Alteryx Designer

Find answers, ask questions, and share expertise about Alteryx Designer.
Andy Uttley, Alteryx ACE, makes music with Alteryx | Math + Music

Seasonality and Trend Forecasting for Multiple Products

Highlighted
6 - Meteoroid
Hi All!

First time poster. Been a long time reader on the community and thankful for all the help that you guys provide.
I'm trying to develop an auto-reordering model and need some help with forecasting data.
 
Looking for a model for Seasonality and Trend Forecasting with several different products that each have their own trend. Attached is the historical sales data which will be the source.

I need Alteryx to group by each SKU and perform the forecasting process based on it's sales history. I have tried following the instructions from CristonS and a couple others' articles in addition to creating batch macros, but keep running into errors.
 
The forecast for each SKU is for the next 1 year. If possible, the next 3 months' forecast needs to be comparatively more accurate than several months into the future.
 
The source data will refresh every week with new inputs(this workflow is already ready). Model needs to account for these continuous changes and make adjustments accordingly for the entire year.
 
Thanks. Look forward to hearing from you guys!!
Highlighted
Alteryx Certified Partner
Alteryx Certified Partner

Hi @vvibes 

 

This should get you started, but I'm not an expert on time series modeling. The attached workflow features a batch macro that will create a new ARIMA model (uniquely evaluating trend and seasonality). for each [INV_ITEM_ID] value, and output the forecast for the next 12 periods (months).

 

To enable seasonal differencing on the ARIMA tool: on the "Model Customization (optional)" tab, there's a couple check boxes and a numeric input to adjust. A setting of 1 will use the trend of the last (1) year in the forecast of each period. 

 

20200624-SKUForecast_Seasonal.PNG

 

 

This should get you up and running (forecasting). Let me know if you have any questions about this workflow. Hopefully others with more expertise in time series modeling can chime in with model suggestions.

Highlighted
6 - Meteoroid

Hi @CharlieS,

Thanks for getting back to me.

I do need a later version of Alteryx to be able to run this. Will upgrade it and let you know how it works. Thanks so much!!

Highlighted
Alteryx Certified Partner
Alteryx Certified Partner

There is a way to "downgrade" the version by editing the XML of the workflow, but 2018.3 is pretty old at this point so maybe it is time for an upgrade. Let me know if you have any questions about either option.

Highlighted
Moderator
Moderator

Hi @vvibes,

 

Have you tried out the TS Model Factory Tool yet? It is a free download from the public gallery and allows you to use a string field for grouping (just convert the [group] to string). This will create a unique model for each group with sufficient data (24 months in this case). You can use the TS Forecast Factory Tool to make predictions from those models.

 

Hope this helps!

Highlighted
6 - Meteoroid

Hi @CharlieS,

Apologies for the delay. I was finally able to get approvals to update my Alteryx and install the predictive tools packages.

Tried the macro that you attached, but the forecasts that it is recommending is almost double when compared the actuals.

I keep reading that we have tools for this like the Model Factory, Forecast Factory and Factory Sample. Do you know how to use these?

 

Thanks

 

 

Highlighted
6 - Meteoroid

Hi @IanWi,

I tried using those tools, but I am not very well versed with using them and keep hitting obstacles like 'A record was created with no fields and input was not initialized' errors.

I have re-attached an updated data set and looking to forecast data for the rest of 2020. One tab has data at a consolidated level and the other tab has data from a plant level. Looking to forecast for both tabs.

 

Thanks,

Highlighted
Moderator
Moderator

Hi @vvibes,

 

You may want to check out the Data Science Learning Path that now includes lessons on Time Series Forecasting!

Labels