Risk & Fraud Glossary

Fraud in which a criminal seizes control of a legitimate user's account — through a leaked password, phishing or a SIM swap — and operates it as if they were the real owner.

Matching the authentication requirement to the risk of the moment: letting low-risk access flow without friction and asking for extra proof only when something departs from the norm.

The notice a creditor must give when it denies credit or offers worse terms, stating the specific principal reasons under ECOA and, when a consumer report was used, the FCRA disclosures.

Screening news and public sources for negative mentions of a person or company — investigations, convictions, scandals — that extend risk knowledge beyond official lists.

The effect of disproportionately attracting the worst risks, because the price on offer is too good for them and too poor for everyone else.

Confirming that a person meets the minimum age required for a service — by document, by an authoritative record, or by biometric estimation from the face — without necessarily revealing their exact age.

Regaining physical possession of a stolen, unreturned or diverted asset — and the window for doing so is measured in hours, not days.

Pooling loans or receivables — auto loans, leases, card balances — into a special-purpose vehicle that issues securities to investors, paid from the cash flows of the pool.

A scam in which the victim themselves is tricked into authorizing a payment to the fraudster — believing in a fake bill, person or opportunity — rather than having their account hacked.

A loan secured by a vehicle the borrower already owns, with the lender recorded as lienholder on the title — from bank auto equity loans to high-cost title loans regulated state by state.

A check of a person's history — criminal, identity, driving, credit, sanctions — before hiring them, renting to them or granting them access to an asset.

Confirming that a bank account exists, is active and belongs to the person who provided it, before using it to pay, receive or validate the account holder's identity.

Recognizing a person by how they interact with a device — typing rhythm, cursor movement, how they hold and tap a phone — rather than by a physical trait like their face.

A risk score built from how a person behaves — mobility, stability, ties and usage — rather than only from the credit history they happen to have accumulated.

The natural person who, at the end of the ownership chain, actually controls or benefits from a company — even when the name on the paperwork is another company or a straw man.

Confirming that the person accessing an account is the same holder who enrolled, by comparing a physical trait — typically the face — against the biometric captured at onboarding.

A test design in which Zarv computes the score without knowing which cases had the event, so the result cannot have been tuned to the answer key.

A technique that estimates the uncertainty of a metric by resampling the data itself with replacement hundreds of times and watching how much the result varies. It is used when there is no closed-form formula for the margin of error.

Distinguishing automated access by scripts and robots from real human users, to block attacks at scale such as mass account creation, card testing and credential stuffing.

A long-game fraud in which the fraudster builds a good credit history to earn high limits and then, at a chosen moment, draws everything at once with no intention to repay.

Shows what percentage of the events is already captured as you walk the base from worst to best score. The sooner the curve rises, the more the risk is concentrated in the worst scores.

A monthly fee for the use of a car that bundles the vehicle with insurance, maintenance and registration, with shorter commitments and more flexibility than a lease.

Insurance that pays for freight lost, stolen or damaged in transit — bought by the carrier to cover its legal liability for the load, or by the shipper to cover the goods themselves.

Stealing goods in transit or in storage along the supply chain — by hijacking, burglary of trailers and yards, or deception such as fictitious pickups — a crime that usually depends on inside information.

The US GAAP accounting standard (ASC 326) that requires lenders to reserve, from the day a loan is booked, for the credit losses expected over its entire life.

The US Bankruptcy Code procedure in which a business keeps operating while it restructures its debts under a court-confirmed plan, protected from creditors by the automatic stay.

The reversal of a card payment that the cardholder disputes with their issuer, which pulls the money back from the merchant through the card network.

A debt a creditor has sent to collections or charged off, reported to the credit bureaus as negative information that can stay on a credit report for about seven years.

The set of policies, controls and responsibilities that ensures a company follows laws, regulations and its own rules — and can prove that it does.

A range around an estimated number that shows the uncertainty of the estimate. A 95% CI is built so that, if the experiment were repeated many times, 95% of the ranges would contain the true value.

Following a financed, leased or insured asset across the whole contract, so risk changes surface before the loss rather than after it.

Keeping customer knowledge current across the whole relationship, re-assessing risk whenever something changes, instead of verifying once at onboarding and never again.

The record a consumer reporting agency keeps of a person's credit accounts, payment history, inquiries and certain public records, governed by the Fair Credit Reporting Act.

A number, usually between 300 and 850, that predicts how likely a person is to repay credit on time, computed from the data in their credit reports.

Assessing a person's or company's ability and willingness to repay, done before extending credit and setting its price and terms.

A search of court and repository records for a person's arrests and convictions, used in hiring, tenancy and licensing, and regulated in the US by the FCRA, equal-employment law and fair-chance rules.

Assigning each customer a risk level — typically low, medium or high — that determines how deep the due diligence is and how intensely the relationship is monitored.

Confirming a person's data by cross-checking it against an authoritative or issuing-source database — not just reading the document, but verifying that the data exists and matches at the source.

Each of the ten 10% slices you get by ordering a base and cutting it into equal parts. In risk, the worst decile is the group of 10% with the worst scores.

How many times a score group concentrates more events than the average of the base. A lift of 1.8x in the worst decile means the 10% with the worst scores have 1.8 times the event rate of the whole portfolio.

The component that turns risk signals into a verdict — approve, decline, review or ask for more verification — applying rules and models automatically and consistently to every case.

The part of a covered loss the policyholder pays out of pocket before the insurance pays the rest — the higher it is, the lower the premium.

An image, video or audio clip synthesized by artificial intelligence to reproduce the face or voice of a real person doing or saying something they did not.

Delinquency is a missed payment past its due date; default is the point where the lender treats the debt as not going to be repaid under its terms — in US banking, typically at charge-off.

A statistical test that tells whether the difference between two AUCs measured on the same base is real or chance. It is the standard way to compare the power of two scores without assuming a distribution.

Recognizing and scoring the device and connection behind an access — from hundreds of browser, system and IP signals — to detect fraud without depending on the claimed identity.

An official, verifiable version of a person's identity issued to their phone — such as a mobile driver's license or state mobile ID — digitally signed by the issuing authority and checkable without the physical document.

The technical examination of a document to establish whether it is authentic — checking security features, fonts, printing and signs of tampering that are not visible to the naked eye.

Producing or using false identity documents, or using another person's identifying information without authority, to commit fraud — federal crimes under 18 U.S.C. §1028 and §1028A.

An investigation carried out before a significant decision — onboarding a vendor, signing a partner, acquiring a company — to uncover risks the other party would not disclose.

The full set of cases that qualify for a test — the population the sample is drawn from — defined by explicit cutoff, deduplication and date rules before any draw.

Technology that compares the features of a face against one or more reference images to confirm or establish a person's identity.

Fraud committed by the legitimate account holder themselves — using their own real identity to obtain credit or goods they never intend to pay for, or disputing in bad faith a purchase they actually made.

The US law that prohibits bribing foreign officials to obtain or keep business and requires SEC-reporting companies to keep accurate books and adequate internal accounting controls.

Deliberate deception to obtain an undue advantage — credit, an insurance payout, an asset, a payment — at the expense of whoever trusted the information presented.

A model in which several institutions share fraud signals — devices, identities and patterns already seen as malicious — so a scam suffered by one protects all the others.

A geographic perimeter defined around a monitored asset, where crossing in or out fires an automatic alert.

A measure of how well a score orders risk: 0% is random ordering and 100% is perfect separation between cases that had the event and cases that did not. For models it equals 2 × AUC − 1.

Checking whether a face already exists in the database before accepting a new enrollment, searching it against every record (1:N) to stop the same person from enrolling several times under different identities.

A layer that coordinates the many checks in an identity flow — document, biometrics, databases, watchlists, anti-fraud — deciding which to run, in what order, and what to do with each result.

The unauthorized use of another person's real data to open accounts, obtain credit or commit fraud in their name — where the data is genuine, but the person using it is not its owner.

The process of confirming that a person is who they claim to be, combining reading the ID document, checking that it is authentic, and proving that the person presenting it is its live owner.

How much a new score adds to the model already in use, measured by the improvement in ordering power when the two run together — not by what the new score does on its own.

Fraud that inserts a fake video or image directly into the verification stream — via a virtual camera or app interception — without going through any real camera.

The instant-payments form of authorized push payment fraud: the victim is persuaded to send the money themselves over a real-time rail like Zelle, FedNow or wire, so the transfer clears before the scam is noticed.

A policyholder's or third party's request for payment after a covered event — collision, theft, fire, damage — which obliges the insurer to investigate it and, if valid, pay.

Faking, causing or inflating a claim — or misrepresenting the risk at purchase — to obtain a payout or a price one is not entitled to.

The written contract between insurer and policyholder that sets what is covered, for how much, under which conditions and at what premium.

The amount the policyholder pays for coverage, set from the insurer's expectation of loss plus expenses and profit, within rates regulated by each state.

Verifying a company before doing business with it — that it exists and actually operates, who controls it, and which individuals ultimately own it.

The process of identifying and profiling a customer — confirming who they are, assessing their risk and keeping that current — before and throughout the relationship.

Analyzing crypto-asset transactions and wallets to assess the risk of the source and destination of funds, using the public blockchain trail instead of relying only on the customer record.

A search of the civil court records in which a person or company appears — as plaintiff or defendant — across federal and state courts, used to size up legal, financial and reputational risk before a decision.

Verifying that the person in front of the camera is present and alive at the moment of capture — not a photo, a screen replay, a mask or a synthetic video.

The ratio of losses incurred to premiums earned — the number that tells you whether the pricing was right.

Adjusting a contract's price during its term, when observed risk changes, instead of waiting for renewal.

Concealing or disguising the nature, source, ownership or control of criminal proceeds so they appear lawful — and the Bank Secrecy Act regime built to detect it.

Fraud committed at sign-up — opening an account or applying for credit with a stolen, synthetic or fake identity — to extract value before the fraud is discovered.

Reading, by tap, the chip embedded in passports and electronic ID documents to obtain the digitally signed official data, instead of relying only on a photo of the page.

US economic sanctions administered by Treasury's Office of Foreign Assets Control, whose Specially Designated Nationals list names people and entities whose assets are blocked and with whom US persons generally may not deal.

The right of consumers to have their financial data shared, at their request, with third parties they authorize — in the US grounded in Dodd-Frank §1033 and a 2024 CFPB rule that is now being reconsidered.

A person who holds or has held a prominent public function — and, by extension, their family and close associates — whose relationships call for risk-based due diligence; US rules speak of senior foreign political figures.

A scam in which a fake email, text, call or website impersonating a trusted company or person gets the victim to hand over passwords, account numbers, one-time codes or card details.

A score computed only with the information available at the reference date — typically the start of the contract — so it does not see the future and the test result stays valid.

An attempt to fool a biometric check by presenting an artifact to the camera — a photo, a video, a mask or a deepfake — in place of the live person.

A document that shows where a person lives — typically a recent utility bill or bank statement in their name — collected at onboarding to satisfy the duty to know the customer.

Frauds that pay earlier participants with money from new ones — a Ponzi scheme disguised as an investment, a pyramid scheme disguised as a business opportunity paying for recruitment.

Drawing a test's cases by pure chance, with a recorded seed, without hand-picking typical examples or filtering by channel, region or product — so the sample represents the base.

Approving, declining or pricing inside the transaction itself, at a latency low enough that the customer never notices the analysis happening.

Buying, receiving, concealing or selling property known to be stolen — the crime that turns stolen vehicles and cargo into money through fences and chop shops.

A representation of the connections between people, companies, addresses, devices and assets, used to surface ties no single record shows.

A model in which a person verifies their identity once and reuses that proof across many services, carrying a verifiable credential instead of redoing KYC from scratch at every sign-up.

Adjusting a sample back to the proportions of the population when the groups were drawn at an artificial ratio, so that estimated frequencies and lift reflect reality.

Renting a vehicle, usually by the week, to people who drive for ride-hailing or delivery apps — a high-mileage, high-turnover segment in which the risk sits in who is driving.

Validation of the score on a sample of your own historical base, with success criteria fixed in advance, to measure the real gain without touching the operation.

Sealing the delivered scores with a cryptographic hash before seeing the results, so no score can be altered once the events are known.

Money or a card authorization held at the start of a rental to cover damage, unpaid charges or a late return, and released when the contract closes clean.

A scam in which the fraudster transfers the victim's phone number to a SIM under their control, taking over the calls, texts and authentication codes that would reach the victim's phone.

The identity, age, location and integrity controls that state-licensed sportsbooks must apply to every bettor, since the Supreme Court left sports betting to each state in 2018.

In the US, insurance is regulated by each state's insurance department, coordinated through the NAIC, under the McCarran-Ferguson Act's rule that insurance is subject to state law.

Someone who appears formally as the buyer, borrower or account holder while another party is the real beneficiary — used to separate risk, or dirty money, from whoever actually carries it.

A three-party guarantee in which a surety company backs a principal's obligation to an obligee — paying if the principal fails to perform — and then seeks reimbursement from the principal.

Fraud committed with an identity fabricated from a combination of real and invented personal information, built to pass verification without corresponding to a person who exists.

A consumer whose credit file is too thin, too stale or entirely absent to produce a score — not because their risk is high, but because they have not borrowed formally enough to leave a trace.

When an insurer decides a damaged or stolen vehicle is not worth repairing or recovering and pays its actual cash value instead, usually after which the car gets a salvage title.

Tracking a customer's transactions over time to detect patterns suspicious of money laundering or fraud and to report them to the authorities.

A rule requiring whoever transfers value — including crypto-assets — to send the originator's and beneficiary's information along with it, so the data "travels" with the transaction across the chain.

A rented or leased vehicle the customer keeps past the end of the agreement without consent — a contract breach that state law can treat as theft or embezzlement once intent is shown.

The state laws that govern how businesses collect, use and share personal information in the absence of a general federal privacy law — led by California's CCPA as amended by the CPRA.

Auto insurance priced on how the vehicle is actually driven — miles, time of day, braking, speed — measured by telematics, instead of only on the profile at the time of purchase.

A device installed in a vehicle that transmits its location and, often, usage data, so the asset can be followed remotely and acted on in case of theft, non-return or default.

The lender taking back a financed or leased vehicle after default, usually without going to court, as long as it does so without a breach of the peace.

The lender's security interest in a financed vehicle, recorded on the state certificate of title, which lets it repossess the car on default and keeps the title from transferring cleanly until the loan is paid.

The federal crime of using interstate or foreign wire, radio or television communications — today, email, phone, internet and electronic payments — to execute a scheme to defraud or to obtain money or property by false pretenses.