We rebuilt Helix’s treasury platform in twelve weeks after two other teams had quoted eighteen months. The rebuild worked — $180M monthly flow, 40ms median API, 99.99% uptime a year later. This is the full ledger of what that speed cost, because "we did it in twelve weeks" is marketing until you show the bill.
What we cut
We cut the admin UI (operations ran on SQL and a spreadsheet export for the first month), multi-currency display formatting (numbers first, polish later), and every nice-to-have integration except the two that moved money. We also cut meetings: one weekly demo, one written update, no standing ceremonies.
What bit back
The deferred admin UI cost us: operations found a reconciliation edge case in week nine that a proper UI would have surfaced in week five. And our gradual-cutover feature flags, brilliant during migration, lingered for months afterwards as dead code nobody dared delete.
The decision we would undo
We matched the legacy system’s ledger semantics exactly to de-risk the migration, planning to "fix the model later". A year on, that compatibility layer is still there. When you have a twelve-week window and the choice between compatible-but-crooked and clean-but-riskier, we now argue harder for clean — later never comes.