3 Replies Latest reply on Jan 24, 2015 10:11 AM by Russell Spangler

    Forecasting Topic Idea

    Steve Kline

      I didn't get to attend the last meeting, but I'm interested in meeting with others to help me ramp up my Tableau skills (I'm pretty much brand new to it). I have a lot of economic information I want to report on, particularly for forecasting machine tool sales. I am also using Tableau for my company's sales data.

       

      One item I would be interested in is using historical trends in my company's sales data to forecast the current year. For example, I know the historic pattern by which we book advertising for the upcoming/current year. We track the day the order was booked and when that order will be delivered (e.g. the January 2015 issue of one of our magazines or a banner for February on of our websites). I've created a chart that compares years with the running total of advertising booked based on how many days prior to the end of the year that amount was booked. Sorry, it's hard for me to explain this in words. Hopefully the picture below helps.

       

      In Excel, I used to use the percent of the year total booked as of a given date to forecast the current year total projection. But, it was hard to do this for anything of the total company since there were too many other variables to manipulate (by brand, by product, by sales person, etc.). If I could get this set up in Tableau, then I could simply filter filter on the various dimensions to see where each one currently stands relative to previous years as well as projected total compared to budget. Another interesting item to be added to the viz is a list of the top 10 or 20 customers that advertised in previous years but we don't have a schedule for the current year yet.

       

      Booked Space - Running Total.jpg

        • 1. Re: Forecasting Topic Idea
          Steve Kline

          Russel responded with the following via email:

           

          In general Tableau does do forecasting:

          http://onlinehelp.tableausoftware.com/current/pro/online/mac/en-us/forecasting.html

          However I do not believe it is very robust, so a lot of people probably leverage R/SAS/SPSS to build their models.  It doesn't sound like you have a very complex model so Tableau might fit your needs.

           

          To be able to filter by the dimensions you are describing you'll need to create a more granular model.  For example: If I create a Monthly Forecast, I can show numbers via. Months/Quarter/or Year, however I lack the ability to do Days (I lack the granularity in my model, I cannot drill down).  The trick is to figure out the most granular level of data you have and build the model from there, that way you can roll up to all the views you want.  So maybe I create a Hourly forecast, I can roll-up hours to Days->Weeks->Months->etc.  Depending on the amount of data you have, that could be a wise decision or a bad decision (how much confidence do you place in your data, as you get more granular the amount of data in each forecast is smaller). 

          • 2. Re: Forecasting Topic Idea
            Steve Kline

            Russell:

             

            I have the data by all the dimensions - sales person, brand, product (3 levels), property, date (both insertion and delivery), company, etc. I guess our challenge is how to do the calculations. I'm struggling putting them together in Tableau, not because the calculation itself is hard if I just wrote it down, but the logic because I'm not familiar with SQL, etc. On the other hand, our IT director said we wouldn't be able to do the calculations outside of Tableau because they would have to be done for every possible permutation to allow Tableau to filter on them. He thought we would have to cube our data, which our developers are not experience with.

             

            Anyway, I'll be interested to meet everyone at the next meeting. Hopefully I'll be in town as I travel a lot.

            • 3. Re: Forecasting Topic Idea
              Russell Spangler

              Awesome, if you can share a sample of the data, please mask any information and it doesn't have to be all the dimensions, that'll be great.

               

              I'll be happy to help try and build the forecast with you.  If the calculation is just % of Sales from Previous year (on the given current day) compared to % of Sales so far this year (on the current day), then that shouldn't be too difficult.  If its something different please send me the details, or even send me an excel sheet with the formula you are using.   You can post it, or send it directly to my email account (listed on the event pages).

               

              I can help to either code solutions in SQL or Tableau (and I can help build cubes in SSAS), either way the calculation has to take place somewhere at some point in time.  If you want it down to each unique value of your dimensions, then the price to pay is the amount of calculations that need to take place to provide that granularity.  With that said a lot of people either aggregate values and create buckets of data or remove any variables that do not influence the model or only reduce their models down to where it makes sense (having only 100K records to model with using 1K variables isn't great, having 100M records with 1K variables is better (but it all depends)).  If you can calculate the number of unique values per dimension and multiple them together that'll help me understand the amount of data we are talking about. 

               

              If you want to wait till the next event that is okay too and I'll be happy to talk before or after the event.