Thin file, yes — but coherent: real address history, an employer, device and contact details that don't reappear across other applicants.
• Activity that looks like living (payroll, recurring bills) rather than priming (rapid limit-building with no real spend).
• No shared device fingerprints, recycled identity fragments, or application data echoing across 'unrelated' people.
• Patience: a genuine customer isn't racing toward a maximum-balance cash-out.

• Stop auto-declining on thinness alone — score behavior and intent, not just file depth.
• Reserve the synthetic flag for corroborating signals (shared devices, recycled fragments, prime-and-bust velocity), not the absence of history.
• Measure your thin-file decline rate as a false-positive cost, not just a risk control.