Buy your own licence when the rent you pay, corrected for the day the sponsor reprices you, passes the annual cost of owning the permission yourself. That crossover arrives earlier than the sticker price suggests, because the rent is not fixed. It rises toward the most it can take from you, and it rises fastest at the moment you can least afford to move.
You did not buy cheaper access. You rented the exposure and kept the risk. The sponsor kept the licence, and with it the right to change the terms whenever their own regulator changes theirs.
This is the decision every founder on a sponsor-bank, agency, or licence-rental arrangement is carrying whether they have priced it or not. You launched fast because a partner already held the permission. Now the permission is a counterparty, and the counterparty prices.
Why these three models
The event is a renewal, or a regulator’s circular, that changes what your rented permission costs. Three lenses read it, and they disagree in useful ways.
The first treats the arrangement as a bargain between two parties whose power shifts after you commit. The second treats it as an arithmetic problem: rent is a variable cost, a licence is a fixed one, and the two lines cross somewhere. The third treats the repricing as a discontinuity: the sponsor tolerates you until you cross a line, then squeezes in one step.
Two of these sit in the same family. Bargaining and the cost crossover are both equilibrium models: they assume each side optimises and lands where the numbers settle. The threshold model is different in kind. It is about tipping, not settling. The natural fourth lens here, a regime-switching model of the regulator flipping states, is deliberately left out. This series has leaned on it too often, and the honesty of an ensemble comes from diverse errors, not from a fuller shelf. Two outcome families, read against each other, are enough to route the decision. I say where the missing third would have changed the answer.
The framework
1. Bargaining: the rent rises to your switching cost
Before you sign, you have options. Several sponsors want your volume, and you can walk. After you build on one sponsor’s rails, wire your product to their APIs, and move your customers’ money through their accounts, walking costs you a rebuild, a migration, and a regulatory conversation you have not had. That sunk, hard-to-move investment is what economists since Klein, Crawford and Alchian have called an appropriable quasi-rent: value that exists only inside this relationship, and that the other party can now reach for.1
The mechanism is not malice. It is structure. Once your switching cost is high, the sponsor can raise the price, tighten a rule, or change a revenue share, and you will still pay, because paying is cheaper than leaving. Your switching cost is the ceiling on the rent, and you built that ceiling yourself.
The Synapse collapse showed the extreme version in public. Fintech apps that were, in the reporting’s own words, “not banks” reached their customers’ money only through a middleware provider sitting on top of a sponsor bank. When the arrangement broke, “the fintech middleman turned off access to a key system,” a court-appointed trustee found up to 96 million dollars of customer funds missing, and one saver was offered 500 dollars against a balance of 280,000.6 Roughly 200 million dollars of customer money sat frozen.7 The switch was never yours to hold.
Bargaining and hold-up
Assumes: your investment in the sponsor is specific and hard to redeploy; the sponsor is rational and reprices to your walk-away point.
Fits because: the rent you pay tracks your switching cost, not the sponsor’s cost of serving you.
Breaks when: switching is cheap (two live rails, portable KYC), or reputation makes the sponsor value keeping you more than squeezing you.
Counteracts: the crossover model, which assumes a fixed rent rate; here the rate itself climbs.
May reinforce: the threshold model, since the squeeze often lands as one step, not a drift.
2. The crossover: rent is variable, a licence is fixed
Renting is a variable cost. You pay a share of revenue or a fee per transaction, so the bill grows with your volume. Owning is mostly a fixed cost, and in this market a large one. In Nigeria, a Mobile Money Operator licence and a Switching licence each carry 2 billion naira in minimum shareholders’ capital, with a matching escrow deposit; a Payment Solution Service Provider licence carries 100 million. That capital is not a fee. It is money you must own and reflect on your balance sheet.5 Add compliance staff, audits, and the reporting a direct licensee carries.
So the arithmetic is a crossover. Let the annual cost of owning be F. Let your effective rent rate be r, a share of each unit of revenue or volume. Cumulative rent is r times your volume. The lines cross at the revenue level where r times volume equals F, roughly F divided by r. Below it, renting is cheaper and the licence would sit idle capital. Above it, you are paying more in rent than the licence would cost, and the sponsor is keeping the difference.
Founders run this once, at the sticker rate, and conclude renting stays cheaper for years. The first model already told you why that is wrong. The rate r is not fixed. Hold-up pushes it up over time, which lowers the crossover and pulls the buy decision toward you. Run the calculation twice: once at today’s rate, once at the rate the sponsor can charge when your switching cost is at its highest. The honest crossover sits between them, closer to the second.
Comparative statics on buy versus rent
Assumes: owning is a known fixed annual cost; renting scales with volume at a rate you can estimate.
Fits because: it produces one number a board can act on, the revenue level where owning wins.
Breaks when: the licence buys more than cost savings (control, a moat, a balance sheet a regulator trusts) that the pure arithmetic ignores.
Counteracts: optimism that rent stays flat; it forces the second, higher rate into the model.
May reinforce: the threshold model, which explains why the crossing, when it comes, is abrupt.
3. Threshold: the squeeze is a step, not a slope
A sponsor tolerates a small partner. You are little volume, little risk to their licence, and not worth the friction of repricing. That patience holds until you cross a line: your volume starts to show up in their own regulator’s questions, or your book grows large enough that the rent is worth taking, or a rule changes and carrying you now costs them. On the far side of that line the treatment flips in one step. The rent jumps, the rule tightens, or the arrangement ends.
Granovetter’s threshold logic is the shape: nothing moves, nothing moves, then everything moves at once. And the trigger is often not you. A regulator can flip an entire category in a single circular. Kenya’s central bank brought digital lenders that had operated freely under a rule that “a person shall not carry out digital credit business in Kenya unless licensed by the Bank,” and gave those already trading six months to apply.2 Uganda’s National Payment Systems Act made it an offence to “operate a payment system or offer payment services” without a Bank of Uganda licence, moving mobile money off partner-bank permission and onto direct licensing.3 Ghana’s Payment Systems and Services Act did the same for payment providers and electronic money.4 And in April 2024 Nigeria’s central bank told four fintechs to stop onboarding new customers, in the middle of a fraud sweep that froze 1,146 accounts and had little to do with those firms’ own conduct.8, 9 The category moved. Everyone renting inside it repriced overnight.
Threshold (Granovetter)
Assumes: tolerance holds up to a tipping point, then behaviour changes discontinuously.
Fits because: sponsor repricing and regulator action arrive as steps, not gradual drift you can plan around.
Breaks when: the sponsor reprices smoothly and often, giving you a slope you can track and pre-empt.
Counteracts: the crossover’s smoothness; the real crossing is a jump.
May reinforce: bargaining, since the step usually fires exactly when your switching cost peaks.
GEER: the moves, cheapest and most reversible first
Start with the move that costs a morning and commits you to nothing.
Read the two clauses that decide your future. Open the sponsor contract and find the repricing clause and the termination clause. Almost every founder on one of these arrangements has never read them closely. What notice can they give. What can they change unilaterally. What happens to customer funds if they walk. That reading is free and it is the whole ballgame.
Compute your switching cost, then your two crossovers. Estimate what leaving this sponsor would cost in rebuild, migration, and downtime. That number is the rent ceiling. Then run the buy-versus-rent crossover twice, at today’s rate and at the rate they could charge once your switching cost is high. You now hold a revenue trigger instead of a vague worry.
Lower the ceiling before you need to. A second rail, a second sponsor, or portable customer KYC cuts your switching cost, which cuts the maximum rent the first sponsor can extract. This costs real engineering, so it sits third, but it is the move that changes the bargain rather than just measuring it.
Start the licence clock early. Acquiring your own permission takes capital and months, and the regulator can move before you do. Beginning the application is costly and slow, which is why it comes last in order and must begin well before the crossover, not after.
RADAR: what to lock in before the next renewal
Do now (T+0 to T+14). Pull the contract. Extract the repricing and termination clauses in writing. Compute switching cost and both crossovers. These are reversible, they dominate in every scenario, and they cost you nothing but attention.
Hedge (T+14 to T+28). Stand up a second rail or a second sponsor even at low volume, purely as insurance against a repricing you cannot yet see. Pre-draft the licence application and ring-fence the capital line. Cheap relative to the tail it covers.
Defer and trigger. Filing for your own licence is close to irreversible once the capital is committed, so pre-commit the observable trigger instead of deciding under pressure. File when your annual rent approaches the annualised cost of owning, or when your volume nears the sponsor’s risk threshold, or the day a regulator opens a consultation on your category. Whichever comes first. Name the trigger now, while you are calm, because you will not be calm when it fires.
CHAIN: what has happened to everyone in this position
Match the reference class on structure, not on sector. The class is: a small operator running live on a larger entity’s permission, where that larger entity answers to its own regulator. Payment fintechs on a sponsor bank belong to it. So do agency-banking outfits on a bank’s licence, lending apps on a partner’s approval, and insurance sellers under a carrier’s authority.
The base rate inside that class is unkind. Take the documented regulator moves in this series: Kenya’s digital-credit licensing, Uganda’s payment-systems law, Ghana’s payment act, Nigeria’s onboarding freeze.2, 3, 4, 8 In each, the operator without its own licence absorbed the change, and absorbed it on the regulator’s clock, not its own. Add the private failures, where a sponsor’s own collapse froze funds the operator did not control.6 The arrangement repriced or ended in every case in view.
Subtract the counterfactual before you over-read that. Some of these firms needed capital and a licence regardless of any sponsor, so not all of the pain is the rental’s fault. What the rental specifically added was the timing: the loss of control over when the change lands. That, isolated, is the cost of renting rather than owning.
One thing would break this whole reading. If a credible regional passport emerged, a single licence honoured across markets, or a settlement layer that let a small operator hold its own thin permission cheaply, the fixed cost of owning would fall and the crossover would move out for everyone. Watch for it. It is not here yet, and planning as if it were is a bet on a rule that has not been written.
What this ensemble is blind to
The models price two rational parties and a rule-bound regulator. They cannot see a regulator acting for reasons that have nothing to do with you. Nigeria’s freeze rode in on a forex-fraud investigation; the fintechs told reporters the order looked misdirected, since most of the flagged accounts sat at commercial banks.8 The models also cannot price a sponsor that simply fails, the way Synapse did, taking your access down with it.6 And there is a fold the arithmetic misses: founders under-weight the repricing tail because a rail you use every day feels like infrastructure, not a counterparty with its own incentives. It is a counterparty.
Here is the action that survives all of that ignorance. Hold a second rail and start the licence clock on your trigger, whether or not you can forecast the specific event. You are not buying a prediction. You are buying the one thing renting took from you, which is control over the day the terms change. Decide that this quarter, before someone else decides it for you.
Sources and notes
- Benjamin Klein, Robert G. Crawford and Armen A. Alchian, “Vertical Integration, Appropriable Rents, and the Competitive Contracting Process,” Journal of Law and Economics, 1978. The paper’s terms are used here: “postcontractual opportunistic behavior” and the “appropriable” quasi-rents that make it possible. Note: the journal name and the word “Vertical” are preserved as the verbatim citation. Open PDF: https://www.edegan.com/pdfs/Klein%20Crawford%20Alchian%20(1978)%20-%20Vertical%20Integration%20Appropriable%20Rents%20and%20the%20Competitive%20Contracting%20Process.pdf
- Central Bank of Kenya (Digital Credit Providers) Regulations, 2022, Legal Notice 46. Regulation 4(1): “A person shall not establish or carry out digital credit business in Kenya … unless that person is licensed by the Bank.” Regulation 59(1) gives providers operating at commencement six months to apply. Primary PDF: https://www.centralbank.go.ke/wp-content/uploads/2022/03/L-.N.-No.-46-Central-Bank-of-Kenya-Digital-Credit-Providers-Regulations-2022.pdf ; human-readable mirror: https://new.kenyalaw.org/akn/ke/act/ln/2022/46/eng@2022-04-22
- National Payment Systems Act, 2020 (Uganda), Act No. 15 of 2020. Section 6 prohibits operating a payment system, issuing a payment instrument, or offering payment services without a licence; section 7 sets the application; Part IV governs electronic money issuance. Full text: https://media.ulii.org/media/legislation/18297/source_file/d1fd12f2a2195d1e/2020-15.pdf
- Payment Systems and Services Act, 2019 (Ghana), Act 987. Cited qualitatively for its licensing requirement over payment services and electronic money; the Bank of Ghana copy is an image-only scan, so no specific section number is quoted from it here. https://www.bog.gov.gh/wp-content/uploads/2019/08/Payment-Systems-and-Services-Act-2019-Act-987-.pdf
- “CBN licences in Nigeria,” TechCabal, 2025, reporting the Central Bank of Nigeria payment-licence categories and capital: Switching and Processing and Mobile Money Operator at 2 billion naira minimum capital each with matching escrow, Payment Solution Service Provider at 100 million; described as “minimum paid-up share capital … you must own.” https://techcabal.com/2025/05/13/cbn-licences-in-nigeria/
- “Synapse bankruptcy: Thousands of Americans see their savings vanish,” CNBC, 22 November 2024. Fintechs “which are not banks” reached accounts via the middleware layer; “the fintech middleman turned off access to a key system”; a trustee found “up to 96 million dollars of customer funds was missing”; one customer was offered 500 dollars against 280,000. https://www.cnbc.com/2024/11/22/synapse-bankruptcy-thousands-of-americans-see-their-savings-vanish.html
- “The spectacular Synapse collapse,” Fortune, 7 March 2025, reporting roughly 200 million dollars in customer money frozen after the failure. https://fortune.com/2025/03/07/synapse-evolve-mercury-bankruptcy-lawsuits/
- “CBN stops OPay, Palmpay, Kuda Bank, and Moniepoint from onboarding new customers,” Nairametrics, 29 April 2024. The directive halted new-customer onboarding pending notice; the EFCC had secured an order freezing at least 1,146 accounts in a forex investigation; the fintechs argued most flagged accounts were commercial-bank accounts. https://nairametrics.com/2024/04/29/forex-deals-cbn-stops-4-fintechs-from-onboarding-new-customers/
- “CBN stops Opay, Kuda, Palmpay, Moniepoint from boarding new customers,” The Guardian (Nigeria), 29 April 2024, confirming the directive to suspend onboarding. https://guardian.ng/news/cbn-stops-opay-kuda-palmpay-moniepoint-from-boarding-new-customers/