Something may be frustratingly slow to do at runtime. Say, distinct count of some field over a large dataset.
If there are a finite number of such calculations you may possibly need to do, or there are infinitely many but some are more likely to be requested than others, then you can pre-compute all (or most) of the ones that may be needed.
Pre-compute for days and days, weeks. Let your machine pre-compute like there is no tomorrow.
To the user, it happened instantly.