How card programs earn yield on customer float

Every card program sits on a balance it doesn't own yet. The question is whether that balance earns anything before it settles.

A card program earns yield on float by classifying customer balances into a liquid tranche that funds daily settlement and a deployable tranche that goes into a yield-bearing position, then sizing the liquid tranche to cover peak draw. When the balance is held as a stablecoin, the deployable portion can sit in an onchain vault and earn continuously, and the position can be exited at the moment of a transaction rather than on a redemption calendar.

That last part is where most float programs either work or fail. Earning yield on idle balances is not difficult; plenty of treasury products do it. Doing it without introducing a delay between a customer tapping a card and the transaction being approved is the actual engineering problem, and it's the reason card float has stayed in cash and Treasury bills at most issuers while the rest of the balance sheet got more sophisticated. The mechanics below are what changes that calculation.

Where float actually comes from in a card program

Float in a card program accumulates in the gap between when a transaction is authorised and when it settles with the network, typically 24 to 72 hours depending on the scheme, the merchant category, and the batch cycle. During that window the issuer has committed funds to a merchant but has not yet moved them, so the balance sits in an operating account doing nothing. Programs that require customers to prefund a wallet or card balance hold a second and usually much larger pool, because customers top up in round numbers and spend gradually. Visa's own work on stablecoins and the future of onchain finance is a useful read on why settlement timing is the variable that changes when the balance is tokenised.

The size of these pools surprises people who haven't looked. A program with $50 million in monthly spend and a two-day settlement cycle carries roughly $3 to $4 million in settlement float at any moment, and a prefunded program can easily hold two to three months of spend in customer balances. Card interchange on that same volume might run 80 to 150 basis points, so a few hundred basis points on the float pool is not a rounding error against the core unit economics. For many programs it's the difference between a negative and positive contribution margin per account. Gauntlet's Idle USDC Report puts numbers on how much stablecoin balance sits unproductive across the market, and the pattern it describes applies directly to issuers.

Three distinct pools are worth separating, because they behave differently and tolerate different amounts of illiquidity:

• Settlement float is committed and short-dated. It has a known outflow date, which makes it the easiest to model and the least tolerant of redemption delay.
• Customer prefunded balances are unpredictable per account but statistically stable in aggregate. Withdrawal behaviour follows patterns that hold up well across a large enough book.
• Program reserves and collateral posted to the network or the sponsor bank are locked, and in most jurisdictions carry restrictions on what they can be held in.

How does a card program earn yield on float without slowing approvals?

Approval speed depends on whether funds are available at the instant of authorisation, so any yield structure has to guarantee availability rather than promise eventual liquidity. The workable pattern holds a buffer in immediately spendable form, sized against the worst observed draw over a rolling window rather than the average, and deploys the remainder into a position that can be unwound atomically when the buffer is drawn down. Authorisation checks against the buffer, so the yield-bearing position never sits in the transaction path.

A common split looks like 20 to 30 percent operational float held liquid, 50 to 60 percent deployed into yield, and 10 to 20 percent held as a regulatory reserve in cash or Treasury bills, though the right numbers depend entirely on the spend profile. A program with lumpy corporate spend needs a much fatter buffer than a consumer program with thousands of small daily transactions, because the law of large numbers is doing more work in the second case.

Two failure modes show up repeatedly. The first is sizing the buffer against average daily draw instead of peak, which works until a payroll cycle or a marketing push moves spend three standard deviations and the program has to unwind a position under pressure. The second is treating a yield position with a redemption queue as though it were cash, which is fine until the day the queue matters and the program discovers it's structurally short.

What "atomic redemption" means for card settlement

Atomic redemption describes a structure where the yield-bearing position and the spendable balance settle in the same transaction, so a customer's tap draws directly against the position without a separate withdrawal step. The vault holds a liquidity buffer sized to expected draw, and a redemption is served from that buffer immediately rather than waiting for a strategy to unwind. From the customer's perspective the balance was always spendable, and from the program's perspective it was always earning.

Upshift is non-custodial vault infrastructure for onchain yield, used by fintechs, exchanges, wallets, neobanks, chains and asset managers to launch custom yield products. The atomic redemption vault is the vault type built for this specific problem. The Vault-as-a-Service model means a program launches a vault configured to its own risk parameters rather than depositing into someone else's pool, which matters when the buffer size and the whitelisted strategies need to reflect a particular spend profile.

Credit sits behind the buffer as a second line. Upshift's prime stack serves over $7 billion in monthly transaction volume and can structure a credit facility against a vault, so a program that draws past its buffer has somewhere to go while the underlying position unwinds and an unusual spend day doesn't force a fire sale. Those figures are as of July 2026.

Comparing the ways to hold card float

The structures below are the realistic options for a program deciding where float sits. Yields move constantly and none of these are fixed, so the figures describe typical ranges observed through mid-2026 rather than anything offered or promised. The onchain rows all settle against the EIP-4626 tokenised vault standard, which is what makes a receipt token transferable and composable rather than locked to one interface.

The weekend column is the one that catches card programs specifically. Card spend rises on Saturdays and Sundays while money market funds and most tokenised Treasury products settle on a business-day calendar, so a program holding float in either structure is running its highest draw against its least available liquidity. Onchain positions settle continuously, which lines the availability of funds up with the shape of the spend. For the lending row specifically, rates float with borrow utilisation, and both Morpho and Aave publish live market data worth checking before modelling a number into a business case.

The wider context is that stablecoin balances are becoming a normal part of how payments companies hold money rather than an experiment at the edges, a shift Bessemer traces in its work on stablecoins moving from DeFi primitive to financial infrastructure. Card programs sit early in that transition because their float is already short-dated and already denominated in something a vault can hold.

What does a program need before it can deploy float?

Deployment readiness comes down to four things, and a program missing any one of them will stall partway through an integration. Getting these settled before scoping saves a cycle.

1. A flow-of-funds map. Where each pool sits today, which entity holds it, what the peak draw has been over the last 12 months, and which balances carry restrictions from the sponsor bank or the network.
2. A custody decision. Whoever holds the vault's receipt token has to be able to hold it, which means the custodian needs to whitelist the vault. This is the step most often discovered late, and it puts the custodian on the deal map as a counterparty rather than a vendor.
3. A regulatory read on customer yield. Passing yield to customers as a savings rate or cashback has different treatment across jurisdictions, and e-money rules in particular constrain what can be done with balances held for customers. Programs that deploy customer float without disclosing the arrangement carry real exposure.
4. Risk parameters someone will actually own. Which protocols are whitelisted, what the buffer floor is, who can change it, and what happens on a drawdown. A vault with a policy engine enforcing restrictions at the chain, protocol, token and function level makes these answerable rather than aspirational.

What are the real risks of deploying float?

Smart contract risk is the one everyone names first and it's the tail risk, which is why audit history and the concentration of exposure per protocol matter more than headline yield. Upshift's contracts have completed 10 smart contract audits across 6 independent firms, and the full audit history is published by firm and date. Non-custodial architecture narrows the blast radius further, because funds only move between the vault contract and whitelisted strategy contracts rather than to any external wallet.

Liquidity risk is the more common failure in practice. A position that pays well and cannot be exited on the day it's needed has converted a treasury problem into an operational one, which is the entire reason the buffer exists and the reason it should be sized against observed peaks. Counterparty risk applies wherever the strategy touches a lending venue or a centralised borrower, and the honest mitigation is exposure limits and transparency on where the capital sits rather than a claim that the risk isn't there. Rate risk cuts both ways: variable yields fall as well as rise, so a program modelling float income into its unit economics should model the downside case too.

Regulatory risk deserves its own line for card programs specifically, because the treatment of yield on customer balances is moving in several jurisdictions at once. A program should expect to disclose how float is deployed, what yield is generated, and how it's allocated between the customer and the issuer. Wharton's Stablecoin Toolkit is a reasonable neutral primer on the financial and market dimensions if the compliance team wants something that isn't written by a vendor.

Always make sure to do your own research and be aware of the above and any other risks before depositing.

Frequently asked questions

How much float does a typical card program hold?

Settlement float generally runs one to three days of spend, so a program with $50 million in monthly volume carries roughly $3 to $4 million at any moment. Programs that require customers to prefund hold considerably more, often two to three months of spend, because customers top up in round amounts and spend gradually.

Can float earn yield without delaying card authorisations?

Yes, as long as authorisation checks against a liquid buffer rather than against the yield-bearing position. The buffer is sized to cover peak draw and the yield position sits outside the transaction path, so approval speed is unchanged. Structures that require a withdrawal step before funds become spendable will add delay.

What is atomic redemption?

Atomic redemption is a structure where the yield-bearing position and the spendable balance settle in the same transaction, so a card tap draws directly against the position. Redemptions are served from a liquidity buffer held inside the vault, which means no waiting for a strategy to unwind.

Is yield on card float passed to the customer or kept by the issuer?

Both models exist. Some programs retain the full spread as revenue against interchange, and others pass a portion to customers as a savings rate or cashback to drive acquisition. The split is usually a regulatory question as much as a commercial one, since e-money rules in several jurisdictions constrain what can be done with balances held on behalf of customers.

What happens if spend spikes above the buffer?

A well-structured program has two lines of defence. The buffer absorbs normal variance, and a credit facility covers the gap while the underlying position unwinds, so an unusual day doesn't force an exit at a bad price. Programs without the second line should size the buffer considerably more conservatively.

Do stablecoin balances change the picture versus fiat float?

They change the settlement calendar, which is the constraint that matters most for cards. Fiat float held in money market funds settles on business days while card spend peaks at weekends, so the availability of funds and the shape of the draw don't line up. Onchain positions settle continuously.

Which custodian needs to be involved?

Whichever one holds the program's assets, because the vault's receipt token has to be a supported asset on their platform. Custodian whitelisting is a gating step rather than a formality, and it's worth raising in the first scoping conversation instead of discovering it during integration.

Keep reading

• Best crypto cards that earn real yield on your balance covers the same mechanics from the cardholder's side, which is useful context for how the product is marketed.
• Why every financial product will run on vaults makes the structural argument for why balance-holding products converge on this pattern.
• How the GENIUS Act and CLARITY Act reshape stablecoin yield is the regulatory backdrop for anything involving yield on customer balances.
• Upshift's risk management framework documents the buffer, the NAV volatility caps and the whitelisting model in technical detail.
• What are DeFi yield vaults is the ground-up explainer if vaults are new to the reader.

Keep reading

• What is an onchain vault: what a vault is, who holds the assets, and how money moves in and out.
• The DeFi yield strategies onchain vaults run: the strategy shapes, the risk in each, and measured returns.
• How to use an onchain vault in a treasury product: the ETP comparison, fees, and what to decide first.