Answers
The questions that actually come up in a brief. Each one answered in the first sentence, with the reasoning underneath. No preamble, no gated download.
Architecture
- Should the ledger be double-entry if we only hold e-money?Yes. E-money still has to reconcile to a safeguarding account, and single-entry balances cannot prove it. Why the shortcut costs more than it saves.
- Can you migrate core banking without downtime?Yes, but the technique is dual-running with the old system authoritative until the new one has proven itself on live traffic. Not a cutover weekend.
- What is a sub-ledger and when do you need one?A sub-ledger records movements at a finer grain than the general ledger, and you need one as soon as a single GL account has to answer questions about individual customers.
- Where should idempotency keys live in a payment flow?At the boundary where the caller first commits to an intent, carried unchanged through every downstream hop. Not generated per service.
- How many chains should a stablecoin be issued on?As few as your distribution genuinely requires. Each additional chain multiplies the reserve, redemption and attestation surface without multiplying demand.
- What breaks first when a payment system scales?Reconciliation, almost always. Before throughput, before the database. The volume that exposes it is the volume of exceptions, not of payments.
Build vs buy
- BaaS or your own licence. How do you actually choose?Choose BaaS when speed matters more than margin and you can accept a partner's risk appetite as your own. Choose your own licence when the economics need the deposits.
- Should we build our own KYC orchestration?Build the orchestration, buy the checks. The vendors are commodities; the decision logic and audit trail are yours and change constantly.
- When does a payment orchestrator stop paying for itself?When routing decisions become a competitive advantage, not a convenience. Usually when the savings from your own routing exceed the orchestrator's fee.
- Is an off-the-shelf card processor enough for launch?Usually yes for launch, and the question that matters is what happens at scale. Specifically whether you can access raw authorisation data and change processors.
- What does “cloud-native core banking” actually mean?Vendors use it for anything from a mainframe with an API to a genuinely horizontally-scalable ledger. Four questions separate them.
Compliance
- What does it mean for screening to “fail closed”?That a transfer is refused when the screening system is unavailable or uncertain, rather than allowed through and reviewed afterwards.
- Who is liable when a sponsor bank offboards you?You are, to your customers. The sponsor's exit does not transfer your obligations. What determines the damage is what you hold and can act on independently.
- How often should stablecoin reserves be attested?As often as the reserve position can be derived without human assembly. Which should be continuously, with the publication cadence a policy choice, not an operational ceiling.
- What does SOC 2 actually cover. And not?It attests that you followed your own stated controls over a period. It says nothing about whether those controls were the right ones.
Programme
- Why do neobank cost estimates vary so much?Because “neobank” describes at least four different products, and the cost difference between them is structural rather than incremental.
- What is the minimum viable team to launch a neobank?Smaller than most programmes run, and the constraint is which competencies are present rather than headcount.
- What actually determines time to first customer?The licence or partner approval, in almost every case. Not engineering. Which changes what is worth accelerating.
Bring us the hard part.
Forty-five minutes with the people who would actually run the build.