The headline said ninety percent. The methodology said nothing.
That asymmetry is the whole story. A Crypto Briefing item crossed my feed this week reporting that ninety percent of businesses using stablecoins at scale consider the effort worthwhile. No sample size. No sampling frame. No geography. No industry distribution. No survey date. No link to the underlying report. No byline. Seven absences and one integer, packaged for a market that is currently chop-bound, direction-starved, and unusually receptive to any number that sounds like confirmation.
I have spent the last four years taking apart documents exactly like this one. In 2021 I burned four hundred hours on Luno's staking contracts while their marketing team was running a viral campaign, and I found a reentrancy path that let users drain liquidity without authorization checks. The team asked me to sit on it for community sentiment. I published instead. The lesson from that episode was not that projects lie. It was that the lie is almost never in the code. It is in the framing around the code. The code spoke, but the logic was a lie โ and the logic was written by the marketing department, not the engineers.
The same structure applies here. The number is probably not fabricated. The number is probably correctly computed. The number is also almost certainly answering a question nobody in the article actually asked. What follows is a dissection of that gap.
Context: who commissioned this, and why the timing matters
The survey was conducted under the Payoneer banner. Payoneer is a Nasdaq-listed B2B cross-border payments company, ticker PAYO, with a client base weighted toward small and mid-sized enterprises, marketplace sellers, and freelancers operating across emerging markets. It holds payment licenses in dozens of jurisdictions. It has real operational history. It is not a shell.
That matters, because it constrains what kind of bias we should expect. A listed company publishing research usually routes the material through legal review. Outright fabrication is off the table. What remains on the table is far more useful to a critic: control over the sampling frame, control over the question wording, control over which cut of the data gets published, and control over the headline sentence that a secondary outlet will quote. That is compliant bias. It is also more effective than fraud, because it survives a fact-check.
The distribution channel is equally instructive. Crypto Briefing ran the item without a byline and without a link to the source report. That means no named human is accountable for the accuracy of the summary, and no reader can trace the claim back to a PDF. What we are reading is a third-hand conclusion: Payoneer's research team produced a finding, a communications function converted it into a press asset, and a crypto-native outlet compressed that asset into a paragraph. Each hop is a lossy compression step. By the time it reaches a reader, the qualifier โ the two words "at scale" โ is the first thing to fall off.
Set this against the 2025 regulatory calendar. The United States moved a federal payment-stablecoin framework into law. The European Union's MiCA EMT and ART chapters are fully applicable and require issuer licensing. Hong Kong's stablecoin ordinance has an active licensing regime. Singapore and the UAE have published or are finalizing their own rules. Every one of those developments is net positive for enterprise adoption, because it converts legal ambiguity into a compliance cost, and compliance costs are budgetable while ambiguity is not.
Which means the value of any adoption survey is entirely a function of when it was fielded. A survey conducted before the frameworks landed measures sentiment under gray-zone convenience. A survey conducted after measures sentiment under real cost. Those are different datasets with different policy implications, and the article discloses neither. That single omission is worth more scrutiny than the ninety percent itself.
There is a second contextual force. Stablecoin payments are the most crowded narrative in crypto right now. Issuers, processors, card networks, and consultancies are all producing supportive research simultaneously. This is what an acceleration phase looks like from the inside: not a single persuasive document, but a high-volume environment of mutually reinforcing ones. Selective disclosure becomes structural rather than individual. Nobody has to coordinate. Everyone has the same incentive.
Core: the denominator is the argument
Start with the sentence as written. Ninety percent of businesses using stablecoins at scale say it was worth the effort.
The denominator is not "businesses that tried stablecoins." The denominator is businesses that already reached scale. That is a conditional probability dressed as a base rate, and the distinction is not academic โ it is the entire difference between a compelling statistic and a meaningless one.
Run the arithmetic. Suppose one hundred enterprises attempted stablecoin settlement. Sixty abandon it inside two quarters: a banking partner severs the on-ramp, a KYT vendor throws a false positive that freezes a payroll run, the auditor refuses the treasury treatment, the finance team cannot reconcile three chains against one ERP ledger. Of the forty that persist, thirty reach what they would describe as scale. Twenty-seven of those thirty report the effort was worthwhile.
The headline writes itself: ninety percent say it was worth it. The full-sample conversion rate is twenty-seven percent. Both numbers are true. Only one of them is a fact about the world, and the other is a fact about the survivors.
If that framing feels abstract, write it as a contract. The logic is four lines:
function reportedSatisfaction(address[] memory firms) public pure returns (uint256) {
uint256 survivors;
for (uint256 i = 0; i < firms.length; i++) {
if (abandoned(firms[i])) continue; // removed from the denominator
survivors += 1;
}
return (survivors * 100) / survivors; // always 100
}The function cannot return anything except one hundred. Not because the data is wrong, but because the filter runs before the division. This is not a metaphor. It is the exact mechanical structure of the published statistic. Trust is a variable you cannot hardcode, and neither is a base rate โ you have to actually count the people who left.
Now inventory what the document does not disclose. Sample size, which determines whether ninety percent means twenty-seven out of thirty or nine out of ten. Sampling frame, which determines whether the respondents came from Payoneer's own client pool โ a population already selected for willingness to use digital cross-border rails. Industry and geography, which determine whether the finding generalizes to Singapore and Switzerland or only to Lagos, Sรฃo Paulo, Manila, and Warsaw. Cost basis, which determines whether "worth the effort" reflects a positive net present value or a sunk-cost rationalization. Fielding date. Source report. Sponsor's commercial interest in the outcome.
Seven fields. Zero disclosed. The confidence interval on this statistic is not wide. It is undefined.
"At scale" is not a unit of measurement
Here is the second structural defect, and in my experience it is the one that survives peer review longest because it hides inside ordinary language.
At scale has no definition in the document. A firm processing one hundred thousand dollars a month in USDC payouts to contractors considers itself at scale relative to its prior wire transfers. A firm processing one hundred million dollars a month in treasury settlement considers itself at scale relative to a mid-cap bank. Both are correct in their own frame. Both answer the same question. The resulting dataset mixes them without stratification.
When I audited three Layer-2 fraud proof implementations during the 2022 retreat, I found the same pattern in a different costume. Two projects described themselves as decentralized. Both routed fault proofs through a permissioned validator set controlled by the founding team. The word decentralized was doing no technical work at all โ it was a label applied to a spectrum, and the spectrum had never been partitioned into categories with thresholds. Their decks were not lies. They were category errors wearing a suit.
At scale is that error in payments form. It cannot be aggregated, compared, or extrapolated until someone defines the boundary: monthly notional, number of counterparties, jurisdictions touched, percentage of total payment volume, number of chains. Absent those thresholds, the ninety percent figure describes an unknown population with an unknown central tendency. Data does not lie, but it does not care either โ it will sit in a slide deck for three years and never once object to being misread.
The cost stack the survey never itemizes
The phrase worth the effort is doing enormous hidden labor. Effort is the cost side of a cost-benefit claim, and the document never decomposes it. In my experience auditing enterprise payment integrations, the cost stack looks nothing like the pitch.
Issuance and redemption fees. On and off-ramp spreads, which are frequently the largest single line item and are almost never quoted as a percentage in vendor decks. Banking rail availability, including the real risk that a correspondent bank simply declines the flow without explanation. KYT and sanctions screening, priced per transaction and prone to false positives that halt settlement. Travel Rule messaging infrastructure, which is jurisdiction-dependent and frequently manual. Accounting and tax treatment, where the auditor's opinion can invalidate the entire program regardless of how well the technology works. Key governance โ MPC, HSM, multi-signature policy, signer rotation, disaster recovery. ERP and treasury system integration, which is unglamorous, expensive, and the actual reason pilots stall.
Note where the friction sits. It is not on-chain. Block space is cheap and finality is solved for the chains enterprises actually use. The friction is in the last mile: the fiat edge, the compliance layer, and the accounting layer. Every one of those is a cost the survey would have had to measure to justify the word effort. It measured a feeling instead.
This is where I would push hardest if I were the counterparty on this research. Show me the abandonment cohort. Show me the firms that reached pilot and stopped. Show me the reasons. In four years of due diligence work I have never once seen a stablecoin payment program fail because the chain was too slow. I have seen them fail because a compliance vendor flagged a supplier, because a bank pulled the account, and because a controller could not get comfortable with the balance sheet presentation. Those failures are quiet. They do not produce press releases. They produce a denominator that quietly shrinks and a percentage that quietly rises.
Where the seigniorage actually goes
Here is the layer the survey does not touch, and it is the layer that determines whether the enterprise economics are as good as advertised.
The stablecoin issuer mints a liability at zero marginal cost and holds reserve assets โ short-dated government paper, repo, money market funds โ that yield at the policy rate. The spread between zero and the policy rate is the issuer's revenue. That is the business. It is a genuine, auditable, non-Ponzi cash flow, and it is entirely invisible in a survey of enterprise satisfaction, because the enterprises are not the ones capturing it.
This matters for two reasons. First, the enterprise is paying fees to a counterparty whose revenue model is rate-sensitive in a way the enterprise is not. When the policy rate compresses, issuer revenue compresses, and the pressure to raise fees or monetize float moves downstream. A multi-year contract signed at current fee levels is a contract exposed to that repricing. Second, the enterprise holds a reserve-backed instrument whose risk profile is the issuer's balance sheet, not the chain's. The technical layer is decentralized in the sense that matters to nobody. The counterparty layer is a single legal entity with freeze and blacklist authority.
That authority is the actual enterprise risk, and it is a compliance risk rather than a cryptographic one. A mainstream stablecoin issuer can freeze an address on law enforcement instruction. An enterprise running payroll through that address has a business continuity problem, not a security bug. Any credible survey of enterprise adoption would have asked about this. This one, on the available evidence, did not.
The distribution fight behind the press release
Strip the research framing and the artifact is a positioning move, which is neither scandalous nor surprising. The stablecoin value chain runs from issuers at one end to enterprise finance departments at the other, with settlement chains, custodians, compliance vendors, and ramps in between. Payoneer sits in the middle layer. It has weak bargaining power upstream โ it buys the same USDC everyone buys โ and strong relationships downstream, where it owns the SMB and freelancer customer.
For a middle-layer company, the strategic asset is not the technology. It is the definition. Whoever defines what counts as mature stablecoin adoption also defines the benchmark against which every enterprise program will be measured, and the vendor that supplies the benchmark has a natural claim to supply the implementation. This is category education with a commercial tail. That does not make the research false. It makes the research interested.
The competitive field makes the interest legible. Stripe bought its way into stablecoin payments. Circle issues the reserve asset and sells the compliance story. Ripple issues its own. Wise watches. A cluster of B2B processors in emerging-market acquiring are running parallel pilots. In a field this crowded, publishing the first benchmark is a cheap way to become the reference point.
They built a palace on a fault line, and the fault line is that the benchmark was constructed before anyone agreed on the foundation. The building will stand until someone publishes a competing survey with a disclosed methodology and a lower number, at which point the ninety percent becomes a liability rather than an asset.
Contrarian: what the bulls got right
The lazy read of this document is that it proves nothing and therefore means nothing. That read is wrong, and it misses the strongest signal in the material.
Consider what has to be true for the finding to exist at all. A payment company with legal review decided that enterprise stablecoin adoption was a large enough commercial category to fund primary research on. That decision is a market signal independent of the research output. Nobody commissions surveys about categories they believe are dead. The existence of the study is evidence about the study's sponsor, and the sponsor's revealed preference is more informative than its stated finding.
Second, the underlying fundamentals are not vapor. Stablecoin settlement volumes are real, the reserve income model is real, and the regulatory frameworks that landed in 2024 and 2025 converted an ambiguous category into a licensable one. This is not a DePIN token with a whitepaper and no revenue. When I modeled Compound's rate curves during DeFi Summer, the failure I predicted came from a mechanism that had no revenue floor underneath it. Stablecoin payments have a revenue floor: real enterprises paying real counterparties across real borders because the correspondent banking network is slow and expensive. That floor does not disappear in a drawdown.
Third โ and this is the counterintuitive point โ the friction is the finding. Enterprises reporting that the effort was worthwhile while simultaneously absorbing bank rail risk, KYT false positives, multi-jurisdiction licensing complexity, and unresolved accounting treatment is a far stronger statement than a clean ninety percent on a cost-free product. Nobody calls a painful process worthwhile unless the alternative is worse. The alternative is correspondent banking with a three-day settlement window, an opaque FX spread, and a rejection rate that nobody publishes either.
What the bulls have wrong is narrower and more consequential. They have confused a sentiment reading with an adoption curve. Sentiment measures how the survivors feel. Adoption is measured by the abandonment rate, the median time to production, and the percentage of total payment volume migrated โ none of which appear here. A category can have ninety percent survivor satisfaction and a thirty percent conversion rate simultaneously, and the second number determines the market size.
Takeaway
One question resolves this document. What was the n, and who was in the denominator?
Until that is answered, treat the ninety percent as a directional signal from an interested party, and treat the missing qualifier as the load-bearing element. When the figure reappears โ and it will, in pitch decks, in vendor one-pagers, in the next consultancy report that cites it without the words at scale โ the qualifier will be gone. Statistical drift is not a hypothetical. It is the default behavior of a number that travels faster than its footnote.
My prediction: within twelve months, at least one competing survey with a disclosed methodology will report enterprise satisfaction in the sixty to seventy-five percent band. That will not be a contradiction. It will be the same reality measured with the abandoners included. The interesting question is not which number is right. It is who publishes the abandonment rate first โ and why nobody in this cycle has volunteered to.
Audit the denominator before you audit the claim.