Data & calculation: 20 ways AI gets it wrong
Arithmetic errors, confused figures and false precision in the numbers. Each failure mode below is phrased as the question people actually ask, with what it looks like in real work — and the layer of the Trust Stack that catches it.
Does AI make basic arithmetic mistakes?
A simple sum in a financial model comes out wrong.
Why does AI use the wrong formula?
A calculation uses a formula that doesn't fit what's being measured.
Can AI misread a table of figures?
A value is pulled from the wrong row or column of a table.
Does AI misinterpret charts and graphs?
A trend in a chart is read backwards from what it actually shows.
Why does AI confuse gross and net figures?
A gross number is reported where the net figure was meant.
Can AI mix up pre-tax and post-tax numbers?
A return is quoted pre-tax in a place that needed the after-tax figure.
Does AI confuse annual and monthly figures?
A monthly cost is treated as an annual one, off by twelvefold.
Why does AI confuse nominal and real returns?
A nominal return is presented as if it already accounted for inflation.
Can AI confuse percentage points with percentages?
A move from 5% to 6% is described as a 1% rise instead of a 1-point rise.
Does AI hide the calculations behind a number?
A final figure appears with no working shown to check it.
Why does AI use unrealistic assumptions in a model?
A projection assumes growth or costs no real business would expect.
Can AI leave fees out of a calculation?
A net return is computed as if fees didn't exist.
Does AI ignore taxes in financial figures?
A take-home figure is shown with no tax deducted.
Why does AI ignore inflation in projections?
A long-term forecast treats future money as worth the same as today's.
Can AI ignore currency risk?
A cross-border return is quoted as if exchange rates never moved.
Does AI ignore the downside scenarios?
A forecast shows only the upside case and skips what happens if it goes wrong.
Why does AI round numbers too aggressively?
Heavy rounding turns a precise figure into a misleading one.
Can AI give false precision in its numbers?
A rough estimate is reported to several decimal places it can't support.
Does AI fail to reconcile numbers that don't match?
Two figures in the same document contradict each other and neither is questioned.
Why doesn't AI flag the inputs it's missing?
A calculation proceeds with a key input simply assumed rather than asked for.
More from the library
One model can’t reliably catch its own mistakes. A council of independent minds can.
Run your work through the councilAll 250 failure modes · See also: the Trust Stack