Forecasting Tool Multiple Inputs

[aldoacosttaa]

Hello everyone,

I am attempting to build a MONSTER forecasting tool that will calculate monthly forecasts based on inputs from 7 columns ( for different items) that are 12 rows each (1 per month). These inputs represent a projected increase or decrease in items produced. I am fairly new to Designer, I am only CORE certified and am currently working on my advanced certification.

But as of now, I have only built one analytic app, and it wasn't to this scale before. So, I am trying to follow an excel skeleton that I made. On my excel skeleton, the inputs for these 84 inputs are connected to different sheets within the same excel, one sheet for each month that compares the estimate to what the past numbers look like.

Looking ahead, I plan on visualizing this data so I can easily filter and see which of the vendors are the bottleneck and cannot produce the proposed increase in demand for each part needed for the construction of the different items. I was wondering if anybody had a good forecasting example with many inputs, some sort of dashboard display with multiple filters, or any extra tips/guidance from all the Alteryx experts. (Or maybe all 3)

I appreciate any help or advice for trying to tackle this challenge. Something I have been thinking is that every sheet on my excel follows a similar format in the approach, the only thing that changes are the VLOOKUP/XLOOKUP cells selected in order to run the formulas and come up with the initial forecast comparison. Thinking how I could automate this even more so that my workflow is as efficient as possible. Another alternative I am debating is to produce 12 different outputs 1 per month that follow the same steps and combine it somehow at the end? Let me know what you guys think and if there is a better approach

Thank you Alteryx community! <3

[KGT]

I don't have an example, but there's plenty of forecasting examples out there. Have a look at Weekly Challenges 24/94/132/147.

You have a big process there and it would help to split it out.

• Bringing the data in and formatting it: You want to get it into one table if possible. You don't want multiple inputs, one of the huge advantages of Alteryx is it makes it easy to get that data together.
• For the actual forecasting part, I read it as you have 12 months data that you will use to forecast the future. To use a statistical method such as Time Series, your confidence intervals will get pretty wide really quickly if trying to forecast multiple months. I expect some seasonality in that 12 months as well which will throw the model off more with only 12 months data.
  • If you already have a method in Excel, you could replicate those formulas.
  • Caution around running multiple models. If so, it should be per item, not per month. Running a separate model per month in the future doesn't really make sense as you would be trying to predict something with a gap in the timeline.
• Display of data: I'll leave this part, as once you get the results, you'll be right...

Your first task is getting all the data into one table and ready, then you can try a couple of different things. No need for an analytic app for this though unless you want to productionise it on gallery later or something.

[aldoacosttaa]

Hey KGT,

Thank you for responding to my post! I'll make sure to check out all of those challenges to get a better idea of how I want to proceed.

 

You made some really great points and saved me a ton of time, since I was about to build out my forecasting model on month and not item! 

So far, I have the large majority of my data in a single table on Alteryx. I don't want multiple inputs but unfortunately I have to gather a bunch of data from raw sources first then compile it into one table, which this part is almost done. Currently I am working on adding in the build rates, avg days worked per location, capacity, etc.. from another source. I will try out the time series method for now and see what happens...

I believe that we are trying to forecast only 2 months in advance so I think a time series model will be a good approach and I'm not too sure about how seasonality will be built into it yet. also thank you for clarifying that I don't necessarily need to build an analytic app for now. I appreciate the guidance, and I hope you have a good week. Would it be possible for you to stay in this thread for a little? I will probably have more questions in the upcoming days :0

.