OptiArms

Application fraud

An identity assembled from parts reuses the parts.

Synthetic identities, stolen documents and mule accounts are built out of details that already exist. The same device across four applications, a document under a second name, an account that took a disbursement last month.

The problem

What this is for

A synthetic identity is designed to pass the checks it knows about. What it cannot avoid is reuse: assembling identities costs effort, so the parts get used again. Seeing that requires reading across applications rather than scoring one, which is the thing an origination flow is not shaped to do.

The workflow

8 stages, and what runs where

Each stage names the module behind it. The chip says whether it runs on your network or ours.

  1. In your systems

    An application is submitted

    Your origination system posts the application, the applicant and the account, keyed on your own identifier.

  2. In your systems

    Your stack supplies the device identifier

    An opaque id your own fraud stack already computed. The platform derives none of its own and measures nothing about a person.

  3. At intake

    The event is accepted or refused, in writing

    Every application is accounted for, including the ones no rule fires on. That record is what an examiner asks to see.

  4. At intake

    The details join the ledger

    Phones, emails, addresses, device ids, account and document hashes, appended against the applicant with the application they came from.

  5. On the platform

    The cross-reference runs across your own book

    A device under four applicants, a document under a second name, an account that has taken a disbursement before. Each writes one sentence with its evidence.

  6. On the platform

    Evidence accumulates on the applicant

    Different rules pool as different witnesses; one rule repeating is discounted. The band is recomputed from the rows and is reproducible by hand.

  7. On the platform

    A case opens for your fraud queue

    Above the floor you set. If you wired one up, a webhook asks your system to pause the application while a person looks at it.

  8. With a person

    Your team decides, and owns the decision

    The case is a prompt with reasons. Any decline, and the notice it requires, is made and sent by you.

What it runs on

Banks, credit unions, fintech lenders, and mortgage, on the lending and onboarding configuration. Verticals are settings over one platform, never separate products.

You supply the applications

Applications, applicants, accounts, device identifiers and document hashes, from your origination stack.

We supply the ledger and the rules

The cross-reference across your own book, the published rules, the case queue, and the audit record of every verdict.

You keep the decision

A pause asks your system to hold an application for a person. Approving, declining and notifying are yours.

Why this shape

What the design buys

It reads across applications

The reuse that identifies an assembled identity is invisible inside one application and obvious across a book.

Reasons an examiner can read

Every finding is a sentence with evidence attached, and the band is reproducible from the stored rows and the published constants.

No model risk added to yours

The rules are published and readable. There is no opaque score to document, defend, or explain to a regulator.

Limits

What it refuses

Stated here rather than found later. Each of these is enforced somewhere in the code, not only written on a page.

This is not a consumer report

It is used for fraud prevention and is never a factor in eligibility, terms or pricing. It is not a credit bureau and does not act as one.

Nothing here declines an application

There is no decline action in the product. The adverse decision and the notice it requires are the lender's, which is where the law puts them.

No body and no habit is measured

No face, no voice, no typing pattern. A device identifier is one your own stack computed and handed over, the way an address is.