In "variable dimension" mode, the app returns a set of results among many more possible options. This set of "chosen" results might not be the ones you expect, or that are meaningful to you. You can modify the results by changing the parameters.
Here is a set of troubleshooting tips you might find interesting.
DROP NON RELEVANT DIMENSION FROM RESULTS
In this example Ward Kermit, a buyer, seems to be the principal "cause" of the loss of sales. In fact the "Buyer" column is probably not a good "explainer" of the change of sales. In this specific case, Mr. Ward simply left the company at the beginning of the year, but the Plan was not corrected to take that into account. So Mr. Ward appears to explain a large part of the difference between Plan and Actual.
Other than correcting the Plan and substituting the new buyer's name to Mr. Ward, the best course of action here is to simply drop the non-significant column from the results.
This will result in a markedly different, and hopefully more insightful set of results that do not have the "Buyer" column.
SHOW DIMENSION IN EVERY ROW
Sometimes, for clarity, we need a given dimension to show up in every result row.
In this same example, it might be interesting to find out in which Store Type the Womens category is having trouble.
This widget forces a given dimension to appear in every row.
This result is returned.
If it is not, the best option is to play along with the parameters to tweak the result. For instance, we can try returning an alternative result row with this widget...
...and see if we get something better.
In this case we get result rows with many dimension, but that have a low variance value. As a consequence the balance value after five results is very high.
Please note that all these different results are computationally correct and show different sets of facts. Setting the widget to a different value...
...will return yet a different result, in this case with similar characteristics.
RESULTS DO NOT CONVERGE TO TOTAL VARIANCE
Sometimes you might get something like this. The result items are too small and therefore do not converge to the total.
The app is choosing too small combinations and therefore it is converging too slowly to the total variance.
One simple fix could be to increase the minimum accepted result size value.
This will partially solve the problem and return a set of larger results that converge more quickly.
Since the issue here was that the app is returning lots of low value results, and not giving enough weight to the amount, another approach could be to change the parameter aggregation algorithm and choose an algorithm that "over-weighs" variance amount.
Now the app returns this set of results which is definitively more weighted on their amount.