As a data and system literate business partner I’m always interested in applying technology to solve business problems.
In recent months I’ve been evaluating business friendly Data Mining apps to open the door to the next level of business analytics.
The reason I’m interested in this is for two main reasons.
Big Data will swing the analytical pendulum towards more use cases that involve a high level of processing and modelling. With BI we often have to aggregate transactions to optimise OLAP processing. To date, this has limited the practicality of analysing transactional data. With Big Data we’ll finally get the opportunity to consume transactional level data and unstructured data on a more frequent basis.
No need for a PhD
Data Mining is within our grasp. The industry standard framework for working with Data Mining is called CRISP-DM.
Data Mining begins with business understanding. This is where the objective is studied and requirements gathered from the perspective of the business. Data literate business people are close to the questions that can be answered by Data Mining.
Data Preparation is akin to self-service data integration. With tools like Power Query you don’t have to be a developer to prepare data. Although a BI developer certainly comes in useful if they are available to you.
This is not a big part of the process. Most effort actually goes into the preceding stages. We don’t have to understand the modelling techniques at a PhD level. Do you understand regression? Know your R2 ? How about confidence interval?
There is some learning involved but you may be surprised at the level of learning required. In my experience with a base level of knowledge it’s better to get started with the doing and learn as you go.
The software and learning material is finally becoming more business friendly. Don’t bite off more than you can chew initially – start with something you feel confident about delivering.
This is where we measure the quality of the model and check the overall robustness. Does it cover all pertinent issues? At this time we decide if the model is good enough to deploy.
We want to avoid this.
Deployment can take many different forms depending on the objective. Think about a model that predicts the customers that are likely to churn. The model also tells us which factors are most likely to lead to the customer churning.
In this deployment type outcomes are fed back into the operational process/system to reduce the likelihood of churn taking place. Think about alerts to trigger actions or retention campaigns. It’s here that you’re likely to hit the IT buffers but this really depends on the systems you have and the IT resources at your disposal.
With the right tools and access to data, data literate business professionals can produce the insight that will add value in different ways to traditional BI. Don’t wait around for the Data Scientists.