The Payroll Float Sandwich
Who actually keeps the money between Tuesday and Friday?
Hey all!
I’m coming at you with some updates, and a technical piece around the Payroll Float Sandwich and Stablecoins. I’ve been thinking about this one for a while, and it took perhaps the greater part of all my running volume this past month to really cement the thought.
Warning: This is straight up technical, good luck everyone.
What I’m Up To:
Almost all my extra time goes to endurance sports!
May 2026 marks the first month I hit 160 miles of running, this includes an international trip! I don’t think I can hit this again as May was unique in that there were two races - a Half Marathon (fastest I’ve done that) and….my first Ultra Trail Run 50k (~32 miles). I had never race on trails, and only started vertical training April 2026 on treadmill, and actual trail training in May 2026.
The Payroll Float Sandwich
Who actually keeps the money between Tuesday and Friday?
TL;DR: Payroll companies are software wrappers for float income (money earned on your money)
My favorite company on payday, ADP, earned $1.19 billion on $37.6 billion of average client funds in fiscal 2025. My paycheck contributed to my favorite company’s earnings.
This $1.19 billion is float income, interest on money that legally belongs to the employee but sits in a trust between the moment the employer initiates payroll and the moment the employee can spend it. This timing gap, is what we will call the dwell, and the the income earned on the dwell, we will call float; yield that accrues to whoever’s balance sheet the in-transit money touches. At scale, the interest becomes a significant positive financial line item. It isn’t a fee, nor was it a service charge that the employee or employer had to be burdened with.
It was the yield on someone else’s wages. Like Venmo, but, prehistoric.
The $1.19 billion is 5.8% of ADP’s revenue. And yet, it is roughly 22% of the company’s operating profit. The gap between the two is the margin - the ones companies like: running a Treasury portfolio costs almost nothing relative to running payroll software, so the float income drive the operating profit, increasing profitability disproportionately at the operating profit level.
The payroll industry is often branded as software. If you look closely, it then looks like a service provider. And if you look really closely, economically, it is float income. The current traditional fiat & bank structure support this model because they are in on it too.
Naturally, the question to ask these days is where do Stablecoins fit in?
Stablecoins do not eliminate the float. On the contrary, stablecoins offer a path to reorganize who holds it (the float), on what custody terms, and inside which regulatory ceiling. Stablecoins are an act of reorganization, not replacement, because the float dollars do not vanish; they change who the custodians and counterparties are and can be.
Let me introduce you to two mental models to make this more palatable:
The first is how to measure the prize (the market): start from the wage base, not from a payroll-software total addressable market chart. The US private-sector W-2 wage base runs about $11.7 trillion, and the float that accrues on the dwell across that base sits between
Narrow range: $4 billion narrow (third-party processors only)
More broader range: $15 to $20 billion (including bank-side margin on either end).
The second is how to read the regulatory arc.
Bridge (it’s a company) holds the OCC’s conditional trust-bank charter as of February 2026, the first federal recognition of a stablecoin custodian as a depository institution.
HSBC was licensed under Hong Kong’s new stablecoin framework on April 10, 2026, a global bank issuing a regulated stablecoin under a fresh regime.
SAB 122 took effect for fiscal years starting January 2025, restoring banks’ ability to custody crypto off balance sheet.
Several native stablecoins launch, including USDsui, which debuted March 4, 2026, routing reserve yield to the network rather than the issuer; that in itself is very different from how other stablecoins have done it. USDC and USDT route yields to the issuer.
Each step pulls float deeper inside the depository system rather than away from it. With who, and where the float actually sit, are the next questions.
The processor takes one slice…the banks take the rest.
TL;DR: Payroll companies and banks make money on your money, before and after it reaches you.
The common version of this story has ADP orchestrating the movement and the banks clipping the yield. That is half right. ADP captures float. It only captures the days-of-dwell slice, not the weeks-of-dwell slice.
ADP’s “interest on funds held for clients” is a line item, disclosed, audited, and ever growing. $1.19 billion in fiscal 2025, on $37.6 billion of average client funds, projected to be $1.29 to $1.31 billion for the next year. Behind this mundane line item is a treasury portfolio: corporate bonds 51%, US Treasuries 25%, Canadian government 9%, asset-backed and agency the rest. AAA/AA across the portfolio. These bonds are laddered (bonds with staggered maturities, a portion maturing each year), up to five years on the extended tier and ten years on the long tier. So ADP is investing long and borrowing short. Functionally speaking, ADP is an investment portfolio embedded inside a payroll-software wrapper, and the wrapper trades on a SaaS multiple. Don’t laugh, that’s MSTR.
The dwell (days) ADP captures is short. Money sits in a trust between the employer-debit moment and the employee-credit moment, and ADP earns yield on those days. Call it the processor’s slice.
The longer dwell sits on either end.
The employer’s operating bank holds the money in the days between when payroll cuts and when the ACH debit clears. The employee’s bank holds the money from credit through whenever the wage actually gets spent or moved to savings. Those windows are weeks, not days. The banks on either end of the sandwich do not break out “payroll float” as a separate line because they do not need to. It is rolled inside net interest margin (NIM, the spread a bank earns between what it pays on deposits and what it earns on investments) and inside basic checking-account economics. No 10-K names JPMorgan, BNY, State Street, or Bank of America as float recipients. They are the silent partners of the sandwich, and they capture the larger slice.
A bottom-up build sizes the magnitude. BEA reports a $13.2 trillion total US wage base at the end of 2025, with roughly $11.7 trillion as the private-sector W-2 line. Run that through observed dwell times, the share processed by third-party providers, and a blended yield in the low single digits, and the industry float interest comes in around $4 billion narrow. Widen to include downstream dwell in employee checking accounts and bank-side NIM and the number runs $15 to $20 billion.
Disclosed float income across the Big Five public payroll companies (ADP, Paychex, Paycom, Paylocity, Dayforce) sums to about $1.8 billion. The full third-party industry sits closer to $3.2 billion. The remainder of the narrow bracket is the long tail of regional providers and PEOs whose float disclosure is buried inside other lines.
So the float fragments. The processor captures one slice. The corporate bank captures another. The employee’s bank captures a third. Each earns NIM on the same dollar at a different point in its journey. Fun fact, Circle pays Coinbase $908 million to manufacturer fragmentation in an effort to get USDC to spread across platforms so that the reserve income pools in multiples place, and Coinbase can skim it. I’d call it the revenue model.
Why hasn’t this collapsed already?
TL;DR: This is so boring, invisible, and regulatory intensive that no one bothered to try.
Truthfully, don’t fix what is not broken, and also, who want’s to fix the least-sexiest thing in the financial market?
There are five reasons though.
Regulatory. The federal stack (the GENIUS Act on issuance, CLARITY on tightening yield definitions, SAB 122 on bank custody effective fiscal 2025 onward) no longer blocks the depository side of the path.
None of it helps a payroll provider trying to use a stablecoin as a qualified holding for trust-account purposes.
State money-transmitter law requires FDIC-insured trust accounts for payroll funds, and most state statutes do not contemplate stablecoin custody as a qualified holding.
No US state has yet approved stablecoin-denominated custody for payroll trust accounts. Whichever state moves first becomes the launching pad, and probably the regulatory model the others copy. I see Federal as the ceiling, and the State is the operating floor.
Historical. ACH was designed in 1974, on the constraints of mainframe batch processing and overnight settlement, and the two-bank payroll plumbing inherited that architecture. Half a century later, the rails still settle on the same overnight cadence, and the entire installed base of US employers runs on top of it.
Thus, path dependence is a real moat, and the energy required to “change it” is immense. It is also incredibly boring, which is probably why it survives. The boring systems that survive are the ones nobody bothers to think of, had a problem with, or rebuild because they keep working. Like, are you going to see another WM Waste Management Company servicing your domicile any time soon?
Risk separation. The bank takes credit risk and lives under the OCC and FDIC supervisory regime. In between, the processor takes operational risk and lives under fifty different state money-transmitter regimes. Combining them means combining capital requirements, examiner cadences, and liability surfaces, which is why most institutions choose to specialize. Because of regulation, it becomes a clear:
Be a processor or be a bank.
Float tends to fragment across custodians because liability fragments across them, and fragmented liability is more appealing as it reduces overall surface area - consider it reverse synergy if you’re goal is to not be noticed.
Customer choice. Employers select payroll vendors on feature set, compliance posture, HRIS integrations, and the price of the SaaS line. The procurement RFP does not ask “how much of my employees’ wage yield are you keeping during the dwell.” It asks about uptime, tax filings, record keeping, and how far in advance do I need to fund the payroll for me to pay employees.
Principal-agent gap. The party who pays the cost of the float, the employee, has no say in vendor selection. The party who picks the vendor, the employer, does not bear the cost. The cost-bearer and decision-maker are different parties, and the rev-share between them is invisible to both. That is what keeps the sandwich stable despite the magnitudes on the table.
The first four reasons presented here make a structure persistent. The fifth makes it specifically un-shoppable. A market does not optimize a cost the buyer cannot see and the payer cannot vote on.
Stablecoins do not directly change reasons one through four. They directly change reason five, which can then lead to influencing 1 to 4. They make the cost visible at the point of receipt. The employee can see the money the moment it hits the wallet, and the dormancy spread becomes a choice rather than a tax.
What changes when the float lives onchain?
TL;DR: Nothing, but as the “changes” happen, we see an opportunity for new players to shuffle in, and old players get shuffled out.
Stablecoins do one thing ACH cannot. They can consolidate the float into a single custodian and make receipt legible at an onchain level. The principal-agent gap closes off because the employee can see the dwell (and the rest of public market looking at the chain).
That consolidation forces a second-order question. If the float lives in one place, who captures the yield on it?
To better understand where we could be going onchain, I am going to introduce the concept of Rent or Recycle.
Rent: The issuer rents the float to a distributor that owns the user surface. All current payroll companies, in the traditional sense, fall into this category.
Recycle: The issuer recycles the float into another reorganization. In blockchain terms, it means there is an inflow back to the protocol network the stablecoin runs on. We haven’t seen this yet, until USD Sui.
Let’s examine these mental models when applied to the payroll industry.
How is the Payroll Industry…Renting?
TL;DR: Payroll companies often rent their float (they are using your money to make money), but take different strategies at preserving this mechanism.
Dayforce (used to be Ceridian) reported $200 million of float in fiscal 2024 on $1.76 billion of revenue:
11.4% of the topline, roughly 192% of GAAP operating profit.
Strip the float and fiscal 2024 operating profit was negative $96 million.
You can tell it’s a rent because if you strip float, the operating profit is negative, which is really like a Landlord who forgot to build anything else and got screwed when their only tenant didn’t pay rent on time.
The market noticed, and Thoma Bravo (private equity) took the company private at $70 a share in August 2025. The float was structurally significant, the rest was overhead, and once rates turned, a private buyer was willing to underwrite the structural slice rather than wait for the public multiple to repair.
Paycom is another live indicator, though it’s priced as software.
$113 million of float on $2.05 billion of revenue, 5.5% of the topline, guided lower into calendar 2026.
The company is publicly forecasting its highest-margin revenue line to shrink and the equity continues to trade on a SaaS multiple. That is the cleanest live mispricing in the public comparables, given that we shouldn’t be pricing on SaaS multiples on companies like this. It also means PE didn’t notice it yet.
Paylocity captures the rate beta on the upside, even though it’s priced as software.
Float grew 24.9 times over three years, and roughly 54% of the three-year operating-income expansion was rate-driven.
Between Paycom and Paylocity, that is two companies, two directions, both priced as software.
Paychex, another renter, shows what the incumbents look like when the tailwind ends.
The company raised $4.2 billion of fresh debt above 5% to fund the Paycor acquisition while its own float yields about 3.4%.
Negative carry, a bet against a falling yield environment.
It’s totally exposed to falling rates while above market debt service, so the float income on this isn’t even saving them due to poor financing. It’s not at all dissimilar to how SVB failed in 2023. P
Now let me remind you of ADP, another renter
$1.19 Billion is 5.8% of ADP’s revenue, but it has a 22% company operating profit. It maintains, and increases its margin, by running a laddered treasury portfolio. This portfolio has nothing to do with their payroll software, it’s just them buying bonds at different maturities and risk levels.
In response to the great “rate cycle in the sky”, ADP responded with laddered duration (Like MSTR). Dayforce responded by being taken private. Paychex responded with leverage against debt. Three different incumbent reactions to the same rate cycle, none of which changes the underlying capture, or that float. The float “isn’t getting better”, the overhead around capturing the float just changes.
That is the reorganization happening in the payroll side, in which companies are chasing the float. There aren’t new dollars. Same dollars, captured by a different counterparty, with a different objective function. Objectively speaking, while each company itself is a permutation of “Rent”, it’s quite hilarious to see each company reorganize…and do different things to preserve that rental float income.
Of course there is also the option of not playing.
Workday is…not a renting example, but not a recycle example either. It’s an abstention
Fiscal 2025 revenue $8.45 billion, customer-cash float zero by design, 4.9% operating margin. Quite a thin operating margin there. This is a consequence of Workday not participating in float, and going in as a large software company.
The company earns roughly $350 million on its own corporate cash, but the income flows through as non-operating: it boosts net income without flattering the operating margin, and the equity market values it accordingly.
So where do stablecoins play in this?
TL;DR: A new path to break the rent cycle is possible, but is it going to be profitable?
As a reminder, this is what I will define rent as: The issuer rents the float to a distributor that owns the user surface. All current payroll companies, in the traditional sense, fall into this category. A hard example today is:
Circle and Coinbase: $908 million of $1.68 billion routed to Coinbase in 2024, roughly 54% of reserve revenue, under the distribution agreement. To be clear, it means the issuer, Circle, who is taking the regulatory risk and getting the reserve yields, has to then pay Coinbase 54% for Coinbase to serve as the primary distributor. Big ouch.
Where stablecoins offer a new pathing is Recycle: The issuer recycles the float into another reorganization. In blockchain terms, it means there is an inflow back to the protocol network the stablecoin runs on.
When it comes to “Recycle”, we haven’t seen this permutation yet, until maybe now. USDsui’s redirect is the structural opposite of that decision. Wherein a lot of stablecoin issuers today follow the RENT Cycle (USDC and USDT). Uniquely, we are seeing an emergent of a possible RECYCLE through USDsui. USDsui’s design routes reserve yield to the network, incentivizing SUI token repurchases and DeFi incentives.
I suppose the equivalent of this in traditional capital market is a company repurchasing it’s stock, for the benefit of it the ecosystem and employees. Just kidding. It’s hard to imagine a “recycle” version that isn’t tied to a co-op, but even then, co-ops don’t scale.
Everyone’s banning stablecoins…yields.
TL;DR: Banks really don’t want to change this.
What is very important these days is the level of regulatory banning on who gets the yield, and for how long you get the yield, should you go through a stablecoin. Yet the same rigor is not applied to the plumbing that supports the existing financial system.
Stablecoins present 4 disadvantages:
The duration moat. ADP runs a laddered portfolio out to five years extended and ten years long, with leverage applied through short-term borrowings. A fully-reserved stablecoin custodian cannot do that by itself.
The GENIUS Act effectively constrains reserves to short-dated Treasuries and equivalent, and 100% coverage prevents the long-and-short trade.
The duration premium ADP earns on the spread between funding cost and portfolio yield is not available to a fully-reserved issuer. Per dollar of float, the issuer earns less.
The yield ban. GENIUS prohibits direct issuer-to-holder yield. CLARITY tightens to “economically or functionally equivalent.” MiCA bans any equivalent yield. The HK Stablecoins Ordinance does the same. The Japan PSA, the Singapore MAS, the UAE framework, all converge.
A holder of a regulated stablecoin cannot earn yield from the issuer. Whatever the float earns, the holder does not see directly.
The structural yield gap. Circle’s effective reserve yield ran about 3.15% in 2024. ADP’s blended portfolio yield ran about 3.2% in fiscal 2025.
The fully-reserved issuer is already structurally below the laddered processor on per-dollar yield (i.e. the reserve issuer underperform when compared to ADP’s investment portfolio), and that’s before the distribution share is paid to the wallet.
The economics seem to get worse from there.
Tax remittance lock-in. The IRS does not accept crypto via EFTPS. State tax authorities do not accept stablecoin remittance for income-tax withholding. Federal Form 941 schedules are fiat by construction. State income-tax withholding aggregates to about $475 billion a year. Federal income-tax withholding runs $1.8 trillion. FICA runs $1.55 trillion.
The arithmetic is mechanical: Form 941 plus EFTPS, applied to every withholding line on every paycheck, locks roughly 60 to 75% of the payroll-cycle’s NPV (net present value, the discounted future value of the float dollars) to fiat by statute, regardless of how fast the on-chain rail settles.
Even a fully stablecoin-native employer has to deliver fiat for the tax slice. The remittance rail is the binding constraint but that is a tactical issue for now.
Where the new capturable positions sit
TL;DR: Welcome to the great reshuffle.
Stablecoins do not eat payroll.
Really, it reshuffles it and offers new opportunities. They reorganize the slice the tax administration leaves on the table, and as well as who the players could be. The reorganization is large in absolute dollars.
Introducing the Stablecoin - Pay roll - Float Stack!
issuer
custodian
orchestrator
card issuer
reserve manager
consumer wallet
Each node is a P&L line in someone’s projections. And currently, Bridge spans most of them.
Who are the issuers?
FinTech.
USDsui (Bridge), USDC (Circle), USDT (Tether), PYUSD (Paxos for PayPal), mUSD (Bridge Open Issuance for MetaMask), CASH (Bridge Open Issuance for Phantom) are some.
The total stablecoin market sits around $321 billion.
USDT carries about $190 billion and USDC about $77 billion. The duopoly of USDT and USDC is 85% of supply.
The issuer slot pays Treasury yield on reserves, gross, before whatever distribution rent the wallet extracts. The Circle distribution share is what the rent looks like in practice.
Circle and Coinbase: $908 million of $1.68 billion routed to Coinbase in 2024, roughly 54% of reserve revenue, under the distribution agreement.
To put this plainly, Circle, the issuer, paid 54% of its reserve earnings to the distributor, Coinbase. Yet Circle holds all the regulatory risks and is responsible for holding the peg.
Who custodies?
Banks basically.
Bridge holds an OCC conditional trust-bank charter as of February 2026. Anchorpoint
HSBC sit in the first wave of HK stablecoin licensees, with HSBC’s license issued April 10, 2026.
The Qivalis 12-bank euro consortium is targeting late 2026 under MiCA.
SAB 122 has been effective for fiscal years beginning January 2025, removing the balance-sheet penalty on US bank custody of crypto.
JPMorgan, BNY, State Street, and Bank of America already custody the ADP-side trust accounts that the sandwich’s bank-side leans on, and under SAB 122, they sit one regulatory step away from the bank-issued-stablecoin rung above (it means Banks are one product decision from issuing their own stablecoin).
The float ends up back inside the depository system, not outside it.
Who orchestrates? (Moves information between HRIS, stablecoin rails, and emplyoee wallet)
Payroll lookin companies.
Toku ran $1 billion of token-payroll volume by January 2026 (want to caveat that $1b in volume is probably crypto tokens) and integrates ADP, Workday, UKG, and Gusto on the HRIS side.
Also want to caveat that orchestrator can use the plumbing of other companies - Toku for example uses the plumbing of ADP.
RISE crossed $1 billion of total payroll volume in November 2025.
Both Toku and RISE are chain-agnostic.
Deel processes a reported $22 billion of annual payroll flow across 150 countries and chose Polygon, Solana, and Base for its stablecoin payouts, not Sui.
BVNK and Payoneer sit in the same orchestration layer with different geographic emphasis. The orchestrator captures a per-transaction fee and, in some cases, a yield share through arrangements like Paxos Amplify.
At this time, there are no Sui Native orchestrators, or Sui-adjacent. That leaves a nice vacancy.
Who manages reserves (currently)?
Asset management companies.
Superstate and Blackrock on USDsui
Blackrock on USDC
Cantor Fitzgerald +Anchorage holds a portion of Tether’s portfolio
The reserve-management slot is structurally similar to a money-market mandate at a large asset manager.
Margins are thin per basis point. The lock-in comes from regulatory comfort and operational integration, not yield differentiation.
Who spends?
Me.
This is the largest economic prize in the stack. Card interchange runs 1.5 to 2.2% of merchant volume in the US. Apply that to $11.7 trillion of private-sector W-2 wages at 30% penetration and the annual fee pool is $50 billion or more.
That is roughly ten times the narrow float pool prevously calculated (like $4b) and two to three times the wide one ($15b).
Rain is currently a contender for this slot.
Bridge plus Visa is positioned for this slot
Whoever wins the spend rail wins a category prize an order of magnitude larger than issuance, because the consumer surface is where the dollar terminates and where the per-transaction tax is highest.
Who primarily holds the wallet (core distribution layers)?
This is the fun new vector/
Coinbase distributes USDC.
Phantom distributes CASH (Bridge & Stripe).
MetaMask distributes mUSD (Bridge & Stripe).
The wallet is the layer closest to the user, which is why it captures the largest distribution rent. Hold the wallet, hold the customers.
The takeaway here is that we’re seeing which positions are getting captured for major chain : If nothing changes, Bridge is the cross-stack winner in its vertical integration play as it:
issuer (USDsui, mUSD, CASH),
custodian (OCC trust),
orchestrator platform (Open Issuance),
reserve-flow co-pilot - they sit with whoever manages the underlying Treasury reserves, so a partner in crime.
Card Issuer - partnered with Visa on a stablecoin-linked card program and Lead Bank as the issuing bank. The card issuer is the second-biggest economic prize after issuance and the easier slot to contest, because the distribution moat is shallower than the regulatory one.
Where there is opportunity
TL;DR: Spot the vacancies, and aim for profitability.
The yield ban is currently global and tenuous. We have, in no particular order: GENIUS in the US as well as CLARITY behind it. There’s MiCA in Europe, the HK Stablecoins Ordinance, the Japan PSA, and the Singapore MAS framework. Every regime converges on the same compromise: 100% reserves, segregated custody, no direct holder yield. That is the universal structural constraint, not a US idiosyncrasy.
Inside the constraint, the yield has to go somewhere. Perhaps a chain where the yield does not reach the holder; it reaches the network first might offer a new paradigm. That design is a possible answer to the global question of “if not the issuer and not the holder, then who,” and it is the answer that does not require a private distribution partner to capture half the value.
The cleanest knockdown of that design lives inside CLARITY’s tightening of GENIUS to anything “economically or functionally equivalent,” under which a future OCC or SEC could read a blockchain protocol’s buy-burn as functionally equivalent yield. CLARITY’s own “bona fide activities or bona fide transactions” carve-out points the other way, because the redirect flows to the network token and the protocol’s incentive layer rather than to a stablecoin holder’s balance. If final rulemaking reads it as functional yield anyway, well any chain that redirects reserve yields to the network instead of the issuer now has a narrower position: “yield-redirection answer” to “yield-redirection answer plus the empty orchestrator slot.”
A question that I’d like to assert is this: What would a protocol’s ecosystem stack be to support payroll float sandwich, do we have players there, and how survivable is it? If there are empty slots, does that mean it’s vacant and an opportunity for business to open? Toku is chain-agnostic but could serve as one of those players that “opens into a vacancy” but it doesn’t make it prime to capture the slot, unless its uncontested. Who are the rchestrator integrated into the HRIS/Payroll/Tax stack for each chain? Is this profitable?
Vacancies close fast when the underlying economics are visible - but I suppose the bigger question is - what are the economics then?
What makes our current times so exciting, and what makes banking so exciting today, is that we’re seeing very possible and very viable changes in financial power structures that aren’t as concentrated as they once were. Perhaps that’s because it’s early, and maybe in 50 years it’ll be the same story. But a statement like that isn’t a reason to compete today.
On the contrary, it means it’s an opportunity to be one of the big ones.
Like what you read? Share it around.
Want me to explore a topic? Drop me a note.
Want to make a correction or stress a thought out? I’m game.
Want to go running with me? Always open, just remember I’m usually slow.