What we did not examine

Why this is separate from the gaps

The previous section listed things this work tried to establish and could not. Those affect how much weight the conclusions carry. They do not change what the conclusions are about.

This section is different. Each item below is a direction that was never investigated — in most cases because of how the work was sequenced rather than because anyone judged it unimportant. If any of them turns out to matter, it does not weaken the analysis so much as redirect it.

None of them is a large undertaking. Everything already on this site — every corridor, every rail, every licensing regime — was researched, cross-checked and written in a single working day. Most of the nine below are the same kind of work at the same scale.

The nine

1. The treasury product, which turns out to be a different business

One finding in this work identified a real proposition — repositioning liquidity across a weekend or a bank holiday, when the settlement systems are closed — and noted that it is sold to a treasurer rather than to a payments lead. Different buyer, different budget, different sales conversation, different competitors. Having correctly separated it, nobody then sized it.

Treasury revenue turns out to be a function of one variable, and it is not product design. Interest on customer balances beats transaction revenue only once the average balance, expressed in days of annual volume, passes a threshold set by the take rate and the yield. The two firms that earn seriously from this hold roughly 29 and 39 days of volume as customer money. A business holding payments in transit for one to three days earns under three basis points from the same mechanism. The float business and the payments business are separated by a factor of fifteen to forty in dwell time.

Two findings inside that were genuinely unexpected. The interest is legally ours to keep across every European regime examined — not because anyone negotiated well, but because the rules forbid passing it on; the market treats as a competitive variable something that is a legal one. And the same rules prevent scaling it, because an issuer barred from paying interest cannot compete for balances on price. The line accrues to whoever owns the account, not to infrastructure sitting behind them. Selling hedging is closed off separately, on the primary text: the carve-out that keeps a forward outside investment-services regulation requires an identifiable underlying transaction, and a hedge against a forecast exposure is exactly what a treasurer wants to buy.

2. Freight, commodities and trade finance — precedents exist; demand does not

Shipments have deadlines with contractual penalties attached, which means a payment arriving late has a cost the buyer can already quantify. That is rare, and it is precisely the condition under which a guarantee about timing can be priced.

The category is no longer evidentially empty. Contour combines digital trade documents, ERP integration and stablecoin-settlement work; Komgo has live multi-bank corporate deployments; and TradeLens shows that technical viability and large institutional sponsors do not guarantee commercial viability. What remains unexamined is the Atlas-specific demand: the trade instruction, document, financing and payment failures a named buyer would pay Atlas to control.

3. High-risk and iGaming — the weakest exclusion, and the arithmetic that settles it

This segment was ranked first on willingness to pay, and it is the one segment whose actual problem is route selection, because providers withdraw from it without notice and operators genuinely need alternatives ready. It was nonetheless set aside on a judgement about regulatory and reputational cost that was never sourced and never quantified — the exclusion least supported by evidence anywhere in this work.

One listed operator's audited accounts settle it. On the same gambling customer base it earns 0.6% on card acquiring and 3.1% on its consumer wallet — around 1,500 operators and $167bn of volume. The 250 basis points between those two numbers is not paid for routing or for approval-rate optimisation. It is paid for being the licensed counterparty of record to the player. Everything an orchestration layer does lives inside the smaller number, and the smaller number is falling by about ten basis points a year.

So the segment should be declined on the take rate rather than on reputation, which is a defensible position rather than a squeamish one. The profitable layer needs a licence and a consumer relationship; the accessible layer pays less than the compliance stack costs at the prevailing ticket size. It is a good business for whoever already holds the licences, and buying into being that party is a different company.

Banking access, which this work had feared and could not evidence, is now sourced from an audited filing: the same operator discloses having lost three important banking relationships for its wallet business, and an operator on the other side of the table discloses that processors may withdraw from the industry as a whole. The fear was correct.

The wider high-risk question resolved differently and more usefully. Everything that makes these merchants risky — chargebacks, reserves, scheme registration, disqualification — attaches to card acceptance. Almost none of it attaches to paying money out. Which means the payout leg is a reasonable business and the acceptance leg is not, and we should not buy the second to reach the first. The deflating corollary is that a payout-only provider bears none of the risks the premium compensates, so it has no claim on that premium — it can charge for the currency conversion, which is a larger pool anyway, and which is what the rest of this work already concluded from an entirely different direction.

4. Starting from a currency other than the euro

Every corridor examined begins in euros. That was inherited from how the question was first framed and never decided on its merits.

It matters more than it sounds, because the conversion carries most of the margin and its economics are not symmetric. A dollar-origin book faces different competition, different liquidity and a different regulatory perimeter. The finding that the deep token markets are dollar-denominated — which forces an awkward double conversion for a euro-origin business — simply does not apply to a dollar-origin one.

5. Domestic and intra-European adjacencies

Two proprietary data assets were identified during this work: the rate at which the European payee-verification check returns a mismatch, and the per-institution value limits that replaced the removed scheme cap. Both were noted as things nobody publishes. Neither was sized as a business.

They are unlikely to be businesses on their own. They may be the residue that makes another product defensible, which is a different argument and one nobody made.

6. Acquisition as a strategy rather than as an outcome — tested, and it did not survive

This work concluded that acquisition is the probable end state and then planned as though independence were the goal. Those are different plans. Building deliberately for a specific acquirer changes which licences are worth holding, which corridors are worth entering and which data is worth accumulating — so the obvious move was to identify corridors that are structurally underserved, become the only credible licensed non-bank in one of them, and be bought for it.

The premise was tested directly, with two instruments, and it failed both tests. The first instrument was a search of every filing made by a sixteen-company set of plausible acquirers over three years, looking for statements of intended expansion. The corridors this work nominated were named by nobody; searched across all filers rather than that set, the phrase "expand into Central America" returns nothing at all.

That result is weaker than it looks, and the researcher said so. Filings are where firms disclose risk and material events, not where they announce appetite — so an absence there may mean the instrument cannot see the thing it was pointed at. Widening it to what acquirers say on their own earnings calls closed that hole, and the answer got worse for the strategy rather than better.

Acquirers are not ignoring scarce corridors. They are entering them themselves, within quarters. One named entering Qatar and Kuwait in May — the two corridors this work had measured as having zero non-bank suppliers three months later. Another added Niger and Mali in the quarter ending six days before this research was written, both of them members of the West African bloc our own scarcity map ranks first. A third named Peru, Colombia, Costa Rica and Nigeria, and described displacing an incumbent processor.

Which is the finding. The buyer's alternative to acquiring the only licensed firm in a thin corridor is to walk into that corridor directly, on a timescale of quarters, for the cost of an application. Scarcity that a buyer can resolve by applying is not an asset you can sell them. And the one region nobody entered, at any point in the window, is the one the strategy nominated — the only statement about it found anywhere predates the window by eighteen months.

The one acquirer that did want depth in those corridors bought it, and paid 0.78 times a declining revenue line — the cheap half of the comparable set, against another transaction in the same window that paid $1.8bn for orchestration plus twenty-five licences and live volume. Its stated rationale, on its own investor call, was ten thousand retail agent locations and twenty million existing customers. The word that never appears as a reason is licence. The buyer paid for demand, not for supply.

And the licence itself prices in a way that inverts the strategy. Two audited purchase-price allocations each valued a payments permission: one in a small African market at 0.5% of the price, one passportable European licence at 35.7%. The same class of asset, one hundred and seventy-eight times apart, with the licence and the customer book behaving as substitutes — the buyer pays for whichever was scarce. Scarcity of supply in a thin corridor is not what commanded the premium. Breadth of reach was.

The arithmetic then closes it. Applying this work's own measured dispersion to the entire euro-origin trade flow through the thinnest corridors, at a realistic share, yields less per year than a single compliance stack costs to run. Those corridors are underserved because the money moving through them is US-origin retail remittance — roughly ten times the euro-origin trade flow, and nineteen times for one of them. It is a different business from this one. And the arithmetic is generous in one further way: the European invoicing-currency statistics it leans on are not published per partner country, so even the euro-origin numerator is an assumption rather than a statistic — commodity trade in particular is routinely dollar-invoiced and settled through a global bank. Every correction points the same direction: downward.

One further thing fell out of putting two of these documents side by side, which neither had drawn on its own. Deals in this sector that involve a licensed money-transmitter target take a median of about twelve and a half months from announcement to completion. Swedish takeover rules cap a permit-conditional public offer at nine months. Those two numbers have never been compared, and they should have been: the public offer route does not fail in some unlucky tail, it fails in the median case. Any exit structured as a public offer for a group holding payment licences needs either a different structure or a regulatory pre-clearance strategy that starts long before the offer does.

This item is the reason the section exists. It was opened expecting to convert a conclusion into a plan, and it removed the plan instead.

7. Corridors outside the shortlist

Four were added later in response to a challenge that the work was Latin-America-focused: South Africa, the European non-euro pairs, the Gulf and China. Three were rejected on evidence and one was added to the sequence.

That leaves genuine absences — intra-African and other emerging-market to emerging-market corridors, which are non-euro-origin by construction and therefore fall outside the framing in item 4; and several large trade corridors that were never on the list because the list was inherited from a ranking of where correspondent banking fails, which is not the same question as where money can be made.

8. Two costs nobody put in the plan

The payments-company interpretation depends on accumulating operational data before its guarantee can be sold — you cannot promise a delivery time you have never measured. That implies running volume unguaranteed first, potentially at a loss, for some period. Neither the cost of that period nor the working capital it consumes appears in any figure here.

9. The banking platform as an existing strategic asset

The platform-company interpretation rests on unified transfer states, beneficiary controls, approval workflows, provider adapters, compliance services, treasury states and reconciliation machinery already present in Atlas. This research has not independently established which of those capabilities are production-live, which are gated or simulated, how much implementation remains, or whether a regulated institution will pay for them.

That requires two different instruments: a capability audit against deployed systems and a paid design-partner process against named bank, EMI, PSP or trade-platform buyers. Without both, the platform is evidence of relevant product IP but not yet evidence of a business.

What it would take to open them

Items 2, 4, 5 and 7 need focused market work. Item 9 needs a capability audit and buyer evidence rather than more desk research. Item 8 is not research at all — it is a planning decision that needs to be made.

None of them was skipped because it was judged unimportant. They were skipped because the work was sequenced to answer one question first, and that question turned out to take all of it.