Glossary

Credit decisioning, in plain terms

Concise definitions of the vocabulary a modern credit operation runs on.

Data

Deterministic data: Deterministic data is verified directly at the source that originates it, such as employment reported by the relevant authority. It contrasts with inferred or self-declared data, which may be estimates or outright false.
Open finance: Open finance is the framework that lets a person authorize their financial data, such as account movements or payments, to be shared with another institution. The data travels with explicit consent through standardized channels.

Product

BNPL (Buy Now, Pay Later): BNPL (Buy Now, Pay Later) is short-term instalment financing decided at the point of checkout. Its challenge is approving credit in milliseconds with minimal friction, which requires a fast decision engine, alternative data sources and policies that can be tuned per merchant.
Embedded finance: Embedded finance means financial products offered inside platforms that are not banks: marketplaces, wallets, delivery apps. Credit appears at the moment of purchase or payout, without the person going to a financial institution.
Loan origination: Loan origination is the end-to-end process from a credit application being submitted to it being approved and disbursed: data capture, identity verification, data-source calls, scoring, policy execution and the final decision. Automating it cuts response times from days to seconds.
Subpolicy: A subpolicy is a decision policy invoked from another one, like a reusable block. It lets you split logic into pieces with a life of their own (identity validation, affordability calculation, fraud rules) and call them from several policies without duplicating them.
Test environment: An environment is an isolated space where policies are designed and tested without affecting real decisions. Separating testing from production lets you experiment with new rules at no risk to the portfolio or the applicant experience.

Regulation

Decision traceability: Decision traceability is the ability to reconstruct, for any given case, which policy was applied, on what data, which version was live and what outcome it produced. It is the baseline requirement for auditing, explaining and defending a credit decision to customers and regulators.
KYC and AML: KYC (know your customer) is the set of checks confirming a person is who they claim to be. AML (anti-money laundering) covers the controls to detect and report suspicious activity. Both are regulatory obligations prior to granting credit.
Model risk: Model risk is the exposure to poor decisions caused by flaws in the design, data or use of a scoring model or credit policy. It is mitigated through governance: independent validation, versioning, performance monitoring and full traceability of every change.

Risk

Affordability: Affordability is how much a person can put toward loan installments without compromising essential expenses. It is estimated from their income and existing obligations, and it defines the maximum reasonable amount to grant.
Alternative data: Alternative data are non-traditional information sources —transactional behavior, open banking, device data, telco or utilities— that complement the credit bureau to assess applicants with little or no history. They make it possible to lend to populations traditional scoring can't reach.
Champion/Challenger testing: Champion/Challenger is a technique for improving credit policies by routing a share of traffic to an alternative policy (the challenger) and comparing its performance against the one in production (the champion). It lets teams validate changes on real data at controlled risk before adopting them.
Credit scoring: Credit scoring is the technique of assigning an applicant a score that estimates their likelihood of repayment, combining internal data with external sources such as credit bureaus, behavioural data and alternative data. That score feeds the lending decision alongside policy rules.
Low and Grow: Low and Grow is a lending strategy that approves a small initial amount for customers with no credit history, observes their payment behavior and raises the amount gradually if they comply. It allows inclusion without exposing the portfolio up front.
Over-indebtedness: Over-indebtedness happens when the total installments a person must pay exceed what their income can sustain. It usually builds up gradually, adding loans across several institutions that look reasonable in isolation.
Payment shock: Payment shock happens when the installment a customer faces rises sharply against what they had been paying, due to a higher amount, a change of term or the end of a promotional period. It is a frequent cause of delinquency among customers who had been complying.
Roll rate: The roll rate is the percentage of accounts that move from one delinquency bucket to the next over a period (for example, from current to early delinquency, or from 30 to 60 days past due). It measures how fast a portfolio deteriorates, with more precision than the overall delinquency rate.
Shared consumer: A shared consumer is a customer who holds active credit with more than one type of institution at the same time (for example, a fintech and a traditional bank). Their true indebtedness is only visible by looking at the whole financial system, not a single institution in isolation.
Thin-file customer: A thin-file customer is someone with little or no information at the credit bureau: no prior financial products, or only very recent ones. It does not mean a bad payer, only a lack of evidence.
Vintage (cohort): A vintage groups loans originated in the same period to track their behavior over time. Comparing vintages shows whether lending policies are improving or deteriorating.

Technology

Batch processing: Batch processing evaluates many cases at once against a policy, instead of answering application by application. It is used to re-score a portfolio, build pre-approved campaigns and test a new policy on historical volume.
Decision engine: A decision engine is software that automates credit application assessment by applying business rules, scoring models and external data to return a decision — approve, decline or refer — in milliseconds. Unlike custom-built code, it lets the risk team configure and change policies without engineering involvement.
Decision tree: A decision tree is the visual representation of a credit policy as a sequence of conditions (if/then) leading to an outcome. In a no-code engine it is built by dragging blocks, which lets business and risk teams design and read the logic without writing code.
Policy versioning: Policy versioning is the practice of storing every change to a credit policy as an identifiable version — with author, date and reason — and being able to tell which one was live at any point in time. It makes policies comparable, revertible and auditable without ambiguity.
Query cache: A cache temporarily stores an external source's response to reuse it in later queries about the same applicant, instead of requesting it again. It cuts the cost per bureau query and the flow's response time.
Single sign-on (SSO): Single sign-on (SSO) lets people access the platform with the corporate credentials they already use, instead of a separate username and password. It centralizes onboarding, offboarding and access policies in the institution's directory.
Webhook: A webhook is an automatic notification the platform sends to a customer system when something happens, for example when a decision finishes. Instead of the system asking over and over whether there is news, it receives the result as soon as it is ready.