8 Replies Latest reply on Jan 28, 2014 12:01 PM by . Tableautester

    Forecasting - how to?

      Hi everyone,


      I am new to Tableau and I really love it - I am querying my dummy DB (sales data) with it and it is great so far.


      However, there is one point which I cannot really understand as of yet: how do you forecast using Tableau?


      So for example I have a data sets which corresponds of all days in May with number of items sold in May.


      Now I apply a trendline and can visually see: sales are increasing, yes! Of course, thats beautiful, but for example to order more stuff from my supplier, I would like to know how much more I will sell in the next x days.


      Or the other way around: I have 1000 units on stock. And based on the sales so far + the trend, how many days of stock do I have left.


      Is it possible to answer these questions with Tableau?




        • 1. Re: Forecasting - how to?
          Richard Leeke

          Good question!


          The short answer is that there's no easy, out-of-the-box, point-and-click answer for you - but with a bit of effort there are things that you can do.  Actually probably quite a lot of effort.  ;-)


          Your question hints at two classes of requirements, one of which is a lot easier than the other.


          1) "Based on historical data, when will I run out" or "Based on historical data, how much will I sell in the next N days" type questions need some sort of trend model, from which you want to calculate a single number.


          You can't get at the results of Tableau's trend models to do further calculations, but as long as you know the maths for the trend model you want to use, that should be possible using table calculations.  There is an active thread on this topic here.


          2) I don't think this is specifically what you were asking, but sometimes you may want to project results into the future and display the next N periods (say) visually.


          That would require injecting additional rows into your data set, and arranging for those rows to contain the projected values.


          That is a complex topic and not one I've seen discussed much.  There are various possible approaches, including doing the projection in your back-end data source if it supports it.  Alex Blakemore posted about the possibility of doing that with the Oracle Model clause, here.  (That thread is very hard to read now - the order of postings got shuffled when the forum was revamped a while back.)


          This thread also discusses similar topics.

          • 2. Re: Forecasting - how to?

            Hi Richard,


            thanks for pointing me into the right directions, I will read through the threads later on.


            So what I can tell by my impression so far is that Tableau is a great tool to display historical data and analyse the past but does not go into the predictive part, right?



            • 3. Re: Forecasting - how to?
              Joe Mako

              With table calcs and parameters, lots of interesting things are possible, and if you can provide sample data and the formula you want want to apply, an example workbook can be made.

              • 4. Re: Forecasting - how to?
                Alex Blakemore
                Peter, Tableau is at its best when doing interactive exploratory data analysis on an existing data set, but you can definitely use Tableau as part of a predictive modeling toolsuite. One approach is to use Tableau to visually explore predictive data generated from another tool. If your predictive model is very sophisticated, that might be the right approach. Tableau does have some features useful for doing basic predictive modeling. It can build linear and polynomial trend lines from your data for example. You could also try using Oracle's model queries with Tableau's custom SQL if your predictive model can be expressed in Oracle's model SQL extensions. Here are some current Tableau features to learn about that can be useful when doing predictive analysis visually
                • trend lines -- making linear and non-linear models to fit data
                • parameters -- to allow the user to vary interactively
                • calculated fields -- for calculations
                • reference lines and drop lines -- for seeing where the trend and data points cross thresholds
                • Take a look at the Sales sample workbook. The Hospital dashboard shows off trend lines and the Sales Forecast and Sales Planning show off using parameters and calculated fields for simple prediction. If your predictive model is relatively straightforward, you could try to define it using calculated fields and parameters. If you do that, you might find yourself needing to synthesize some dummy future data points in your data to give your calculated fields some rows with future dates to build upon.
                • 5. Re: Forecasting - how to?
                  Hassim423 pal


                  We facilitate the provision of independent analysis to support expert testimony, regulatory or legislative engagements. Frequently, this work includes economic, financial and statistical studies of varying data analysis, technical and http://www.potentiamed.com

                  • 6. Re: Forecasting - how to?

                    I would like to bring this tread back because I am trying to get a forecast for the next month of data and need help.  I have attached a peice of sample data; the superstore data.  I would like to have data on the last 5 months (includeing current month) of sales.  With the current month I would like it to include the acutal data for the days in the month that have past and a projection of what the sales will be for the remaing days in the month. I would like it so that when new data it entered and it moves on the next month, the calculations will move on as well, and project the next month.


                    So right now in the sample data I would like a way to build a projection for January 2013 based of the data in the past 4 weeks; and off the last four weeks the most recent week is weighted more heavily than the other weeks and so on.  For example the most recent week is where 40% of the projection comes from, the following 30%, followed by 20%, and finally the furthest week 10%.


                    Thus I would like what has happened in the last four weeks in december to project what January's sales will be. And I would like with the projection of Jan 2013 to still inculde the acutal data from the previous 4 months, which on my current sheet I include.


                    If anyone can help create this forecast, I would be greatly thankful!

                    • 7. Re: Forecasting - how to?
                      Paul Morgan

                      Tableau 8 has this feature (or did in Beta 5) - not too much dox on it but it seems it has the possibility to create forecasts and confidence of said measure