Where estimates go wrong

Bias, not arithmetic.

Systematic error

Random errors cancel; systematic ones do not. If every assumption was optimistic, the total is optimistic by their product, which can be an order of magnitude.

Check whether all your guesses lean the same way.
Anchoring on a number someone mentioned first is very hard to shake off.

Compounding uncertainty

Multiplying five figures each uncertain by a factor of two gives an answer uncertain by far more. Say so, rather than presenting the product as a fact.