Why now

The cost of waiting is rising

Credit history is real evidence, but it was never complete evidence. What changed is the pressure around it: AI makes fakes cheaper, thin files are still misread, and every edge-case decision has to be easier to explain.

Length: 489 words
Read time: 3 min read

The moment

Bounded credit evidence is under new pressure

For years, the bureau-visible record was imperfect but workable. Now it is being asked to separate three things it was not built to tell apart on its own: a person with no file, a person recovering from a setback, and a fabricated identity designed to look clean on paper.

Synthetic pressure

Fake lives are easier to manufacture

Estimated U.S. synthetic-identity fraud losses crossed $35 billion in 2023, and public reporting warns that generative AI is making fraud cheaper and faster. Boston Fed.

Thin-file pressure

Real people are still invisible

Roughly 32 million U.S. adults cannot be scored or have files too thin to read. That is not a niche edge case; it is a persistent underwriting gap. Federal Reserve.

Score-shock pressure

One shock can dominate the picture

In 2025, 2.2 million borrowers saw scores drop more than 100 points in one quarter after student-loan delinquencies returned. New York Fed.

The category gap

The market optimized the wrong question

A decade of capital went into scoring the applicant differently: more identity signals, more fraud models, more alternative data, more automated decisions. That helped, but it did not change the shape of the answer.

Most tools still return a number the lender has to trust. The harder problem is producing evidence the lender can open, explain, and replay when the decision is challenged.

The wedge is not another score. It is a reviewable record of why this person, this request, and this moment fit together.

The landscape

Everyone is solving a different corner of the problem

Map the field on two axes — whether the lender gets a record it owns or a score it rents, and whether the tool does one job or unifies detection and decisioning — and the gap becomes clearer. Most providers solve one corner. Kenshiki is built for the upper-right: two jobs, one evidence layer, and a replayable record.

Synthetic identity fraud: competitive landscape

How the field maps on ownership of evidence versus breadth of the job — and the quadrant Kenshiki is built to occupy.

A record you own proof, replayable

A score you rent confidence, black-box

Records, but narrow

Model-risk and audit tooling

Produces documentation, but for one function — not the decision itself.

Two jobs · owned record

Kenshiki

Catches the synthetic and approves the thin file off one evidence layer, sealed into a record the lender owns and an examiner can replay.

One job · a score

Fraud tools, decisioning, and data rails

Fraud tools (SentiLink, Socure), decisioning platforms (Zest, Upstart), and data rails (Plaid, Truv, the bureaus) each do one job and hand back a score or signal the lender rents.

Two jobs · still a score

The bundlers

Incumbents stitch detection and decisioning together — for example, Socure + Scienaptic — but still output a score, not an owned record.

One job detect OR decide

Two jobs one evidence layer

Every established approach either solves one job, rents the lender a score, or produces a narrow record for a single function. Kenshiki’s claim is different: one evidence layer that can catch fabricated lives and recover real applicants, sealed into a record the lender owns.

The governance shift

Defensibility is becoming part of the product

Regulated lenders do not just need a better guess. They need a reason they can inspect before they approve, decline, price, or refer. That makes governance a product requirement, not a policy appendix.

Kenshiki is built for evidence that is defined in advance, bounded to the approved question, and replayable after the fact, so governance is part of the record instead of a reconstruction exercise.

The wedge

Number checks do not prove a lived pattern

The government’s number-checking service can confirm a name-and-number match, but it explicitly does not verify identity. SSA eCBSV documents that boundary.

Kenshiki operates in the gap after the number matches: is there a real life here, does that life fit the obligation, and can the lender explain the answer without pulling raw private data into the wrong place?