Estimates, at the time the Story Points are decided, may have a margin
of error of as much as 100% (well, actually the margin would be +/- 50%)*. That's why we say that all 3 point story
are roughly the same, even of one takes 1 day and other take 2 days. If
one 3 point story takes1 day and another take 5 days then something went
wrong, either in the estimation or in the execution.
- When estimating in hours, a buffer needs to be put at the end of the plan just in case. By estimating in Story Points, the buffer is created when setting the initial Velocity of the team, and then adjusted on each iteration.
- Speed of team members are different. A 3 points story can be implemented in 1 day or 2 days depending on the people involved.
A 3 person team saying "we can deliver 12 story points per 2 week iteration", is the same as if they were saying "we can deliver these features that are estimated for a total of 150 hours work in 80 hours of duration, leaving 90 hours of buffer in case that the estimates are wrong or some problem comes along the way".
At the end, estimating in Story Points and estimating in hours has exactly the same basis: Both depends on Gut Feeling. But it is easier to sell Velocity than to sell buffers and estimates with confidence range. And it is a lot easier to think in the size of the task ("hmm, this may involve changing 3 modules, but the change is simple") than to think in the time to execute it ("hmm, this may take about a week").