The typical approach to predictive analysis is to start up a new analysis process from scratch for any new business request that is made. For example, if a non-profit organization wanted to use time-series forecasting methodologies to predict future donation amounts, the analyst would gather operational data, determine the methodology to use, train the model over multiple iterations until an acceptable configuration was found, produce the results, and deliver the findings back to the user. Any new request would start the process over again. There are two things I’d like you to consider relative to this work.
A lot of analysts use the operational data itself when gathering information needed for forecasting. I would like to suggest that you use your modeled data instead. There is an order to analysis that allows an appropriate data foundation to be defined – descriptive to diagnostic to predictive to prescriptive. The descriptive and diagnostic stages are critical to cleaning and preparing operational data for analysis. If you use this modeled data as the source for your predictions your forecasting will be based on much cleaner and more accurate data that reflects how the business functions – this will produce better predictions. Also, your forecasted numbers will play well with your diagnostic visuals since they will be based on the same dataset. Having two reports show different numbers for the same metric is a quick way to hurt the integrity of your BI platform. The price you pay to properly describe your data is well worth it.
Something else to consider when doing predictive analytics is to build a framework to house your various methodologies that you can configure and use repeatedly. There is a natural progression in any technological advancement to start first with everyone doing their own thing. I remember when web page development was first beginning – there were people using Frontpage, Visual Studio, and even notepad to produce websites. Over time as website development has matured, production has greatly accelerated – developers can now use established platforms to quickly customize and configure their websites. Coders no longer must start from scratch, instead they can leverage frameworks that help produce quality secure work quickly. This progression will eventually happen in the predictive world. I suggest that you future-proof your organization now and start building a framework to handle your various forecasting methodologies. Instead of starting new with each business request, start with a basic framework that handles what can easily be managed. Then over time you can grow your platform to handle more and more aspects of the analysis. This will allow you to produce consistent results and be able to handle the increased business requests as word of your successful predictions starts spreading throughout the organization.
These two suggestions can greatly benefit your BI program and help you make the leap into predictive analysis successfully. Here at ConradBI, we can help you safely navigate into the advanced land of forecasting. Contact us today!
