Since the second and third categories add up to the third category, you probably dont want a stacked column chart.
Im also going to make the (quite possibly wrong) assumption that you want time to be the X-axis of such a chart.
1) step 1 -- define the search you need.
You'll need to ultimately get a search that gives you results where the columns are
time | # total visitors | # attempting b | # completing b
1a) get a search that has all users doing everything.
1b) if your rows do not each contain all info to determine whether the action was completed or merely initiated, you may have to use stats or transaction to get yourself to that point. (where all rows have complete information)
1c) eval yourself a field called like 'type', just on the rows that match 'action=b'. You'll want to eval one version if the rows have terms indicating it was completed, and other rows where it was not completed (see http://answers.splunk.com/questions/7144/graphing-multiple-data-sources-in-one-chart/7162#7162 for a similar solution)
1d) <your search> | timechart count by type
2) that's it really. Once you get the search language giving you that table it's really just a non-stacked column chart that you probably want. Or if you modify 1c slightly, you can make it so you have 1 category for 'attempted b', one for 'completed b', and one other for 'everyone else'. At that point they would sum up to the total so you could use a stacked column chart and it would make sense.
hth
