In short
What a neobank costs to build is set by four decisions, not by engineering hours: whether you hold your own licence or operate as an agent on someone else's, whether the core ledger is built or bought, whether card issuing is in scope for launch, and how many jurisdictions you enter at once. A single-market product on a sponsor's licence with a bought core sits an order of magnitude below a licensed multi-jurisdiction build with an owned ledger and a card programme. Any quote produced before those four are settled is pricing a different project from the one that will actually be built.
Why the range is so wide
Ask five firms what a neobank costs and the answers will span two orders of magnitude. That is not evasion or padding. It is that "neobank" describes at least four different products which happen to present a similar app to the customer, and the cost difference between them is structural rather than incremental.
The useful move is to stop asking for a number and start settling the four decisions that produce one. Each is a fork, not a dial: taking the left branch does not make the programme somewhat cheaper, it makes it a different programme with a different team, a different timeline and a different risk profile.
Decision one: your licence, or someone else's
Operating as an agent or distributor on a sponsor's licence removes the single largest cost and the single longest lead time from the programme. It also means you inherit their risk appetite, their compliance decisions and their product boundaries. When a sponsor declines a feature, that feature does not ship.
Holding your own licence inverts every one of those. You carry the application cost, the regulatory capital, the compliance function and the timeline. And in exchange the economics change shape, because you hold the deposits and the interchange, not a share of them. The decision is rarely about what you can afford; it is about whether your business model works at all without owning the regulated layer.
The trap is treating this as reversible. Migrating from a sponsor to your own licence later is viable, but only if the ledger was built to be portable from the first line. Teams that skip that discipline discover in year three that their customer balances live in a structure they do not control.
Decision two: the core ledger
The core ledger is where the largest silent cost sits, because it is usually chosen on the wrong criteria. Selection tends to optimise for integration speed and vendor support, which are project concerns. The ledger is not a project concern. It encodes the operating model of the bank, and every product decision for the following decade inherits its assumptions about postings, limits, holds and reconciliation.
A bought core is faster to stand up and slower to change. A built core is the reverse, and it is only the right answer when your product needs something the market does not sell. Unusual settlement timing, multi-asset balances, an embedded compliance path. If your product is a conventional current account, building a ledger is an expensive way to arrive at what you could have licensed.
The question that separates the two is not "which is cheaper" but "how often will we need to change how money is recorded". Teams that cannot answer that have not yet designed the product.
Decision three: cards at launch
Card issuing is frequently treated as a feature and priced as one. It is closer to a second programme running in parallel: scheme certification, a processor relationship, BIN sponsorship, dispute and chargeback handling, fraud controls, physical fulfilment, and a settlement path that has to reconcile against the ledger every single day.
It is also, for most consumer propositions, the thing customers actually came for. The decision is genuinely difficult, and the honest framing is that launching with cards roughly doubles the scope of a first release while significantly improving its odds of mattering. What it should never be is a line item discovered in month five.
Decision four: how many jurisdictions
Each additional jurisdiction adds a licence or a passporting argument, a local compliance regime, local payment rails, local reporting, and a set of product rules that will contradict the ones already built. Two markets is not twice one market; it is the point at which every assumption baked into the first has to be re-examined and generalised.
The cheapest version of a multi-market bank is one designed as multi-market from the first architecture and launched in a single country. The most expensive is one built for a single country and extended under commercial pressure two years later.
What to do with this
Before asking for a price, write down your four answers. If any of them is genuinely undecided, that is the work to do first, and it is a matter of weeks rather than months. A discovery phase exists precisely to close them.
A firm that quotes confidently without them is not being decisive. It is pricing the version of the programme that is easiest to sell, and the gap between that and the one you needed shows up as change requests somewhere around month seven.