Skip to content

FinTech glossary

30 terms we use on this site, defined the way we mean them. Every entry links to the work behind it rather than to another definition.

Payments and money movement

The vocabulary of the money path: what carries a payment, what proves it happened.

Tokenization
Replacing a card number with an opaque reference that means nothing outside the system that issued it. The card details live with the processor or a vault, and your systems move the token instead, which is what keeps the card data out of your estate in the first place.Payments & Money Movement
PCI DSS SAQ A
The lightest PCI DSS compliance tier, available when card data never touches your servers. Hosted fields put the card form in an iframe the processor controls, so the page collects nothing sensitive and the annual assessment stays a questionnaire rather than a full audit.Payment Gateway
Payment orchestration
Routing each transaction to a processor chosen at run time by geography, currency, method or health, with automatic failover when one degrades. The point is that a bad night at one provider costs a retry rather than a night of declines.Money movement
Double-entry ledger
A book where every movement is recorded twice, as a debit and a matching credit, so the totals prove themselves. A balance is the replay of its entries rather than a number someone updated, which is why a trial balance can demonstrate the books rather than assert them.Payments & Money Movement
Idempotency key
A caller-supplied identifier that makes a repeated request produce the same single result. Networks retry, users double-click, and queues redeliver, so without one a retry becomes a second charge.Payment Gateway
3-D Secure 2 and SCA
The authentication step that shifts chargeback liability to the issuer, and the European rule that requires it. Because the rule also lists exemptions, an exemption engine decides per transaction whether friction is required at all rather than applying it to everyone.Cards & wallets
Account-to-account (A2A)
Payment straight from one bank account to another, over local instant rails instead of card networks. Cheaper per transaction and irreversible in a way cards are not, which changes how refunds and disputes have to be designed.Money movement
Reconciliation break
A line that appears in one record and not in another: the processor file, the bank statement and your ledger disagreeing. Automated matching resolves the mechanical majority; a break is what is left for a person, and the useful measure is how few of them reach one.Payments & Money Movement

Lending and credit

What a lender has to be able to show, months after the decision was made.

Automated underwriting
Running an application through the same checks in the same order every time, so a clean file clears without a queue. It only works in regulated credit when the outcome arrives with the reasons attached, which is what separates it from a score nobody can interrogate.Lending & Credit
Adverse action notice
The explanation a US lender owes an applicant it declines, under ECOA and Regulation B. It has to state the actual principal reasons, so the decision path has to record its inputs and reason codes at the moment it runs rather than reconstruct them later.Lending & credit
Alternative-data scoring
Scoring cash flow, rent, payroll or telecom history alongside bureau data. It is how a thin-file or self-employed borrower stops looking riskier than they are, and it only holds up under fair-lending review when each input is documented and governed.Lending & Credit
Model drift
A model growing less accurate as the world it was trained on moves away from it. Undetected, it degrades quietly rather than failing, so monitoring is part of the system rather than a review someone remembers to schedule.AI in Finance

Fraud, identity and compliance

The checks that decide who gets in, and the record that proves the decision.

KYC and KYB
Verifying who a customer is, and for a business, who ultimately owns and controls it. Neither is a one-off: the obligation continues after onboarding, so the design question is what triggers a re-check rather than what happens at signup.KYC & KYB digital onboarding
Sanctions and PEP screening
Matching parties against sanctions lists and politically exposed person data. Name matching produces false positives by nature, so a review queue is part of the control: hard-blocking on a common surname loses good customers to a spelling.Fraud, Identity & KYC/AML
Transaction monitoring
Scoring payments as they happen rather than reviewing them in a batch afterwards. Behavioural and device signals travel across channels, and each decision is written to a record that survives to the point a report is filed.Fraud detection
Account takeover (ATO)
A real account used by someone who is not its owner. Credentials alone cannot detect it, because the credentials are correct; what gives it away is behaviour that stops matching the person the account belongs to.Fraud, Identity & KYC/AML

AI in finance

Terms worth being precise about, because the loose versions are how these projects lose credibility.

Retrieval-augmented generation (RAG)
Answering from documents fetched at query time instead of from what a model absorbed in training. The knowledge stays governed and updatable, and an answer can cite the document it came from, which is the part that matters in a regulated process.AI in finance
Confidence routing
Scoring how sure the model is per field and sending the uncertain ones to a person. It is what keeps extraction usable at volume: the clear majority passes through, and the values worth doubting reach a human instead of a decision.Finance process automation
Model governance
The record of what a model is built on, what it has seen, what it returned, and who reviewed it. A regulator asks for this, and assembling it after deployment is considerably harder than writing it as you go.AI in Finance
Evaluation set
A fixed set of cases with known correct answers, run before deployment and after every change. Without one, quality is a matter of impression and a regression is invisible until someone downstream notices.Applied & Generative AI
Pilot accuracy
A figure measured on a controlled sample during a trial. It is a demonstration, not a production average, and presenting it as one is the most common way these projects lose credibility.AI in Finance

Platform and delivery

How a system gets replaced without the business going dark.

Strangler fig
Putting a routing layer in front of a legacy system and moving traffic to its replacement slice by slice. The old path stays live until the new one has earned the traffic, so there is a rollback point at every step instead of one weekend.Legacy Application Modernization
Ports and adapters
Keeping business rules free of the things they talk to, with processors, storage and rails behind interfaces. Adding an acquirer or a market becomes an adapter rather than a change to checkout, and the rules stay testable on their own.FinTech Product Architecture & Engineering
Characterisation test
A test that captures what a system does today, including behaviour nobody intended. Legacy code is under-documented and over-relied-on, so this is what a refactor gets to be correct against.Legacy Application Modernization
Data contract
A schema a producer agrees to publish against, enforced in the pipeline. Breaking changes then fail in CI rather than in a dashboard on Monday morning.Financial Data Engineering & Analytics
Banking as a Service (BaaS)
A licensed institution providing the regulated pieces (accounts, cards, money movement) through an API. It is easy to start on and hard to unwind, so the decisions worth making early are where the money is recorded and how replaceable the provider is.Embedded Finance
Open banking APIs
Standardised, consent-based access to account data and payment initiation at other institutions. The standard differs by jurisdiction, which is why the connectivity belongs behind one interface rather than one integration per provider.APIs & open banking

Billing and revenue

Where invoicing, collection and recognition stop agreeing with each other.

Dunning
The retry and notification sequence after a payment fails. A hard decline and a temporary one need different behaviour, so a fixed schedule that ignores the reason code churns customers who would have paid.Billing, Subscriptions & Revenue
Revenue recognition (ASC 606)
Recording revenue when the obligation to the customer is met, which is not the same moment as the invoice. Deriving it from contract events rather than from invoices is what keeps the books defensible when the two diverge.Billing & subscriptions
Metered billing
Charging on measured usage rather than a flat plan. Usage has to arrive as an event stream rather than a nightly total, because a customer asking why a bill changed is asking to see the events behind it.Billing, Subscriptions & Revenue

Tell us where
it breaks

Bring the part of your fintech stack that is hard right now. We start with a working consultation and scope to outcomes.