Patterns are all around us, but whether we can see them and make sense of them is a different thing entirely.
While it may be easy for you or me to look at our credit card statement to determine if there were any fraudulent transactions, it is a bit more challenging for a credit card company to look at all transactions taking place across the globe to determine if a transaction is fraudulent before it is authorized. Predictive analytics is about applying statistical techniques to large amounts of data, finding patterns and then using them to classify customers, predict orders, recommend complementary products and more.
BlumShapiro Consulting has developed a predictive analytics methodology that has helped our clients to transition ideas into reality in a streamlined and repeatable way.
Consisting of 6 iterative steps, our Predictive Analytics Methodology (PAM) can be applied to a variety of business problems including demand planning, production planning, staffing planning and predictive maintenance. The following sections provide an overview of these steps.
Our Predictive Analytics Technique:
1. Identify Business Problem
The value in a predictive analytics solution is in being able to be hyper-focused on a specific business challenge. Perhaps you’re a manufacturer trying to predict the outcome of a manufacturing process dependent on a variable material input. Maybe you’re a distributor of goods with a short shelf life or a long lead time, and having a better sense of demand would improve your bottom line. Regardless, it is critical to define not only the broad business challenge, but the specific problem that can be addressed through predictive analytics.
2. Assess Data
“Cash is King” used to be the mantra of many an executive, but that has been replaced by the adage “Data is King” where having good data is the key. When looking at a business problem, it’s important to first determine whether data is available to address the business problem. For example, if sales history at a SKU level for each store by day/hour for the last 2-3 years isn’t available then it may be challenging to predict demand. When working with an organization we look to not only assess the impact of a business problem, but also the ability for that organization to solve for it with relevant data.
3. Transform and Refine
With a clearly defined business problem and relevant data available we then work to transform and refine the data so that it is more useable and consumable by predictive analytics tools. This often involves data profiling to ensure we have a clear understanding of the data in question as well as data enrichment to supplement data with additional, related data that may not have been part of the original data set.
4. Build Model
This is where the heavy lifting of data science comes into play. Whether we are working on classification, regression, clustering or recommendation models, our data scientists use the Azure Machine Learning Studio, R Studio and Cortana Analytics tools to create a model that exemplifies the business problem in a predictive fashion.
5. Validate and Deploy
Once we have a working model, we then work with our clients to validate the results in a limited test to ensure that the model is working as intended. This typically takes place through a parallel process where we continue using the current method of planning or forecasting, but run our model alongside. Through this process, we can identify any potential opportunities for improvement in the model. Once any improvements are made we deploy the model through a web API.
6. Incorporate and Evaluate
Having an effective model in place with a web accessible API we can incorporate the model into your business process. This can be accomplished very simply by connecting the API to something like Excel for simple data entry or by incorporating it into a line of business application for real-time decision making.