My two pennies...and on a similar theme to the others:
Pivot Tables IMO should be the first thing anyone learns -- if nothing else it teaches you the value of laying out your data correctly in the first instance whilst not requiring a great in depth knowledge of XL formulae.
All of the following:
VLOOKUP (& HLOOKUP), INDEX, MATCH
The latter two in particular.
With any formula try to make a point of knowing which is best when and moreover why ... ie why would you opt for one over the other in a given scenario.
Demonstrating an understanding of the mechanics of a function is one thing but understanding why it's the optimum approach is another ball game altogether and in the world of big financial models of immense value... I'm still learning myself but having any/some grasp is better than none
In Finance knowing how to work with Dates is of obvious importance -- learn the basic functions but also make a point of knowing the functions available to you via the Analysis ToolPak -- the likes of EDATE, NETWORKDAYS, WORKDAY, EOMONTH etc... can save you a lot of time and head scratching but obviously try to have an alternate approach available should the ToolPak not be a viable option.
Again related to big models -- try to get to grips with Volatile functions - these can have a serious impact on performance -- examples of Volatiles being things such as: INDIRECT,OFFSET,TODAY(),NOW() etc... I've seen lots of BIG models with lots of volatiles... they're slow to calculate for a reason
And if you still have the will to live... Arrays & Sumproduct ... though often where you require lots of Sumproducts a Pivot Table is invariably your better option