Executive teams tend to price a platform conversion like a software upgrade. It has more in common with a transplant. Everything has to come out of one body and go into another, intact, and the patient has to be reporting to a regulator on Monday morning.

The target design is rarely what goes wrong. Vendors are good at target designs. What goes wrong is the history — the two decades of records that have to arrive whole, and the handful of them that were never quite standard to begin with.

Three areas account for most of the overrun we get called in to rescue.

Corporate actions recorded creatively

Splits, spin-offs, bond amortization and anything with a bespoke accounting treatment agreed years ago. Legacy platforms often store these using internal logic that made sense at the time and was never documented, because the system writing the record was also the system reading it. A modern target does not share the assumption. If you are lucky the ingestion fails loudly. If you are unlucky it succeeds, and the position is wrong by an amount too small to notice until an auditor notices it.

Instruments the migration tooling has never met

Off-balance-sheet exposures, custom OTC swaps, private placements, bank loans. These need precise pricing rules and complete accrued-interest history, and off-the-shelf conversion tooling handles the standard ninety-five per cent beautifully. The remaining five per cent is where the money sits, where the audit risk sits, and where the person who understands it is one of about four people in the building.

Generic technical resource on domain-specific work

A good engineer understands the code. Trade-date versus settlement-date accounting is not a coding question. Neither is amortized cost versus fair value, nor which of two defensible treatments your statutory reporting has assumed since 2009. The gap does not surface in testing, because the tests were written by the same people who did not know to ask.

What actually reduces the risk

Count the awkward things first — before the plan, before the timeline, and specifically before anyone says the word straightforward in front of a steering committee.

It is an unpopular phase. It produces a list of small, unheroic, slightly embarrassing findings that make the project look worse on day one. It also makes it look dramatically better on day ninety, which is the only day anybody remembers.

None of these are hard to fix eleven months out. All of them are brutal eleven weeks out, when the budget, the plan and somebody's credibility are all pointing at a date that will not move.