THE SQL Server Blog Spot on the Web

Welcome to - The SQL Server blog spot on the web Sign in | |
in Search

Alberto Ferrari

Thinking in DAX: Counting Products in the Current Status with PowerPivot

One of my readers came to me with an interesting formula to compute in PowerPivot. Even if I don’t normally post about very specific scenarios, I think this time it is interesting to write a blog post since the formula can be easily created, if you think at it in DAX, while it is very hard if you are still approaching it with an MDX or SQL mindset. Thinking in DAX is something that comes after a lot of formula authoring, something that all BI professionals should strive for, as Vertipaq in the new SSAS engine is approaching pretty fast!

The scenario is pretty simple. A processing capability handles items (A, B, C) and, during the working process, the status of a product changes. Each time a product changes its status a new row is added to a logging table, along with a value, indicating the cost of the transformation. The raw data can be simulated with this table:


You can see, from the highlighted cells, that product A changed its status twice on the first of January, first it became an Apple, and then an Orange (don’t ask me why my reader decided to use fruits to model status, this overcomes my comprehension Sorriso but I still find it amusing, so I have not changed it).

Now, with this set of data, the requirement is to compute, at the end of each day, how many products are in a specific status. Thus, product A, at the end of the first of January, was an Orange, even if it has been an Apple for some hourse. Moreover, it should be noted that A keeps the status of Orange during the period from the first to the fifth of January, even if no more rows were added to the table. No new rows, in this case, means “keep the last status”.

When I received the workbook for some testing, it contained two tables: one pretty standard calendar table and the History table, as I have shown it in the previous picture. Moreover, a relationship was holding between the two tables, based on the date.

In such a scenario the relationship is not only incorrect, it is misleading. From the functional point of view, there is no relationship between the fact table and the calendar table. For example, on the second of January, there are no data in the fact table but all of the measures should be computed taking into account the dates before the current one. Thus, the first step I have done is to delete the relationship so that I will leverage DAX only to solve the scenario. PowerPivot will continue to ask you the permission to auto-detect relationships… disable that feature, we know that, in this scenario, we don’t want to leverage automatic relationships.

Now, to solve the scenario we should use a variation of the SCD handling formulas I provided in a previous post. In fact, the table looks very similar to a slowly changing dimension but, this time, we want to compute the last status of a product in a given point in time, where the point in time is not fixed but it changes following the filters imposed by the PivotTable. Said in other words, we are going to compute the SCD status of a product at a variable point in time.

Please stop one second and think twice at what I have said, since this is where “DAX thinking” come to play. In this data model we don’t have a Product dimension, nor we have any canonical SCD handling. Nevertheless, by leveraging the tremendous speed of the Vertipaq engine, we are going to create a sort of “running SCD” on the fly, during the formula computation. This kind of computation is something that you would never try with MDX or SQL, it is a feature that can be leveraged only with DAX, where heavy leaf-level computation is carried on at a blazing speed.

Now, it is time to show the final result and the formula, which I will describe in a few moments:


The result is straightforward: the grand total always shows three products (A, B and C) while the count at the status columns show the number of products that, at the end of that day, are in the given status. You can see that data is shown even for days where no fact table data exists, since the formula will leverage previous information to compute the SCD status at that day.

Now the formula:

        FILTER (
            HistoryT[Id] = 
                CALCULATE (
                    MAX (HistoryT[Id]), 
                    ALL (HistoryT),
                    HistoryT[ItemId] = EARLIER (HistoryT[ItemId]),
                    HistoryT[ShortDate] <= VALUES (Dates[Date])
    LASTDATE (Dates[Date])

Let us spend some words to describe its behavior.

First of all, we use CALCULATE to introduce a filter context that shows, from the calendar table, only the last available date, using LASDTATE. This is useful if the PivotTable does not filter a single date but a range of them (as I am doing with the slicer, that filters only the first fifteen days of the month). In that case, we use the last available date of the selected dates set to perform computation.

Inside that filter context, we use the FILTER function to recreate the SCD status at that date. Let us see how we manage this:

We FILTER the History table removing all the lines that have an ID earlier than the last ID for that specific product, using an inner CALCULATE. Let us check the filters inside the inner CALCULATE:

  • ALL(HistoryT) is necessary since in the current filter context, only the products with a specific status are shown. For example, during computation of <1/1/2020, Apple>, only products with “Apple” status are visible, while we want to check if the product has changed status. Thus, the ALL filter removes the condition on the current status, showing all the products.
  • HistoryT[ItemId] = EARLIER (HistoryT[ItemId]) filters all the instances of the current product, where “current” means the current iteration of the row context introduced by FILTER.
  • HistoryT[ShortDate] <= VALUES (Dates[Date]) filters all the rows that comes before the current calendar table.

Now, it is important to remember that these conditions are ANDed together and the iteration of FILTER produces a table which is, at its end, the current SCD status of the History table at the last date in the calendar selection. It is enough to count the number of rows of this filter to get the number of products that are in the selected status. Always remember that the “selected status” is injected in the inner CALCULATE due to the presence of a filter context on the status which has not been removed by the inner CALCULATE and, thus, it is yet working, filtering only the rows of the SCD that have the correct status.

Working directly with the fact table this formula might not be a top performer for a really huge table, if we need to improve performance, then some more data modeling should be applied, normalizing the product table or performing similar tasks. Nevertheless, what is important to note here is that DAX let you perform very complex computation without any ETL nor modeling, just plain DAX.

If you think at how to solve the same scenario with MDX and… well, I would not even try it, UDM is great but it has a rigid structure and this scenario is not easy to solve without a deep data modeling change. On the other hand, DAX lets you shape the data on the fly and provides a very elegant and neat solution to a complex scenario like this one. The only problem is that, to solve the scenario, you need to think in DAX.

This is just another example of what PowerPivot can do for your reporting needs. If you want to touch the real Karma of PowerPivot, don’t miss the opportunity to follow one of the workshops I and Marco Russo are bringing all over Europe in the next few months.

You can find more info on Hope to see you there!

Published Tuesday, March 8, 2011 9:00 AM by AlbertoFerrari
Filed under:



SQLBI - Marco Russo said:

Last week Alberto published an interesting post about Counting Products in the Current Status with PowerPivot

March 17, 2011 11:46 AM

SQLBI - Marco Russo said:

Chris Webb posted a wonderful solution to the Last “ever” non empty calculation in MDX . I never thought

March 27, 2011 5:49 PM

Oleksandr said:

I wish there were analogy in MDX. I'm struggling with this problem now, need to make a report with MDX.

December 14, 2012 10:53 AM

John Bradley said:


We are looking to implement something similar to this concept but where we have a table that is using percentages to get a payout for our partners against the total for sales.  I have tried to use values instead of countrows to bring in the percentage.  If you had to use the "value" from your HistoryT table instead of trying to just get the count of rows how would you approach that.  The actual problem is always more difficult then that but that is what we are trying to solve as simply stated as possible.

Kind Regards,

July 23, 2013 5:42 PM

AlbertoFerrari said:

John, I would need a sample workbook to look at, it is the easiest way to dive into a formula. Please drop me an email, I will look at it as soon as I have some time.

July 23, 2013 6:13 PM

John Bradley said:

Hello Alberto,

I sent the workbook to what I assumed was your email but wasn't certain.  It is coming from jbradley at nmrdirect com.  Please let me know if you haven't gotten and I will attempt to resend.

Thanks in advance for any help that you can provide.

Kind Regards,

July 30, 2013 12:36 PM

Yong said:

I am confused why “selected status” is injected in the inner CALCULATE because inner CALCULATE has ALL to remove any filter on status.

January 9, 2014 12:58 AM

AlbertoFerrari said:

@Yong, the ALL comes with other condition, the most relevant of which is HistoryT[ItemId] = EARLIER (HistoryT[ItemId], which filters only ItemId which come from the outer filter context introduced by FILTER. Because FILTER filters HistoryT, in the outer filter context, it feels the selected status coming from the slicer, row or columns.

January 9, 2014 2:00 AM

Chris said:

Hi, this is a great example... but i'm really struggling to replace the countrows function with a sum command so that at an point into the future we can see the last known value of product by state..

Can you offer any pointers?

July 28, 2014 12:55 PM

Ty said:

"Hi, this is a great example... but i'm really struggling to replace the countrows function with a sum command so that at an point into the future we can see the last known value of product by state..

Can you offer any pointers?"

Same here... any pointers Alberto?

January 26, 2015 10:08 AM

AlbertoFerrari said:

Ty, Chris... I honestly don't fully understand the specs of your request. Can you provide me an example of the query you want to run? "Replacing COUNT with SUM" does not help me...

January 26, 2015 10:25 AM

Ben said:


I'm able to get a closing balance total for a unique date using the following but it only works if I select a unique year.

Any thoughts on how to make it work so that I can add year as a row attribute so that it shows the running closing balance for each year?

Hi Chris, I did find a DAX solution (see below) but it only works for a unique date filter. Any thoughts on how to make it work with a date (or date hierarchy) as a row attribute in a pivot?





   ,"LastEverNonEmptyDate", CALCULATE( MAX( vwCPAccountRevenueRecognition[RecognitionDate] )

                                       , DimDate[CalendarYear] <= VALUES( DimDate[CalendarYear]) )


   , "EndingDeferredIncomeCP", CALCULATE ( SUM( vwCPAccountRevenueRecognition[EndingDeferredIncomeCP] )

                       , FILTER(ALL(Dimdate), DimDate[FullDate] = [LastEverNonEmptyDate] ) )


, [EndingDeferredIncomeCP] )



May 6, 2015 1:27 PM

Aleksandr said:

Can anyone translate that DAX formula into MDX to get the same results but in Multidimensional model?

February 11, 2016 12:12 PM
New Comments to this post are disabled

About AlbertoFerrari

Alberto Ferrari is a Business Intelligence consultant. He his interests lie in two main areas: BI development lifecycle methodologies and performance tuning of ETL and SQL code. His main activities are with SSIS and SSAS for the banking, manufacturing and statistical sectors. He is also a speaker in international conferences like European PASS Conference and PASS Summit.
Privacy Statement