What are the top tips for SMB marketeers getting started with predictive modeling?
Updated
Start with a simple predictive modeling use case that you can understand. At the end of the day, you are using past data to automatically generate business rules, many of which you might already know. As those rules are automatically discovered from your data by the analytics tool, it will give you additional confidence in the results if you come across some familiar ones.
Use free in-product or email support from services like BigML to get you started, get your data right, understand the models, etc. These tools come with free educational material covering use cases and capabilities for beginners.
The infamous Garbage in Garbage out (GIGO) rule is very much applicable in this context. You need to have a good handle of how your marketing data ties together and whether or not it is complete, accurate and free of outliers that tend to skew results. Depending on your roles and responsibilities this could mean getting some help from your CRM system admin or similar resource with access to and knowledge of your marketing database.
Predictive Modeling is seldom a one-shot deal but rather it is an ongoing refinement process. Identify and document ways your model can be improved in the next iteration. As is the case with many analytics projects, answering certain questions or being able to predict a given business metric opens up new possibilities that are worth exploring.
Do close the loop on your model by putting it to test against new data it has not yet seen. Some models are great with historical data, but fail to transfer this success to new instances due to something called overfitting. BigML lets you split your historical data into test and training subsets with a single click of a button so you can ensure that you have a validated model before you put it to use. Finally, closing the loop also means measuring the business impact of your predictions on a regular basis be it ROI, cost reduction or increase in sales.