Because the word describes at least four different products that happen to present a similar app to the customer, and the difference between them is structural rather than incremental. A single-market product on a sponsor's licence with a bought core and no cards, and a licensed multi-jurisdiction bank with an owned ledger and a card programme, are not the same project at different budgets. They are different projects. Any estimate given before those choices are settled is pricing something other than what will be built.
The four decisions that move the number most are the licence model, whether the core ledger is built or bought, whether card issuing is in scope for launch, and how many jurisdictions you enter at once. Everything else, meaning team composition, technology stack and location, matters far less than any one of these.
This is why quotes diverge so wildly between firms. They are frequently answering different questions in good faith, having each assumed a different set of the four. Comparing them is only meaningful once the assumptions are stated, and asking each firm to state theirs is the fastest way to make the comparison honest.
The estimate becomes reliable at the point those decisions are made, not before, which is what a discovery phase is actually for. A few weeks spent settling them changes the accuracy of the number more than any amount of estimation technique applied to an undecided scope.
Treat a confident price given before that as information about the firm rather than about the project. It usually means they are pricing the version they know how to build.