THE GUIDE · THE BUILDER SIDE · 12 MIN

How anyone launches a card.

Every fintech card — the neobank debit, the startup corporate card, the rewards app — is a stack of rented parts: a bank's licence and BIN, a processor's rails, and an app on top. Here is how it fits together, and who earns what.

SCROLL ↓
IN PLAIN WORDS — READ THIS FIRST

Every fintech card is three rented parts: a bank's license (the legal right to issue cards), a processor's rails (the plumbing that answers each tap), and the fintech's own app on top. The fintech earns interchange — a small fee from every swipe — and pays rent to the other two.

Two moments decide whether the whole thing works. The webhook: when you tap, the fintech's own code gets asked "yes or no?" and must answer in a fraction of a second. And the reconciliation: the app's ledger and the bank's records must agree on whose money is whose. Both show up below — working, and broken.

PART 01

The stack behind a fintech card.

Step through standing up a program, issuing a card by API, authorizing a swipe with just-in-time funding, and splitting the interchange.

PART 02

The parts you rent.

A fintech card is three rented layers plus your app. Here is each one.

THE ISSUING STACK

"Rent the bank, own the app."

A fintech card = sponsor bank (licence + BIN) + issuer-processor (auth rails) + program manager (the fintech's product and risk). Each owns a different piece of the licence, the funds and the liability.

BIN SPONSOR

"Whose licence is it, really?"

The card runs on the sponsor bank's BIN and regulatory licence. The bank is legally the issuer and carries the compliance obligation — which, after Synapse, banks are policing far more closely (see Field Notes).

PROGRAM MANAGER

"The brand you actually see."

The fintech is the program manager: it owns the app, the customer, the marketing and much of the risk decisioning. To you it looks like the bank; legally it is a partner riding a sponsor's rails.

ISSUER-PROCESSOR

"The engine under the card."

The processor authorizes, clears and settles each transaction and exposes the APIs. The modern, API-first generation — Marqeta, Lithic, Stripe Issuing, Galileo — is what made programmable cards possible, versus legacy cores like TSYS and FIS.

JUST-IN-TIME FUNDING

"Decide and fund at the swipe."

Instead of pre-loading a card, the program funds the exact amount at authorization via a real-time webhook. It lets software enforce balances, spend rules and fraud checks at the instant of the tap — the backbone of expense and corporate-card products.

INTERCHANGE AS REVENUE

"The swipe pays you."

For an issuer, interchange is income. A card program's P&L is roughly interchange in, minus rewards, fraud losses, and processor + sponsor fees. Get the mix right and the card funds the whole app — the neobank playbook.

The words, one at a time.

Six terms carry the builder side of issuing. These are the ones that appear in every partner-bank meeting.

FBO account
one big account, holding everyone's money
A pooled account at the sponsor bank — "For Benefit Of" customers — holding all customer funds together, while the program's ledger records whose is whose.
Your $500 balance in a neobank app is a ledger row pointing at a share of one pooled bank account.
Why it matters: if the ledger and the bank disagree, customers can't reach their money. Synapse made this real.
Reconciliation
checking the two sets of books agree
Comparing the program's ledger against the bank's actual balances — ideally daily, transaction by transaction.
The app's rows add up to $10.0M; the FBO account holds $9.7M. That $300K gap is now the only thing that matters.
Why it matters: post-Synapse, sponsor banks demand proof of this discipline before boarding a program.
Consent order
the regulator's stop sign
A public enforcement agreement between a bank regulator and a bank, usually freezing new programs and features until the named problems are fixed.
A wave of consent orders hit US sponsor banks through 2024–25. Every fintech on those banks waited.
Why it matters: your program can freeze because of your bank's problems, not yours.
Stand-in
the answer when your code doesn't pick up
Pre-agreed default rules the issuer-processor applies when the program's webhook misses its time budget — approve the small and familiar, decline the rest.
Your endpoint times out. The $6 coffee still approves under stand-in rules; the $900 laptop declines.
Why it matters: without stand-in rules, your outage becomes every cardholder's decline at the till.
Durbin-exempt
the small-bank interchange rate
US debit interchange is capped for banks over $10B in assets and uncapped below. Programs on small sponsor banks earn far more per swipe.
The same $40 debit swipe pays a program roughly twice as much through a small sponsor as through a big bank.
Why it matters: that gap funds "free" banking, and it's why the Durbin litigation matters existentially here.
Interchange split
who keeps the swipe fee
Interchange flows to the issuing side and gets divided by contract between the program, the sponsor bank, and the processor.
On a $100 exempt-debit swipe, the program keeps the biggest slice; the bank and processor take contracted cuts.
Why it matters: this split is a card program's entire revenue line. Everything else on the P&L is cost.
PART 03

The webhook that funds the swipe.

Just-in-time funding is the primitive under every modern card. This is the moment, in miniature.

// JIT FUNDING — the authorization stream

→ auth.request   { card: tok_9f2, amount: 40.00, mcc: 5814 }
   your code: balance ok? spend rules ok? not fraud?
← approve   { fund: 40.00, from: program_pool }
← or decline   { reason: over_limit }

// You checked a balance, enforced policy, and moved money — in ~200ms, at the instant of the tap.
PART 04

Why the model works.

$10B
the sponsor-bank asset line below which debit interchange is uncapped — the engine of neobank economics.
~200ms
to check balance, apply spend rules and fund the card at the instant of the tap (JIT).
API-first
Marqeta, Lithic, Stripe Issuing and Galileo turned card issuing into a few API calls.
Virtual
most programs mint a usable card number in seconds and ship plastic later.
WHEN IT BREAKS

When the stack fails.

A card program fails at one of three altitudes: the auth loop (milliseconds), the ledger (days), or the unit economics (quarters). One real failure from each, then a tree for finding the altitude fast.

FAILURE 01 · THE LEDGER BREAK
The app says $500. The bank can't confirm it.
WHAT YOU SEEThe app shows your balance, support says "we're working on it," and withdrawals stay frozen for months.
WHYWhen the BaaS firm Synapse collapsed in 2024, its ledger and the partner banks' FBO accounts didn't reconcile — roughly $265M froze, with a $65–95M gap nobody's records could explain. When the map disagrees with the vault, everything stops.
THE FIXFor builders: reconcile daily and let the bank see the ledger. For users: know which bank actually holds your money — FDIC insurance covers a bank failing, and only reaches you if the records say what's yours.
FAILURE 02 · THE WEBHOOK TIMEOUT
Your outage declines their coffee
WHAT YOU SEECards start declining at random. Balances are fine. The processor dashboard shows authorization timeouts climbing.
WHYJust-in-time funding puts your code inside the tap's ~200ms budget. A slow deploy, a database stall, or a traffic spike on your side turns into declined transactions at checkouts everywhere.
THE FIXAgree stand-in rules with your processor before you need them, keep auth handlers fast and idempotent, and treat authorization success rate as a pager metric rather than a weekly report.
FAILURE 03 · THE REWARDS MATH
Growth that loses money faster
WHAT YOU SEESignups surge, volume doubles, and the program loses more money every month it grows.
WHYThe rewards promised exceed the interchange actually earned on the real card mix. A program modeled on rich exempt-debit rates discovers its users spend on lower-rate transactions, while fraud losses and processor fees stay stubbornly fixed.
THE FIXUnderwrite the P&L per swipe before scaling: real interchange on the real mix, minus rewards, losses, and fees. If the unit math is negative, growth is just a louder way to lose money.
A CARD PROGRAM IS FAILING. AT WHICH ALTITUDE?
1 · Are taps declining while balances look right?
THE AUTH LOOP — MILLISECONDSCheck webhook latency, timeout counts, and stand-in logs first. The JIT loop is the usual suspect, and it fails one tap at a time.
AUTHS ARE FINE — KEEP GOINGMoney approved but stuck? Go to step 2.
2 · Do auths work, but money can't move in or out of the app?
THE LEDGER — DAYSA reconciliation break or a compliance freeze at the sponsor. Ask one question first: can the bank verify the ledger right now? If not, that's the whole problem.
MONEY MOVES FINE — KEEP GOINGGo to step 3.
3 · Does everything work, while the P&L bleeds?
THE ECONOMICS — QUARTERSRewards plus losses plus fees against real interchange on the real card mix. Growth multiplies whatever the per-swipe margin is — including a negative one.
NAME THE ALTITUDE FIRSTMilliseconds, days, or quarters. Each altitude has a different owner, a different dashboard, and a different fix.
COMMON QUESTIONS — ASKED PLAINLY

The things everyone wonders.

Five questions people ask about the card in their phone.

IS MY NEOBANK ACTUALLY A BANK?
Almost certainly not. Scroll to the bottom of the app's website and you'll find the tell: "banking services provided by [a bank you've never heard of]." The app is a program manager — it owns the product and the brand, while a sponsor bank holds the license and the money. This isn't a scam. It's the standard structure this whole chapter describes. But it means the question "is my money safe?" is really a question about the sponsor bank and its record-keeping, not about the app's design awards.
WHAT HAPPENS TO MY MONEY IF THE APP SHUTS DOWN?
In the clean version: nothing dramatic. The money sits at the sponsor bank in an FBO account, the ledger says what's yours, and you're paid out. The ugly version happened in 2024: Synapse failed, the ledgers didn't match the bank balances, and real people waited months for money that was "there" but unattributable. The lesson: FDIC insurance protects you if the bank fails. If the record-keeper fails, insurance never triggers — the bank didn't fail, and yet nobody can prove what's yours. Reconciliation, boring as it sounds, is the actual safety feature.
HOW DOES A NO-FEE CARD APP MAKE MONEY?
Mostly from the swipe itself. Every time you pay, the merchant's side gives up interchange, and on debit cards issued through small (Durbin-exempt) sponsor banks that fee is uncapped — roughly double what big banks earn. Multiply by millions of swipes and interchange funds the free account, the cashback, and the ad budget. Add interest earned on deposits parked at the partner bank and, for some apps, subscription tiers. You are not the product exactly — your payment habit is.
WHY DID MY CARD DECLINE WHEN I HAD MONEY IN THE ACCOUNT?
On a fintech card, "the bank said no" is usually the wrong mental model. More likely: the program's spend rules blocked the merchant category, the JIT webhook timed out and stand-in rules declined it, the program is under a temporary freeze, or a fraud rule fired. The tell is randomness — a decline with money present, at a normal shop, that works on retry, points at the plumbing rather than your balance. The decline decoder covers what the codes mean on the traditional side.
CAN ANYONE REALLY LAUNCH A CARD?
Anyone with capital, patience, and a business the banks want to underwrite — yes. The API platforms turned the technical side into weeks of work. The hard part moved: sponsor banks now underwrite you — your compliance program, your funding, your reconciliation discipline — far more carefully since Synapse. Expect months of diligence, ongoing audits, and real capital requirements. The stack is rentable; the accountability isn't.
FIELD NOTES — THE PRO LAYER

For the professionals.

The stack up close — the Durbin-exempt engine, sponsor-bank risk after Synapse, the modern processors, spend controls, and reading a program P&L.

DURBIN-EXEMPT — THE ENGINE OF FREE BANKING
Here's the trick behind 'free' neobank accounts: US debit interchange is capped for banks over $10B in assets, but exempt below it. So fintechs deliberately partner with small sponsor banks and earn roughly double the interchange a big bank could on the same swipe. That uncapped debit revenue — not fees, not lending — funds no-fee accounts, cashback and the customer-acquisition machine. It's also why the Durbin fight (see the card-types chapter) matters existentially to this model.
SPONSOR-BANK RISK AFTER SYNAPSE
In 2024 the BaaS middleware firm Synapse collapsed, and because its ledgers didn't reconcile with the partner banks, real customers couldn't reach their money. The fallout reset the industry: regulators made clear the sponsor bank is fully accountable for its programs' BSA/AML, KYC and reconciliation of the for-benefit-of (FBO) accounts. Banks now underwrite and monitor programs far more heavily. Building a card program in 2026 means proving you can keep the ledger honest.
THE MODERN ISSUER-PROCESSORS
The reason programmable cards exist: a generation of API-first issuer-processors — Marqeta (Square, DoorDash, many expense cards), Lithic/Privacy, Stripe Issuing, Galileo (Chime) — replaced batch-era cores (TSYS, FIS) with webhooks and real-time controls. Their key primitive is the authorization stream: subscribe to auth events and decide, per swipe, whether and how much to fund. Teen cards, fuel cards and fraud controls are all built on that hook.
SPEND CONTROLS & AUTH LOGIC
Because the program decides at auth, controls live in code: allow only certain MCCs, cap per-transaction or daily amounts, restrict to named merchants, require the app be open, block cross-border. Corporate-card and expense products are essentially rules engines on the authorization stream. It's also where real-time fraud defence sits — the same hook that funds the card can decline a suspicious one.
READING A CARD PROGRAM'S P&L
A program's economics: interchange plus any subscription/FX fees come in; out go rewards, fraud and credit losses, and processor + sponsor + network fees. Debit programs live or die on interchange volume; credit programs add interest income and underwriting risk. The classic failure mode is over-rewarding to grow, then discovering the interchange never covered the points — the same trap co-brand cards manage for a living.
PART 05

Remember three things.

1
A fintech card is a stack it rents: a sponsor bank's licence and BIN, an issuer-processor's rails, and the fintech's own app and risk logic on top.
2
Interchange is the revenue, not the cost. The business is interchange in, minus rewards, losses and fees — and on Durbin-exempt debit it's rich enough to fund 'free' banking.
3
Just-in-time funding puts software at the moment of the swipe. A webhook lets a program check balance, enforce rules and fund the exact amount in a fraction of a second — the foundation of every modern card product.
← PREVIOUS
FX: the invisible profit center