The Economy Beneath the Minimum
Fixed fees do more than price digital payments: they decide how small exchange can become, which business models can exist, and which markets disappear before statistics ever see them
Keywords: digital payments; fixed fees; market formation; merchant acceptance; micropayments; fast payments; financial inclusion; platform pricing; AI agents; machine-to-machine commerce
Abstract
The modern payment debate is usually conducted after a transaction already exists. Regulators compare interchange rates, processors advertise merchant discounts, central banks measure payment volumes, and technologists celebrate faster settlement. That sequence misses a prior economic question: what if the structure of the payment price determines whether the transaction, service, or commercial relationship comes into existence at all? A fixed charge is not merely a small tax on completed exchange. At sufficiently low values it becomes a threshold. Below that threshold, a mutually valuable interaction can become privately irrational through a particular payment rail, a merchant can impose a minimum, a platform can bundle tiny services into subscriptions, an API can refuse pay-per-use pricing, or a machine can accumulate obligations rather than settle them individually. The most important consequence is statistical as well as commercial. Activity excluded by a payment threshold does not appear as a failed payment; often it does not appear in payment data at all. This essay develops the broader idea of an “invisible lower tail” of exchange. It distinguishes fixed from proportional fees, transactions from relationships, engineering cost from merchant-facing price, and faster settlement from economically finer settlement. It then applies the idea to small merchants, fast-payment systems, digital services, AI agents, and machine-to-machine commerce. The claim is not that every missing transaction should occur, nor that low fees automatically improve welfare. It is that the minimum economic scale of exchange is itself an institutional design variable, and payment architecture helps set it.
1. The market that disappears before anyone counts it
Payment statistics describe an economy that happened. They tell us how many card payments settled, how much value moved through bank accounts, which payment instruments consumers used, and how quickly funds became available. Those measurements are indispensable. The Federal Reserve’s most recent detailed national-payment release reports 153.3 billion U.S. general-purpose card payments in 2022 with a value of $9.76 trillion, while the Atlanta Fed’s 2023 consumer diary reports that Americans made 45.6 payments per month on average and used credit or debit cards for most payments by number (Board of Governors of the Federal Reserve System, 2024; Federal Reserve Bank of Atlanta, 2024). Europe shows the same broad digitalisation from another angle: the European Central Bank’s 2024 consumer study finds digital payments becoming more important even while cash remains disproportionately common for small-value purchases (European Central Bank, 2024). We therefore know a great deal about transactions that crossed a payment boundary.
We know much less about exchange that never crossed it.
That distinction is easy to miss because conventional payment data begin at the point of observation. A merchant accepts a card, a consumer initiates a transfer, a processor clears a message, a bank posts a debit, and the resulting record becomes measurable. If the merchant refuses the instrument, requires a minimum purchase, bundles several small obligations into one invoice, insists on a subscription, accepts cash instead, delays settlement, or simply never offers a micro-priced product because collecting the money would be absurd, no corresponding card transaction exists to analyse. The absence is not encoded as a neat zero in the payment dataset. It is usually absent from the dataset’s universe.
This produces a form of economic blindness. We often infer the shape of commerce from the shape of successful settlement. Yet successful settlement is selected by the rules, prices, acceptance decisions, technologies, and contractual arrangements that determine what reaches the rail. The measurement system is not neutral. It is part of the selection mechanism.
That point matters most at the bottom of the value distribution. Large transactions can absorb many small fixed charges without becoming uneconomic. Tiny transactions cannot. A 30-cent fixed payment component is irrelevant to a $10,000 invoice and enormous to a 50-cent digital service. The same pricing schedule can therefore be innocuous in the centre of the market and prohibitive at its edge. The result is not merely “higher cost.” It can be a different feasible set of business models.
The broader proposition of this essay is simple: the minimum viable scale of exchange is partly determined by payment architecture. Once that proposition is taken seriously, several familiar debates look different. Financial inclusion is no longer just about whether people possess accounts. Fast payments are no longer merely about speed. Merchant acceptance is no longer a yes-or-no property independent of ticket size. AI commerce is no longer just an API problem. And payment statistics no longer tell us the full distribution of demand that would exist under a different settlement technology or tariff.
2. Why a fixed fee is economically different from a percentage fee
A proportional fee is a slope. A fixed fee is an intercept. That difference sounds elementary, but it changes the economics of low-value exchange.
Consider two stylised payment prices. The first charges 3% of the transaction. The second charges 3% plus 30 cents. On a $100 payment, the difference is modest: $3.00 versus $3.30. On a $1 payment, it is dramatic: 3 cents versus 33 cents. On a 20-cent payment, the second schedule is economically nonsensical for almost any ordinary merchant because the fixed component exceeds the value being transferred before one even considers the percentage fee.
Current merchant prices make the point concrete. Stripe’s published standard U.S. domestic-card price is 2.9% plus 30 cents per successful transaction, and Square’s current U.S. rates include 2.6% plus 15 cents for standard in-person card payments and 2.9% plus 30 cents for API card payments (Square, n.d.; Stripe, n.d.-a). These are not universal prices and they are not measures of resource cost. They are observable examples of the two-part tariff form that matters for low-value exchange.
Suppose a seller has 10% of transaction value available as gross surplus from which to pay the payment charge. That assumption is not meant to describe a representative firm; it simply makes the geometry visible. Under a 2.9% plus 30-cent price, the payment charge equals the available surplus at roughly $4.23. Above that value, this narrowly defined payment comparison can work. Below it, the seller would lose more to the payment schedule than the assumed surplus available to absorb it. Change the seller margin and the threshold changes. Add a better outside payment option and the threshold changes. Add compliance, integration, fraud, or reconciliation burdens and it changes again. But as long as a positive fixed component remains, the system can produce a positive break-even point.
Figure 1. Illustrative economics, not an estimate of a representative merchant. The payment price uses Stripe’s published 2.9% + $0.30 standard U.S. domestic-card price; the 10% seller-surplus rate is an explicit assumption chosen only to expose the threshold geometry (Stripe, n.d.-a).
The crucial insight is that reducing the percentage rate and reducing the fixed component are not equivalent policies. If the fixed component remains positive, a lower proportional charge usually lowers an existing threshold but does not remove the basic low-value discontinuity. If the fixed component falls to zero, the tariff-induced positive minimum can disappear entirely, subject of course to other fixed burdens. Percentage pricing scales with economic value. Fixed pricing does not.
This is why debates framed only in terms of “average payment cost” can be misleading. Two payment systems can have the same average merchant charge over an existing mix of transactions while imposing radically different incentives on the smallest transactions. One system may recover more revenue through a percentage rate and little through a fixed component; another may do the opposite. Average cost can look similar while the set of feasible low-value transactions differs sharply.
There is an analogous regulatory lesson. U.S. Regulation II places a cap on covered debit-card interchange of $0.21 plus 0.05% of transaction value, with an additional fraud-prevention adjustment where eligible (Board of Governors of the Federal Reserve System, n.d.). In the European Union, the Interchange Fee Regulation permits domestic debit interchange structures that may include a fixed per-transaction component within specified constraints (European Parliament & Council of the European Union, 2015). Those rules concern upstream interchange, not the final merchant-facing tariff, but they illustrate the same structural fact: a fixed amount and a percentage amount are different instruments. One cannot infer the low-value effect of a policy merely from the statement that a fee was “capped.”
3. Merchant minimums are not an eccentricity; they are a clue
A card minimum is often treated as a small annoyance at a café or local shop. Economically it is much more interesting. It is revealed evidence that the value of the sale matters to the merchant’s willingness to accept a payment instrument.
The merchant does not compare card acceptance with nothing. The outside option may be cash, a bank transfer, another card, a wallet, a higher basket size, a lost sale, or a customer who simply buys elsewhere. The relevant question is therefore not whether card acceptance yields positive gross revenue. It is whether the merchant prefers the card transaction to the best available alternative after payment charges and other burdens are considered.
That is why a checkout minimum can exceed a bare processing break-even point. A merchant may still complete the sale if the card is refused because the customer switches to cash. If so, the merchant is comparing the net surplus from accepting the card with the net surplus from preserving the sale through another instrument. The better the fallback option, the less attractive it is to incur a fixed card fee on a tiny transaction.
The phenomenon is consistent with the broader pattern of payment choice. In the euro area, cash remains particularly important for small-value point-of-sale payments even as digital methods gain share overall (European Central Bank, 2024). In the United States, the Atlanta Fed diary likewise shows that cash remains widely used despite the dominance of cards in aggregate payment counts (Federal Reserve Bank of Atlanta, 2024). These broad surveys do not prove that fixed fees cause every low-value cash payment. Habits, privacy, acceptance, speed, budgeting, and merchant characteristics all matter. The important point is narrower: observed payment choice varies systematically with transaction value, which is exactly where a fixed component should matter most.
World Bank work on merchant acceptance reaches the issue from the development side. Ardic and Perez-Galindo (2026) argue that many countries now possess digital rails and widespread account access yet still encounter a stubborn merchant-acceptance gap among small shops. The problem is not simply whether a payment technology exists. Small merchants face thin margins, onboarding burdens, devices, service charges, and operational frictions. Denyes et al. (2021), in the World Bank’s Merchant Payments and Digital Financial Services Handbook, similarly treat merchant economics, acquisition, onboarding, business operations, and provider incentives as central to making digital acceptance commercially viable.
This changes the concept of financial inclusion. An account is not the same thing as a usable market. A person may own a wallet, a merchant may possess a QR code, and a country may operate a fast-payment system, yet low-value commerce can remain cash-heavy if the merchant-facing economics do not support digital acceptance at the relevant scale. Payment inclusion therefore has an intensive margin as well as an extensive one: not only who can access the rail, but how small a commercially sensible transaction can be once access exists.
4. The invisible lower tail: why settled data cannot reveal everything
The deepest implication concerns identification. Suppose a payment rail becomes uneconomic below some threshold. What can one learn about demand below that threshold from data generated by the rail itself?
Less than is commonly assumed.
Imagine two economies. In both, the distribution of successful transactions above $5 is identical. In the first economy, almost nobody wants to transact below $5. In the second, millions of potential low-value relationships exist, but the payment architecture makes them uneconomic and they therefore never reach the incumbent rail. If the researcher observes only successful incumbent payments, the two economies can look identical.
Figure 2. Conceptual illustration. The two latent economies are constructed to have different mass below the incumbent cutoff and the same observable tail above it. The point is identification, not a fitted distribution.
This is not merely a problem of choosing the wrong statistical distribution. A better Pareto, lognormal, spline, mixture model, neural density estimator, or machine-learning algorithm cannot recover information that the observation process systematically removes. If the mechanism itself excludes activity from the sample, more sophisticated estimation on the same truncated sample does not magically identify the missing mass.
The error is familiar in other domains. We do not infer the distribution of rejected mortgage applicants solely from the performance of approved borrowers. We do not estimate the ability distribution of children who never attend school from exam scores of enrolled pupils. We do not infer the productivity of firms prevented from entering a regulated industry from the accounts of incumbents. In each case, selection into observation is part of the economic process.
Payments have the same problem, but the selection mechanism is easy to overlook because a transaction record feels like a primitive fact. It is not. It is the endpoint of a chain of decisions: the service is offered, the customer chooses it, the merchant accepts the instrument, the payment is initiated, the processor accepts it, and settlement occurs. A low-value economic relationship can disappear at any earlier stage.
This is why the Federal Reserve Payments Study and consumer diaries are immensely valuable but should not be asked to answer questions they were not designed to answer. They measure activity that enters observable payment channels (Board of Governors of the Federal Reserve System, 2024; Federal Reserve Bank of Atlanta, 2024). They do not, by themselves, identify the population of commercial relationships that would exist under a different fixed-fee structure. That second question requires a threshold shift, an alternative rail, merchant-level relationship records, experimentation, or external bounds on the latent population.
The methodological lesson is severe: absence from settlement data is not evidence of absence from economic possibility.
5. The unit that matters is often the relationship, not the transaction
A second source of confusion is the tendency to treat a payment as identical to the economic relationship that generates it. They are not the same object.
A household may pay an electricity bill once per month even though electricity is consumed continuously. A cloud-computing customer may be billed monthly for millions of API calls. A creator platform may accumulate thousands of micro-earnings before making a payout. A logistics firm may net hundreds of reciprocal obligations before settling. A machine may consume bandwidth every second but pay once per day. A worker may perform many tasks but receive one wage payment.
The economic activity occurs at one frequency; settlement occurs at another.
This distinction is central to low-value commerce because batching can neutralise a fixed payment fee. A 20-cent fee is devastating if it must be paid on every 5-cent event. It is almost irrelevant if ten thousand 5-cent events can be accumulated into a single $500 settlement. The relevant object is therefore not the value of the underlying event alone. It is the value of the settlement unit after economically and legally feasible aggregation.
This is why the rise of fast payments should not be understood only as a race to instantaneity. The Committee on Payments and Market Infrastructures defined fast payments in terms of immediate or near-immediate availability, often on a 24/7 basis, and examined both their benefits and the factors that can support or hinder adoption (Committee on Payments and Market Infrastructures, 2016). The World Bank and CPMI later placed payment infrastructure within a broader financial-inclusion framework that includes access, usage, product design, risk, interoperability, and the surrounding institutional environment (Committee on Payments and Market Infrastructures & World Bank Group, 2020).
Speed is useful. But economically finer exchange requires more than speed. It requires a pricing architecture, access model, and settlement design that do not make small obligations irrational to express separately.
Consider FedNow. Its 2026 price schedule lists a $0.045 origination fee for a customer credit transfer, alongside other message and participation charges (Federal Reserve Financial Services, 2026). That is dramatically below many retail card fixed components, but it is a wholesale service price charged within a banking architecture, not a guarantee that an end merchant or end user faces four-and-a-half cents. Banks, processors, software providers, compliance services, account products, and business platforms can add their own charges. A low wholesale price changes the feasible opportunity set; pass-through determines whether the opportunity reaches the commercial edge.
The same distinction is visible in ACH. Nacha reports 35.2 billion ACH Network payments worth $93 trillion in 2025, including large volumes of business-to-business, consumer, payroll, and same-day payments (Nacha, 2026). Stripe, meanwhile, advertises ACH Direct Debit at 0.8% with a $5 cap for standard settlement, but also lists separate charges for instant account validation, disputes, and failed payments (Stripe, n.d.-b). No single headline number captures the economic burden of using the rail. Payment price, exception risk, integration, timing, compliance, and internal handling all matter.
The relationship perspective therefore protects against two opposite mistakes. The first is to exaggerate the importance of tiny event-level payments by assuming every event must settle separately. The second is to underestimate the importance of fixed fees by assuming all activity can be bundled indefinitely without cost. Real systems sit between those extremes. The key institutional question is what may be aggregated, for how long, at what risk, and at what cost.
6. Batching is an economic institution, not merely a technical trick
Engineers tend to describe batching as an efficiency feature. Economically it is more interesting: batching changes the boundary between feasible and infeasible exchange.
Suppose a platform charges 20 cents per settlement. A service that generates one cent of value per event cannot pay 20 cents after every event. But if 10,000 events can be settled together, the fixed charge is effectively spread over $100 of underlying value. The platform has not removed the fixed fee. It has changed the object to which the fixed fee attaches.
This is already how much of the economy works. Payroll batches labour into pay periods. Utilities batch consumption into bills. Credit cards batch purchases into statements for the consumer even though merchants settle separately. Advertising platforms batch impressions into campaigns and payouts. Marketplaces aggregate seller balances. Cloud providers meter at extremely fine resolution but invoice at coarser intervals. Telecom firms measure packets and bill plans. Subscription models are, in part, economic technologies for avoiding a payment negotiation at the level of every unit consumed.
That observation complicates the popular narrative of “micropayments.” The economically relevant innovation is not necessarily the ability to send one million individual one-cent transfers. It may be the ability to meter one million events, maintain auditable state, price them individually, and settle the resulting net obligation at the optimal frequency. The payment rail matters, but so do the ledger, contract, credit, netting, and finality arrangements around it.
This is especially important for AI agents. An agentic system may purchase tokens, model calls, storage, datasets, electricity, network bandwidth, identity attestations, or execution time in very small increments. If every atomic action requires irrevocable payment finality, a fixed fee can dominate the economics. If obligations can be netted or channelled into periodic settlement, the same underlying market can function even with a positive per-settlement charge. The design question is therefore not “Can an AI send money?” That is trivial. The harder question is “At what granularity can the AI make economically meaningful commitments without payment overhead forcing artificial bundling?”
The answer will vary by risk. A trusted counterparty can accumulate unsecured obligations. An unknown counterparty may require prepayment or rapid collateralisation. A regulated utility can bill monthly. A machine buying scarce computation from anonymous suppliers may require far tighter settlement. The optimal settlement interval is therefore endogenous to trust, volatility, credit exposure, enforcement, and technical finality.
This is why claims that “blockchain solves micropayments” or “instant payments solve machine commerce” are incomplete. Either technology may contribute, but the market boundary depends on the total institutional package. The cheapest message in the world is irrelevant if merchant onboarding costs $500, dispute handling is manual, compliance requires repeated review, or counterparties cannot safely accumulate exposure. Conversely, a payment rail with a non-zero fixed charge may still support extremely fine economic metering if aggregation is cheap and legally acceptable.
The unit of analysis is not the packet, click, token, or joule. It is the economic relationship plus its feasible settlement partition.
7. Faster is not the same as finer
Payment policy has spent much of the past decade celebrating speed, and with reason. Fast-payment systems can reduce settlement delay, improve liquidity, extend operating hours, and support more responsive business models. The World Bank reported in July 2026 that people and businesses in 137 countries had access to instant payment services on a 24/7 basis, describing fast payments as infrastructure capable of accelerating both liquidity and information (Banka & Bossone, 2026). These are significant gains.
Yet “instant” and “economically granular” are different properties.
A payment can settle in 200 milliseconds and still be uneconomic at $1 if the merchant-facing fixed charge is 30 cents. Another payment can take several hours and still support tiny economic relationships if the merchant-facing charge is near zero and obligations can be accumulated efficiently. Speed concerns when value becomes final. Granularity concerns how small an economically sensible unit of settled value can be.
The distinction becomes clearer when one compares wholesale and retail prices. FedNow’s 4.5-cent customer-credit-transfer origination fee in 2026 is a wholesale service price within the Federal Reserve’s participant structure (Federal Reserve Financial Services, 2026). Stripe’s standard card price contains a 30-cent fixed retail merchant component (Stripe, n.d.-a). Square’s standard in-person card price includes 15 cents, while its API card price includes 30 cents (Square, n.d.). These products provide different services, liability arrangements, acceptance networks, dispute rights, interfaces, and risk allocation. It would be wrong to treat the wholesale price as a direct substitute for the merchant retail price. But the contrast reveals a policy question: which parts of the fixed burden are technically necessary, which compensate valuable services, which reflect risk, and which survive because market structure permits them?
That question cannot be answered by engineering cost alone. Payment markets are institutional markets. A card acceptance price can include processing, fraud handling, rewards, chargeback rights, network access, software, compliance, acquirer services, and margin. A bank transfer can have lower interchange but impose different failure, mandate, reconciliation, or integration costs. A public fast-payment backbone can reduce wholesale switching cost without forcing any particular downstream retail tariff.
The Payment Systems Regulator’s review of U.K. card acquiring is instructive because it found that the supply of card-acquiring services did not work well for smaller merchants and identified search, switching, transparency, and pass-through problems (Payment Systems Regulator, 2021). That is a reminder that lower upstream cost does not mechanically become lower merchant price. Industrial organisation sits between technology and the user.
The correct policy vocabulary is therefore not “make payments faster.” It is “make the settlement architecture capable of supporting finer exchange where finer exchange is socially valuable.” Sometimes speed helps. Sometimes interoperability matters more. Sometimes merchant competition matters more. Sometimes integration cost dominates. Sometimes the fixed component of the retail tariff is the binding object.
8. Small merchants reveal the problem first
Large merchants can amortise complexity. Small merchants cannot.
A global retailer can negotiate bespoke acquiring terms, integrate several processors, route transactions strategically, manage fraud with dedicated teams, reconcile multiple feeds, and spread fixed implementation costs across enormous volume. A corner shop, market stall, sole trader, or informal microenterprise faces the same categories of burden at a radically smaller scale. That is why the economics of merchant acceptance are not captured by national account-ownership statistics.
Ardic and Perez-Galindo (2026) describe the small-shop acceptance problem precisely: the rails may exist and consumers may possess digital accounts, yet the transaction still stops at the merchant because the commercial case for acceptance is weak. The World Bank’s merchant-payments handbook likewise emphasises that successful digitisation requires providers to understand merchants’ business models rather than merely distribute accounts or acceptance devices (Denyes et al., 2021). The World Bank’s 2022 merchant survey instrument goes further by treating acceptance as something to be measured through merchant-specific costs, practices, experiences, and constraints rather than inferred from consumer account access (World Bank, 2022).
The fee schedule interacts with these burdens. A merchant facing a low percentage fee but a large onboarding cost may reject digital payments entirely. Another may accept cards but impose a minimum. Another may steer customers to bank transfer for large invoices and cards for convenience purchases. Another may absorb card fees on high-margin goods but not on low-margin staples. Merchant acceptance is therefore a portfolio of conditional decisions, not a binary technological state.
This also explains why cash persists. Cash has costs: handling, theft, counting, depositing, errors, security, and forgone data. But many of those costs are not charged as a conspicuous per-transaction line item. A merchant deciding whether to accept a $2 digital purchase sees the digital fee immediately, while some cash costs are pooled across the business. The comparison is therefore institutional as well as technological. A policy that wants to displace cash must compete with the way merchants perceive and bear costs, not merely with the engineering efficiency of electronic settlement.
The European evidence reinforces the point. The ECB reports that cash remains the most frequently used payment method at the point of sale in the euro area and is especially prevalent for smaller payments, even though most consumers say they prefer cards or other cashless methods overall (European Central Bank, 2024). This does not establish a causal fixed-fee effect, but it does show that value, acceptance, and instrument choice remain linked in mature payment markets.
For development policy, the implication is straightforward. The last mile of financial inclusion is not merely account opening. It is making digital acceptance economically rational for merchants whose average tickets and margins are smallest. That requires attention to merchant-facing fee structure, hardware and integration, liquidity, settlement timing, customer support, dispute processes, and competition among acceptance providers. A free account attached to an expensive transaction is not inclusion in any meaningful commercial sense.
9. The hidden cost of bundling: when payment frictions rewrite the product
One of the least appreciated effects of payment architecture is that it can alter not only whether a sale happens but what is sold.
If the payment system makes tiny transactions uneconomic, firms redesign products around the constraint. They bundle songs into subscriptions, articles into memberships, API requests into monthly invoices, electricity into billing periods, transport into stored-value balances, and creator earnings into payout thresholds. These arrangements may be efficient for independent reasons, but payment friction can push the market toward bundling even when consumers would prefer finer pricing.
This is economically significant because bundling changes competition. A consumer who wants one article can be forced to buy a monthly publication. A developer who needs twenty API calls can be pushed into a minimum monthly commitment. A small creator may wait until earnings exceed a payout threshold. An energy user can consume continuously but settle infrequently. A game may sell packs of virtual currency rather than individual actions. Each arrangement transforms the unit of purchase in order to make the payment layer economical.
Bundling can therefore conceal willingness to pay. Suppose consumers value individual digital objects at 10 cents, 20 cents, or 50 cents. If no feasible retail rail can collect those amounts efficiently, the market will not generate a dataset full of failed 20-cent purchases. The products will be bundled, financed by advertising, cross-subsidised, or never offered. Researchers later observing subscription revenue may conclude that consumers “prefer subscriptions.” Perhaps they do. But the observed contract form is partly endogenous to payment technology.
The same logic applies to platforms. A platform choosing between ad-funded, subscription, commission, and pay-per-use business models is not merely choosing marketing strategy. It is choosing around a payment constraint. If the settlement layer makes direct tiny payments expensive, the platform has an incentive to internalise transactions, create stored balances, issue credits, or move exchange onto an internal ledger. This is precisely why so many digital platforms create quasi-money inside the platform: points, credits, wallets, gift balances, game currencies, ad credits, and prepaid accounts. Internal accounting can occur at arbitrarily fine granularity because the expensive external settlement event is postponed.
The irony is that this often recentralises the system. A payment architecture that is expensive at fine granularity encourages firms to create private internal ledgers where micro-exchange is cheap. The economy does not abandon small exchange; it hides it behind a platform boundary. External settlement happens only when the user enters or exits the platform.
That is one reason the future of digital money cannot be evaluated only by transactions per second. The economically important question is who controls the internal ledger, who can interoperate with it, what it costs to move value across organisational boundaries, and whether tiny balances can migrate without being trapped by fees. A system can be technically fast and commercially siloed.
10. AI agents will make the minimum viable transaction a first-order design problem
Human commerce tolerates friction because humans transact relatively slowly. Agentic commerce changes that assumption.
An AI agent can, in principle, make thousands of economically meaningful choices per minute: buy a specialised data query, rent a few seconds of GPU time, pay for a software function, purchase a verification service, reserve bandwidth, compensate another agent for information, or bid for access to a scarce resource. The economic value of each action may be tiny even when the aggregate value of the agent’s activity is large.
This creates a mismatch with payment systems designed around human purchase sizes. A fixed 30-cent charge is trivial in a $200 retail purchase and absurd in a 3-cent machine decision. If every machine action must cross the same retail tariff, the payment system will determine the minimum economically expressible action.
There are four possible responses.
The first is aggregation. The agent records many obligations and settles them periodically. This is efficient where counterparties trust each other or where credit exposure can be bounded.
The second is prepayment. The agent deposits value into a wallet or service account and then spends against an internal balance. This removes repeated external settlement but creates custody, counterparty, and interoperability questions.
The third is netting. Reciprocal claims are offset before final settlement. This can dramatically reduce the number and value of external payments but requires a rule system for calculating and enforcing net positions.
The fourth is low-fixed-cost external settlement. If the rail itself supports cheap final transfers, agents can settle more frequently without moving onto proprietary internal ledgers.
These approaches are complements, not substitutes. The economically optimal architecture will depend on trust and finality. A model buying computation from a long-term cloud provider can tolerate invoicing. A model buying a one-off computation from an unknown machine across the world may prefer immediate or collateralised settlement. A grid-balancing agent may need near-real-time settlement because exposures change quickly. A content-recommendation agent may safely aggregate royalties for days.
Fast-payment infrastructure makes some of these designs easier. The global spread of instant-payment systems, the use of aliases and request-to-pay overlays, and the increasing integration of merchants into real-time rails can lower latency and coordination costs (Banka & Bossone, 2026). But the same warning applies: wholesale speed does not dictate retail fee structure. The market for machine payments will emerge only where total transaction costs—including integration, identity, fraud, compliance, liquidity, and dispute resolution—are compatible with the value of the underlying action.
The more autonomous economic choice becomes, the more important the payment minimum becomes. Humans can tolerate a world in which the smallest convenient purchase is a few dollars. Machines may not.
11. Low fees do not automatically mean social gain
It would be easy to turn the preceding argument into a simplistic policy slogan: fixed fees are bad, therefore abolish them. That conclusion does not follow.
A fee is a price, and prices can pay for real services. Card payments involve fraud management, dispute rights, network operation, credit provision, rewards, customer service, compliance, settlement, and software. Bank transfers involve their own authentication, account, exception, and compliance infrastructure. A low merchant price funded by cross-subsidy elsewhere is not necessarily socially cheaper than a higher transparent price. Nor is a privately excluded transaction automatically one that should occur from a welfare perspective.
The right benchmark is social surplus net of real resource costs and externalities. Suppose a tiny transaction creates 10 cents of private value but requires 20 cents of genuine compliance and fraud-management resources. Preventing it is efficient even if the payment rail could technically process the message for a fraction of a cent. Conversely, suppose a 30-cent merchant fee largely reflects market power or cross-subsidy while the real marginal social resources are tiny. Then the private threshold can sit above the socially efficient threshold.
This is where cost studies and competition analysis matter. The World Bank and CPMI stress that payment inclusion requires attention to the entire institutional ecosystem rather than to technology alone (Committee on Payments and Market Infrastructures & World Bank Group, 2020). The U.K. Payment Systems Regulator’s acquiring-market review shows why competitive pass-through and merchant switching can matter alongside underlying network costs (Payment Systems Regulator, 2021). The policy problem is to separate genuine resource requirements from transfers, rents, and inefficient contracting frictions.
A fixed fee can even have efficiency advantages. It may simplify pricing, allocate fixed infrastructure costs to users, discourage spam, prevent pathological transaction fragmentation, or compensate providers for per-event risks that do not scale with value. Fraud investigation can cost similar amounts whether the disputed payment was $5 or $500. Identity verification may be largely fixed. Support tickets are not proportional to purchase price. A purely percentage-based tariff can therefore underprice small events relative to their true expected service burden.
The economic claim is not “fixed fees are wrong.” It is “fixed fees create thresholds, and thresholds should be analysed as market-design variables rather than ignored as accounting details.”
12. Regulation should target the instrument that creates the problem
Payment regulation often focuses on the wrong layer because the industry contains several vertically related prices.
A merchant may pay an acquirer or payment service provider. The acquirer may pay network fees and interchange. The issuer receives interchange and provides account, credit, fraud, or reward services. A public payment system may charge participating banks. Software platforms may add gateway and orchestration fees. The final merchant price is therefore the product of a vertical chain.
Regulating one upstream price does not guarantee an equivalent movement in the merchant-facing fixed component. The Payment Systems Regulator’s card-acquiring review explicitly examined whether reductions associated with interchange regulation were passed through to merchants and found weaknesses in the acquiring market for smaller businesses (Payment Systems Regulator, 2021). Regulation II in the United States likewise caps covered debit interchange, but the cap itself is not the final merchant price (Board of Governors of the Federal Reserve System, n.d.). The European regulation distinguishes interchange from the merchant service charge and permits different structures for domestic debit within its cap architecture (European Parliament & Council of the European Union, 2015).
This creates a general rule for policy analysis: identify the price instrument that actually enters the user’s participation decision.
If the policy goal is to reduce the minimum viable merchant transaction, an upstream percentage cap may help, hurt, or do little depending on how processors and acquirers reprice. A merchant fixed-fee cap acts more directly on the threshold but can induce compensating percentage increases, subscription charges, reduced service quality, or eligibility restrictions. Interoperability policy may lower integration cost without touching the settlement fee. Fraud infrastructure may lower expected loss. Better digital identity may reduce onboarding costs. Open APIs may reduce software switching costs. Each instrument acts on a different component of the merchant’s effective burden.
This is why regulation by headline category is dangerous. “Interchange cap,” “instant payments,” “open banking,” and “low-cost rail” describe institutional changes, not final economic incidence. The commercial edge is determined by the whole stack.
13. What evidence would actually reveal the missing market?
If settled incumbent data cannot identify excluded activity, what could?
The strongest evidence would come from threshold movement. If a merchant-facing fixed fee changes sharply while other conditions remain reasonably stable, one can observe which transactions or relationships appear between the old and new thresholds. This is far more informative than extrapolating a statistical distribution into a region where the incumbent system previously generated no observations.
A second source is alternative-rail data. If one payment method has a lower fixed burden than another and the same merchant or customer can use both, the lower-cost rail may reveal activity absent from the higher-cost rail. The comparison is difficult because rails differ in fraud protection, settlement timing, reversibility, mandate structure, integration, and user experience. But matched merchant data across payment methods can begin to identify the relevant margin.
A third source is merchant relationship data rather than payment data. A merchant knows customers, invoices, subscriptions, attempted checkouts, abandoned carts, minimum-policy decisions, rejected tenders, and perhaps the payment method ultimately used. Those observations can reveal demand that does not appear in a card network’s successful transaction file.
A fourth source is experimentation. Payment providers can randomise fixed components, waive fees below certain values, change batching options, or vary payout thresholds. The resulting response would directly identify behavioural elasticities at the low-value boundary. Such experiments are commercially feasible because providers already use promotional pricing and differentiated plans.
A fifth source is external population bounds. Even without observing every latent relationship, credible information about the number of potential endpoints, firms, devices, invoices, or accounts can place an upper bound on the invisible region. This does not produce a point estimate, but it converts an unknowable magnitude into a bounded one.
The World Bank’s merchant survey architecture is relevant here because it treats merchant acceptance as a measurable object in its own right, with questions about costs, instruments, barriers, and business characteristics (World Bank, 2022). That is exactly the direction required: observe the participation decision rather than infer it from transactions that already survived the decision.
The central methodological rule is modest: do not estimate the invisible economy solely by extending the visible distribution downward. Use evidence that observes the boundary moving, observes an alternative rail, or observes the relationships directly.
14. The macroeconomic significance is market formation, not just cheaper payments
Most policy documents describe payment innovation through cost savings, speed, resilience, competition, or inclusion. Those are important, but the fixed-fee perspective adds another channel: market creation.
A lower payment burden can matter even when the saving on existing transactions is small. Suppose a new rail saves only a few cents on a typical $50 purchase. Measured over current transactions, that looks modest. But if the same price structure makes previously uneconomic $1 relationships viable, the relevant gain is not merely the cents saved on the $50 purchase. It is the possibility of an exchange that did not previously exist in that form.
This is analogous to other infrastructures. Broadband does not matter only because it reduces the cost of sending emails that would otherwise have been posted. It enables cloud software, video streaming, remote work, telemedicine, and businesses that make no sense under narrowband constraints. Container shipping did not merely shave freight costs from existing trade; it altered supply chains and the feasible geography of production. A lower fixed payment burden can similarly change product design, contract length, payout frequency, platform structure, and the set of counterparties worth serving.
The macroeconomic importance of fast-payment systems is increasingly being framed in those broader terms. Banka and Bossone (2026) emphasise liquidity and information acceleration: faster funds release working capital while structured payment histories can improve access to finance. The CPMI’s earlier work likewise recognised that fast payments alter availability, service design, and stakeholder incentives rather than simply reducing clock time (Committee on Payments and Market Infrastructures, 2016).
The fixed-fee perspective extends that logic one step further. Payment infrastructure can alter the minimum efficient commercial granularity of the economy. It affects whether a creator can be paid per use rather than per month, whether an API can charge per call rather than sell bundles, whether an energy market can settle locally rather than aggregate nationally, whether a micro-merchant can accept a $2 digital sale, and whether autonomous software can purchase an atomic service without routing everything through a dominant platform.
This is not a claim that trillions of dollars of hidden demand are waiting below an arbitrary threshold. The magnitude is precisely the difficult part. The claim is about mechanism: when price structure changes the feasible set, the welfare and industrial-organisation consequences need not be proportional to cost savings measured on transactions that already occur.
15. The political economy of the minimum
Once a payment minimum is understood as a market boundary, questions of market power become unavoidable.
A platform with a large installed base may prefer a tariff that maximises total profit across existing users rather than one that maximises the number of viable low-value relationships. Lowering a fixed fee can attract small transactions while cannibalising revenue from larger ones. Raising a percentage fee to compensate can alter demand elsewhere. Rewards can shift surplus to consumers while increasing merchant prices. Bundling payment services with software can obscure the standalone settlement price. A firm optimising across several instruments may therefore pass a cost reduction into the component that protects margin rather than the component that expands the low-value boundary.
This is why a reduction in engineering cost does not imply a proportional reduction in merchant fixed price. The processor chooses a tariff, not merely a markup over one homogeneous marginal cost. Competition can improve pass-through, but network effects, switching costs, proprietary integrations, data lock-in, and merchant inertia can weaken it. The PSR’s acquiring-market findings are a real-world reminder that merchants may face difficulty comparing and switching even in a mature market with multiple providers (Payment Systems Regulator, 2021).
The issue becomes more important as payment functionality is embedded into software. A restaurant does not buy “payments” in isolation; it may buy point-of-sale software, inventory, payroll, lending, loyalty, online ordering, and payment acceptance from one provider. A marketplace does not buy “settlement” in isolation; it buys compliance, onboarding, escrow, identity, payouts, and risk management. The relevant fixed burden can therefore migrate from the visible transaction fee into a monthly platform charge or integration requirement.
This creates a regulatory trap. A regulator can successfully compress one fee category while leaving the economic threshold unchanged because the provider reprices elsewhere. The right object is the effective burden of participation, not the nominal label attached to one line item.
16. A better way to think about payment innovation
The standard technology narrative asks whether a payment system is faster, cheaper, more scalable, or more decentralised. Those are useful engineering descriptors but incomplete economic ones.
A more informative framework asks five questions.
First: What is the fixed merchant-facing burden? This includes not only a visible per-transaction amount but also fixed per-settlement operational costs, dispute charges, verification, manual handling, and any event-level cost that does not fall proportionally with value.
Second: What can be batched? A system that supports cheap aggregation may enable fine underlying economic activity even with non-zero settlement fees. A system that requires every event to settle separately may impose a much harsher boundary.
Third: Who bears the burden? Wholesale price, interchange, processor cost, merchant price, and consumer price are different objects. Pass-through across the vertical chain matters.
Fourth: What is the outside option? A merchant who can accept cash, steer to bank transfer, or preserve the sale through another instrument has a different threshold from one who loses the customer completely.
Fifth: What remains unobserved? Any claim about latent market size should state whether it is measured, bounded, or merely extrapolated from transactions that already settled.
This framework is deliberately technology-neutral. Cards, ACH, instant bank payments, digital wallets, stablecoins, CBDCs, tokenised deposits, private ledgers, and public blockchains can all be analysed using the same questions. The label on the rail matters less than the economic perimeter.
The current market already demonstrates why that neutrality matters. Stripe’s pricing page now lists cards, instant bank payments, stablecoins, and local methods under materially different fee structures (Stripe, n.d.-a). Stripe’s ACH page adds caps, validation fees, failed-payment fees, and dispute fees (Stripe, n.d.-b). Square offers different prices for in-person cards, online cards, APIs, ACH transfers, and cash (Square, n.d.). FedNow exposes a different wholesale price architecture again (Federal Reserve Financial Services, 2026). None is “the cheapest payment system” in the abstract. Each is a bundle of price, risk, functionality, acceptance, and integration.
The economically important comparison is which architecture supports the desired form of exchange at the lowest total burden consistent with trust and regulation.
17. The invisible economy is not automatically a lost economy
There is a final conceptual trap. Once we recognise an invisible region below a payment threshold, it is tempting to call every unseen transaction a “missing market” and treat its absence as a social loss. That is too strong.
Some invisible exchange is not worth creating. Some tiny transactions would consume more attention, fraud-control resources, compliance effort, energy, or dispute capacity than the value they create. Some consumers prefer bundles. Some merchants rationally use minimums to simplify operations. Some platforms pool micropayments because customers hate constant payment prompts. Some services depend on subscription commitment to finance fixed costs. The social optimum is not an infinite number of infinitesimal settlements.
The correct claim is conditional: where positive social surplus exists below a privately imposed payment boundary, reducing the effective fixed burden can expand welfare-enhancing exchange. The existence of a private threshold is evidence of a constraint, not proof that the constraint is inefficient.
This distinction should discipline policy rhetoric. “Zero-fee payments” can be funded by hidden spreads, data extraction, cross-subsidy, float income, or monopoly elsewhere. Public systems still require infrastructure and governance. Compliance is costly because society has chosen to impose legal obligations. Fraud protection is valuable. Chargeback rights are valuable. Identity can be valuable. Reliability has a cost.
What matters is not whether the price is literally zero. What matters is whether the pricing and institutional architecture allows exchange to occur at the granularity where its social value exceeds its social cost.
That is a much harder question—and a much more interesting one.
18. Conclusion: the smallest viable payment is an economic institution
Payment systems are usually presented as plumbing. Money moves from A to B; economists then study the price, speed, and volume of what moved. That description is incomplete.
The plumbing shapes the building.
A fixed merchant-facing charge can create a positive economic minimum. A percentage fee usually changes the slope; a fixed fee can create the intercept that bites hardest at low values. Batching can move the effective minimum by changing what counts as one settlement. Merchant outside options can raise or lower the acceptance threshold. Fast-payment systems can reduce latency without guaranteeing fine-grained retail economics. Competition and vertical pass-through determine whether lower underlying costs reach the merchant-facing instrument. And because excluded activity never reaches the incumbent rail, settled-payment data can be fundamentally uninformative about the size of the invisible lower tail.
This way of thinking changes what we should ask of payment innovation. The question is not merely, “How much cheaper is this transaction?” It is, “What new transaction, contract, payout, service, or relationship becomes viable because the payment architecture changed?”
For small merchants, that may mean accepting a $2 digital sale without sacrificing the margin. For creators, it may mean receiving value closer to the moment of use. For software, it may mean charging for one function instead of forcing a subscription. For energy and communications, it may mean pricing consumption at finer intervals. For AI agents, it may mean purchasing data, compute, identity, or execution at machine scale without being trapped inside one platform’s proprietary wallet. For financial inclusion, it may mean moving from nominal account access to commercially useful participation.
The statistical implication is equally important. We should stop assuming that the visible payment distribution is a neutral map of economic desire. It is a map drawn after institutions have filtered the terrain. Different fee structures, settlement rules, trust arrangements, and batching possibilities can produce different visible economies from the same underlying willingness to exchange.
The economy beneath the minimum is therefore not a number waiting to be extrapolated from a chart. It is a research question about market formation.
And the smallest viable payment is not merely a technical specification. It is an economic institution.
References
Ardic, O., & Perez-Galindo, S. (2026, March 17). Why digital payments still stop at the small shop—and how to fix it. World Bank Blogs. https://blogs.worldbank.org/en/psd/why-digital-payments-still-stop-at-small-shop-how-to-fix-it
Banka, H., & Bossone, B. (2026, July 8). How fast payment systems drive economic growth and financial inclusion. World Bank Blogs. https://blogs.worldbank.org/en/psd/how-fast-payment-systems-drive-economic-growth-and-financial-inc
Board of Governors of the Federal Reserve System. (2024, November). The Federal Reserve Payments Study: National payment volumes, detailed data, NPIPS (CY 2021 and 2022). https://www.federalreserve.gov/paymentsystems/2024-November-The-Federal-Reserve-Payments-Study.htm
Board of Governors of the Federal Reserve System. (n.d.). Regulation II: Average debit card interchange fee by payment card network. Retrieved September 4, 2026, from https://www.federalreserve.gov/paymentsystems/regii-average-interchange-fee.htm
Committee on Payments and Market Infrastructures. (2016). Fast payments: Enhancing the speed and availability of retail payments. Bank for International Settlements. https://www.bis.org/publications/fast-payments-enhancing-speed-and-availability-retail-payments
Committee on Payments and Market Infrastructures, & World Bank Group. (2020). Payment aspects of financial inclusion in the fintech era. Bank for International Settlements. https://www.bis.org/cpmi/publ/d191.htm
Denyes, L. S., Ramji, M., Makin, P. C., Davies, K. S., Lesher, N. G., & Coleman, R. D. (2021). Merchant payments and digital financial services handbook. World Bank Group. https://documents.worldbank.org/en/publication/documents-reports/documentdetail/358361613558168171
European Central Bank. (2024). Study on the payment attitudes of consumers in the euro area 2024. https://www.ecb.europa.eu/stats/ecb_surveys/space/html/ecb.space2024~19d46f0f17.en.html
European Parliament, & Council of the European Union. (2015). Regulation (EU) 2015/751 of 29 April 2015 on interchange fees for card-based payment transactions. EUR-Lex. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX:32015R0751
Federal Reserve Bank of Atlanta. (2024, June 3). 2023 Survey and Diary of Consumer Payment Choice. https://www.atlantafed.org/research-and-data/surveys/survey-and-diary-of-consumer-payment-choice/2023-survey-and-diary
Federal Reserve Financial Services. (2026). FedNow Service 2026 fee schedule. https://www.frbservices.org/resources/fees/fednow-2026
Nacha. (2026). ACH Network volume and value statistics. https://www.nacha.org/content/ach-network-volume-and-value-statistics
Payment Systems Regulator. (2021, November 3). Market review into the supply of card-acquiring services: Final report (MR18/1.8). https://www.psr.org.uk/publications/market-reviews/mr1818-market-review-into-the-supply-of-card-acquiring-services-final-report/
Square. (n.d.). Calculate your rate with Square. Retrieved September 4, 2026, from https://squareup.com/us/en/payments/fee-calculator
Stripe. (n.d.-a). Pricing & fees. Retrieved September 4, 2026, from https://stripe.com/pricing
Stripe. (n.d.-b). Local payment methods pricing. Retrieved September 4, 2026, from https://stripe.com/pricing/local-payment-methods
World Bank. (2022, May 12). Electronic payments acceptance merchant survey instrument: Electronic Payment Acceptance Package. https://documents.worldbank.org/en/publication/documents-reports/documentdetail/099830005172285415