Cheap Math, Expensive Questions

Cheap arithmetic changes which work is scarce. When the marginal cost of another financial model collapses to a rounding error, the analyst's defensible value stops being the math and starts being the question, the assumption surfaced, and the meaning of the answer carried into a decision. The bottleneck migrates from constructing the calculation to framing the problem.

PYMNTS makes the shift legible with a stress test any FP&A team would recognize: a 6% unit volume drop, a 4% input cost rise, one underperforming geography, and a two-quarter hiring delay. Run that once and you get a single answer. Run three and you get a conversation. Run three hundred and you get a map of the actual decision surface, the one that used to be too expensive to draw.

First-wave generative AI in finance preserved the existing division of labor: humans asked, machines summarized. The PYMNTS piece frames the next wave as boundary-disturbing because the machine is now constructing the analytical structure itself. The reframe is not that analysts are replaced. It is that the question becomes the scarce input, with 77.9% of CFOs already ranking the cash-flow cycle as very or extremely important to strategy.

The honest caveat: the PYMNTS report notes the arguments were generated by the model, then prepared into manuscripts with human involvement and formalized into Lean. A formal certificate is not yet an EBITDA model that survives an audit committee. The divide in finance will run between teams that treat the question as their product and teams that still treat the spreadsheet as their product.