Editor’s Intro: Given the many sources of information about consumers available throughout many departments of companies, I have long thought that “insights Departments” risk becoming the “Department of some kinds of insights based on some kinds of data.” For Insights Departments to stay relevant, they need to focus more on prediction and less on insights. Lynnette Leathers gives us a valuable discussion about predictive analytics, and the value it brings.
The practice of predictive analytics allows us to make predictions on future behavior using historical data. Predictive models and their application in businesses will only increase moving forward. Did you know that you are likely to encounter (and use) predictive analytics each day, but may not realize it?
When you search on Google and possible suggestions appear it’s because predictive analytics are being used. When you shop on Amazon and other items you may like are suggested it’s because they are employing an algorithm which uses predictive analytics.
Predictive Analytics started in the 1940s and was used by government agencies. By the 1970s the process began to become more mainstream. By 1998, Google used algorithms to produce relevant search results. The short story is, this isn’t new and it’s being used more frequently by businesses, governments, manufacturers in multiple applications in the technology field, supply and demand and financial forecasting. You can employ a predictive analytics model just about anywhere you want to be able to forecast outcomes and responses: Sales, Marketing, Manufacturing, Customer Service, Healthcare, Government, Financial forecasting, Acquisition, Development, Investment, and the list goes on. The utility of data can be highly beneficial across an organization.
When creating a predictive analysis it helps to frame the terminology. The term model in predictive analytics refers to a “representation” of the specific world or universe we are talking about, an attempt to relate one set of variables to another.
Predictive analytics is the process that brings together data management, information technology, and modeling. The process seeks to bring together meaningful (statistically relevant) relationships among variables and represent these relationships in a model.
There are many types of predictive models that have been used over the years. However, two types of commonly used predictive models are: Regression and Classification.
Regression looks to predict a change in response (such as cost, price, increase, decrease, etc.)
Classification is used to predict a categorical response (yes/no, which ones, who, what, etc.)
At my firm Mindspot Research, we employ predictive analytics frequently in one of our services which measure Customer Experience attributes and relative performance. Once we collect the customer data, we then analyze the data and determine which attributes drive retention. We use this data to streamline any improvements. We can tell of the opportunities to improve – focus on this first. We overlay any variation by using a six sigma methodology and validating any correlations and regressions to determine the optimum order (which) variables will improve business (Revenue/Retention/Customer Satisfaction) the fastest.
Here are some examples:
- Using ‘customer service’ as the main improvement point first followed by two other variables a mature quick-service restaurant increased their revenue 5% year-over-year.
- By focusing on innovation a commercial printer and improved their business 11% over the next year even though their market was flat to declining.
- A paint manufacturer was able to provide existing sales and price data to build an algorithm to inform product ‘bundling’ opportunities for gaining more margin while providing even more value for customers.
There are 6 Stages in Developing a Predictive Model:
- Define- The purpose
- Collect- Determine the information that you have and the information that you need
- Build- Once you have collected your data set(s) you can begin create
- Test- Validate your assumptions
- Implement- Run the model to provide insight
- Improve- Take every opportunity to add relevant data and refine
There are certain assumptions made in producing a predictive model. However, it is the wisest approach to keep assumptions at a minimum from the beginning stage when selecting the predicting variables most important to your model. As you move forward and explore forecasting capabilities of your model, remember to remain flexible and cut away unnecessary assumptions. Your goal, when it comes to assumptions in your model, is to have as few of them as possible! This means, know your industry well and look to have reality-driven variables—this way, you can produce a predictive model that is accurate and useful.
Accuracy in predictive models is determined by the quality of data, the amount of data, the relevance of the data applied, the way the data is prepared, and how old the data is. Simply put, if the data is irrelevant or of poor quality going in—sadly, you will get nothing of consequence coming out. In essence, you are measuring the probability that predicting your stated outcome is better than chance mechanisms.
I find that the more types of data that you work with and the more iterations that are run the better your skill. That’s why when it comes to the interpretation the expertise that is brought to the table should not be discounted. It’s worth it.
It’s been my experience that everyone would like to increase their odds of winning. The point being when making a go-no-go decision, spending hard-earned capital or risking your brand’s reputation in the market, it makes sense to bring a predictable outcome to big decisions.