# How a single transaction becomes bankable data
You tap your card at a coffee shop. The payment clears in under two seconds. In that blink, a chain of systems has captured, checked, enriched, and routed your transaction across at least four institutions. By the time you pocket your card, the event exists as structured data in multiple databases, each version serving a different purpose.
That single swipe is the atomic unit of banking data. Understand its journey and you understand where every fraud model, spending report, and regulatory filing gets its raw material.
A card payment happens twice. This surprises most people.
Authorization is the real-time check. When you tap, the merchant's system asks your bank a question: "Is this card valid, and is there enough money or credit?" Your bank answers yes or no in milliseconds. No money moves yet. A hold is placed.
Settlement is the actual movement of money, usually hours or a day later, in batches. This is when funds leave your account and reachreachThe number of unique people exposed to your message in a given period. Unlike impressions, reach counts each person once, no matter how often they see it.Voir la définition complète → the merchant.
Why split them? Speed and reliability. The merchant needs an instant yes to hand you the coffee. Moving real money instantly across banks is slower and riskier, so it happens later in bulk.
For data teams, this matters enormously. The authorization record and the settlement record are two separate data events describing the same purchase. Reconciling them (matching auth to settlement) is a core banking data problem, because they can differ. A $50 authorization at a gas pump might settle at $47.32.
A typical card transaction involves four players, each generating and storing their own copy of the data.
The card network defines the data format everyone must speak. For decades that format has been based on the ISO 8583 message standard, a compact, field-based structure for financial transaction messages. The industry is migrating toward the richer ISO 20022 standard, which carries far more structured detail (see the ISO 20022 resource centre).
An authorization message is not just "amount and card number." It is a packed set of fields. A simplified view of the core data captured at tap:
{
"pan_token": "tok_4f9a...c21", // tokenized card number, not the real PAN
"amount": 5.40,
"currency": "USD",
"merchant_id": "MID_00874512",
"merchant_category_code": "5814", // MCC: fast food restaurants
"terminal_id": "TERM_39",
"timestamp": "2026-03-14T08:12:44Z",
"auth_response": "00", // 00 = approved
"entry_mode": "contactless"
}A few of these fields do heavy lifting downstream:
The transaction is born structured. That is the key insight. Unlike a customer email or a call transcript, a card swipe arrives as clean, typed fields ready for a database.
Raw fields are terse. A merchant name might arrive as "SQ *BLUE BOTTLE 4471" with no location. Enrichment is the step where the bank adds context.
Common enrichment steps:
Enriched data is what you see in a mobile banking app, with a logo, a clean name, and a category. That polish is the product of a pipelinepipelineAll active sales opportunities across the stages of the sales process, together with their combined potential value and probability of closing.Voir la définition complète →, not the raw swipe.
🎬 [VIDEO: "How Credit Card Processing Works" — youtube.com — a clear walkthrough of the authorization and settlement flow between issuer, acquirer, and network]
The transaction now lives in several places, each tuned for a different job.
The core banking system is the bank's system of record for accounts and balances. It holds the authoritative version: your balance changed, this posting is final. Core systems prioritize accuracy and durability over analytics. You do not run marketing queries against the core.
A near-real-time copy feeds fraud detection and the mobile app. Fraud models need transactions within seconds, so this layer is optimized for fast lookups and streaming.
For analytics, transactions are copied (usually in nightly or streaming batches) into a data warehousedata warehouseA central repository that consolidates data from many source systems into a structured, query-optimized store designed for analytics, reporting, and business intelligence.Voir la définition complète →, a system built for large-scale querying and reporting. Increasingly banks use a lakehouselakehouseA hybrid architecture combining the flexibility of a data lake with the analytical capabilities of a data warehouse, on a single storage layer.Voir la définition complète →, which combines a low-cost data lakedata lakeA data lake is a centralized repository that stores large volumes of raw data in its native format, from structured tables to unstructured files, until needed.Voir la définition complète → (raw storage for any format) with warehouse-style query power.
Here the transaction joins years of history. It becomes one row among billions, feeding:
Regulators expect banks to explain where a number came from. Data lineageData lineageData lineage maps how data moves and transforms across systems, from origin to consumption, showing where it came from, what changed it, and where it goes.Voir la définition complète → is the documented trail showing how a figure in a report traces back to source transactions. After the financial crisis, the Basel Committee issued principles known as BCBS 239BCBS 239Principe du Basel Committee on Banking Supervision imposant aux grandes banques une traçabilité stricte des données de risque, ayant catalysé la création de nombreux postes de CDO dans le secteur bancaire. on risk data aggregation, pushing banks to prove their data is accurate and traceable. That is why a single swipe carries metadatametadataDonnées sur les données, informations décrivant le contexte, la structure, la provenance et les caractéristiques d'un asset de données (auteur, date, format, source, définition). about its origin all the way through the pipelinepipelineAll active sales opportunities across the stages of the sales process, together with their combined potential value and probability of closing.Voir la définition complète →.
Vérification des acquis
1. Why do card payment systems separate authorization from settlement rather than moving money instantly?
2. A gas pump authorizes $50 but the transaction later settles at $47.32. What core banking data challenge does this illustrate?
3. During the authorization step of a card payment, what actually happens?
4. Select ALL correct answers about why a single transaction produces multiple versions of data across institutions.
Sélectionnez toutes les réponses correctes.
5. Select ALL correct answers describing roles in the four-party card model.
Sélectionnez toutes les réponses correctes.
A recurring headache: two reports show different totals for "yesterday's spend." Usually neither is wrong. They are measuring different versions of the same transaction.
Add timing. A transaction authorized at 11:58pm might settle the next day, so it lands in a different reporting period. Cross-border transactions add currency conversion timing.
This is why banking data teams obsess over definitions. "Number of transactions" is meaningless until you specify: authorized or settled? Approved only or including declines? Gross or net of reversals? A good data model captures every state a transaction passes through, so any question can be answered consistently.
Tap, authorize, enrich, settle, post, reconcile, warehouse, model. Each step writes or updates data, and each downstream system reads a version suited to its purpose. The coffee you bought is now a permanent, queryable, governed record that will help score the next transaction for fraud and feed a report a regulator reads next quarter.