So far PowerPivot does not support many to many relationships. It is still early days and I am sure that things will change soon. Moreover, there are a few ways to implement many-to-many relationships for measures in DAX. Two notable posts on the subject can be found on Marco Russo’s and Alberto Ferrari’s blogs:
Unfortunately, if we have multiple measures, the approach is somewhat cumbersome as we need to build numerous DAX measures. I have been playing around with modelling options and I just came across an idea which, while not perfect, can help us to mitigate to some extent the lack of this crucial bit of functionality. In fact, “not perfect” is quite an understatement – the more appropriate expression would be “a hack”, but it could still be useful in some scenarios – hence this post!
To put some background, let’s assume we have this simplistic model:
An account table
A customer table
And a fact table
Here we have two accounts, A with an Amount of 100 and B with 50.
Now, let’s assume we have the following mapping table between Accounts and Customers:
Account A maps to customers X and Y; Account B to X only.
Now, instead of trying to directly map the relationships here, we add the Amount to the mapping (bridge) table:
And we then create our relationships in PowerPivot. The problem here is that when we slice by Account A (with a key of 1), we would get the 100 Amount doubled. If we had more than two customers mapped to the account, it could have been tripled, or in fact multiplied by N, where N is the number of mapped customers.
The trick I am offering is adding a negative amount to this impromptu fact table against the same account (A) and against a Z customer with a key of -1, which acts as an unknown/adjustment bucket. Therefore, the Customer table would become:
And our fact/bridge table:
Now, if we create relationships between this table, the account and customer tables we would see the following:
Well, we do get one extra row with negative amounts, but all else seems just fine. The total is correct and the amounts per account and customer are also correct. If we have more than one measure, we will have to negate each of the additional measures as well if we want to achieve the same behaviour.
I also created a slightly more complicated model, which included years:
Here we have to compensate for two rows with negative amounts but the experience of using this model is not much different than the previous:
Some other common scenarios could also work quite well. I have tried with two M2M relationships and while the number of rows gets progressively larger, the model still seems to work. Another common scenario would be to have more than 2 attributes in either of the tables (whoa!), and that seems to work very well, too. The key is to remember to negate/adjust for each mapping after the first one. If we had a third node in our X,Y table mapped to A, we would have two additional rows in the fact table – one with a positive and one with a negative amount.
Unfortunately I have not come up with a way to simplify the data modelling through DAX, so at this stage I would do it in the source. However, it should not be too difficult to achieve this in SQL, for example.
I would be interested to see if anyone comes up with scenarios where this could not work and cases in DAX where the additional rows are a problem (beyond the “too much data”, of course).