While businesses are discovering the need for advanced analytics to be able to manage change, organizations do not necessarily all have the ability to leverage data in the same ways. Over time, as analysts gain more expertise they will be able to gain new levels of sophistication. Below you’ll find the four stages of maturity that companies go through in gaining insights from their data.
Stage 1: Collecting, cleansing, integrating, and reporting on data.
During this stage you can use this data to query and analyze current and past business performance. Before you can make sense of data, it is critical to understand what problem your company is trying to solve. What do you want to do with this data and why? When the organization understands the business goals, it is the right time to put a strategy in place. This stage is important in creating a baseline of consistent and trusted knowledge about the business. You can’t begin to accurately predict where your company is headed if you don’t have a clear understanding of where you have been. Data cleansing and data integration ensure that senior executives understand that reports from business units such as sales, operations, and finance are accurate. Any predictions of future performance based on current and historical information make the implicit assumption that business operations are stable. This approach assumes a Bayesian approach that begins with a hypothesis and then uses a statistical analysis to reach conclusions. Therefore, the rate of change is considered only for the rate of growth or decline in the overall market.
Stage 2: Trend analysis for forecasting.
This stage uses basic modeling capabilities to make business forecasts based on an analysis of historical trends. For example, a clothing buyer for a retail chain looks at past sales across the company’s stores and forecasts next year’s sales by store prior to placing a new order. The model is likely to account for various factors that differ across each of the stores such as climate, store location, and demographic characteristics of shoppers. The buyer may apply what‐if analysis to adjust the sales forecast based on changes in selected variables. For example, what if next season has 5 or 10 additional heavy snowfall days? The forecast can be adjusted downward to account for less traffi c in the store due to snowstorms. Although creating a forecast for the future based on past performance is a good place to start, these models were not designed to capture and account for change as it is happening. For example, the buyer in this example may end up with a lot of unsold merchandise after overlooking rapid changes in fashion trends among a certain demographic. The outcome from these systems tended to be based on the ability to codify current knowledge and report from those findings. In essence, the results of leveraging these systems were not predictive in nature. Rather, the results are based on a structured and well‐defined set of problems.
Stage 3: Predictive analytics.
This stage is defined by the use of statistical or data‐mining solutions that consist of algorithms and techniques that can be applied to both structured and unstructured data. Multiple sources of both structured and unstructured data types can be used individually or together to build comprehensive models. Some of the statistical techniques used in this phase include decision tree analysis, linear and logistic regression analysis, data mining, social network analysis, and time series analysis. A key factor in predictive analytics capabilities is having the ability to incorporate predictive models with business rules into the operational decision‐making process. This makes the modeling process more actionable and helps businesses to improve outcomes. The focus of predictive analytics is on anticipating trends before they happen so that you can act to minimize risk for the business. Although predictive analytics has been used for many years by statisticians in certain industries, advances in software tools combined with increasing compute power has made this technology more accessible and more widely used by business users. Predictive analytics models are designed to analyze the relationships among diff erent variables to make predictions about the likelihood that events will take place. For example, an insurance company may build a model that analyzes the components of fraudulent claims and use this model to fl ag new claims that have a high probability of being fraudulent. Another common use case for predictive modeling is to help call center agents understand the next best action to take when they are interacting with customers. Based on the individual customer’s profi le, specifi c product recommendations can be made at the point of interaction between the agents and customer.
Stage 4: Prescriptive and cognitive.
Prescriptive and cognitive approaches take predictive analytics to the next level by applying machine learning algorithms, visualization, and natural language processing. Companies want their models to look beyond their internal assumptions about customers and products so that they are better prepared to respond to changing market dynamics. If models are designed to continuously learn based on each new interaction, the accuracy will improve. For example, a mobile service provider uses analytical models to help customer service agents reduce the level of customer churn. These models analyze information about a particular customer and predict what action the company needs to take to retain this customer. The company’s predictive models were updated infrequently and lacking in accuracy and sensitivity to competitive changes in the market. The company significantly improved its customer retention rate by designing a new model that is more prescriptive. The model is designed to be self‐learning by feeding each new interaction back into the model, capturing changing market conditions. In addition, the model incorporates social analytics to understand the customer’s interactions with and infl uence on others. These changes improved the model’s capability to help drive accurate decision making regarding what the next best action should be to support the customer. Models that are designed to adapt and change are beginning to be used by companies to predict when a machine is likely to fail so that corrective action can be taken before a catastrophic event occurs. For example, patterns identified in streams of machine data coming from sensors in a train can be used to build models that will anticipate equipment failure before it happens. By using adaptive learning, the model’s accuracy can be continuously improved to provide a real‐time warning of equipment failure in time for the company to take corrective action.