The Use Cases for Analytics

Defining Analytics Use Cases

 

Today, artificial intelligence (AI) might be taking the business world by storm, but many of the models found in AI’s toolset are just analytical methods like logistic regression, random forests, and bayesian inference dressed up as AI, which speaks as much about the power of analytics than the inherently misleading messaging of AI marketing. Whereas standard machine learning or deep learning modeling techniques like neural nets can be rather opaque and difficult to grasp, understanding how random forests — decision trees on steroids — work is much easier to grasp.

 

 

The concept of analytics is nothing new. Archaeologists hypothesize that data and analytics have been integral to organizational decision‑making for thousands of years. Evidence of humans counting and recording quantities in notches on sticks or bones goes back over 45,000 years. Today, analytics is considered to fit within one of four types —  descriptive, diagnostic, predictive, and prescriptive — and while each has its unique purposes, they add the most value when they all work in concert with each other.

 

Analytics extends along a scale running from descriptive, which is the easiest to implement, up to prescriptive, which is the hardest to accomplish. As with most things in business, the return on investment (ROI) correlates closely with an endeavor’s difficulty level. Unsurprisingly, the analytical process is no different; descriptive analytics can help a business understand its website traffic and provide information on how to tweak a website page to ensure it keeps a customer engaged. Prescriptive analytical models can provide the most value, as they can help a business like an integrated resort predict the most profitable price to sell time-limited items like hotel rooms and concert seats or even a seat at a table game on the casino floor. This can generate healthy profits, and it unquestionably helps ensure use. An empty airline seat holds no value once a plane takes off, and yesterday’s empty hotel room is obviously useless to both the managing hotel and a potential patron.

 

An enormous amount of data is required for these revenue optimization models to accurately predict the exact price that will tempt a buyer to purchase a product as well as provide the maximum potential profit to a business, but the ROI usually makes an investment in analytical software to define these prices well worth it to the business. Although implementing an analytics solution is not easy, the price of the tools to build these models has come down substantially over the past few years, and the software is getting easier and easier to use. An analytics team filled with Ph.D. candidates is no longer a requirement to build highly sophisticated models, as the software has become much more accessible to laymen.

 

 

 

Four Types of Analytics

 

The four types of analytics exist on a graph that rises with the difficulty of implementing, which correlates well with the value the analytics bring to the business. While descriptive analytics tries to answer the question “what happened?”, diagnostic analytics explains why it happened. Predictive analytics tries to forecast what will happen, and prescriptive analytics attempts to answer the most important of all analytical questions — “how can we make it happen/prevent it happening again”?

 

As a business ascends the analytics ladder, there is a corresponding increase in ROI value. Prescriptive analytics provide the most value, as these models can be used in revenue optimization models, which predict the most profitable price a company can sell a time-limited product, like an airline ticket, a hotel room, a cruise line cabin, or a casino table game seat.

 

Although implementing an analytics solution is not easy, the price of the tools to build the models has been reduced substantially over the past decade, and tools have become so simplified that Ph.D. candidates are no longer required to build some highly sophisticated models, so there really is no better time to add analytics to a business.

 

 

Descriptive Analytics

 

Descriptive analytics is a technique that uses data to describe and analyze the characteristics of an entity or phenomenon. Descriptive analytics can be used to answer questions like, “How many people have a particular disease?”, “What are the symptoms?”, and “What is the prevalence of this disease in a certain region, state, or country?”

 

Market basket analysis is a descriptive analytics technique that provides answers to questions like, “What do customers buy on shopping trips?”, “How much do they spend on each item?”, “How often are certain items bought?”, and “How many different products are bought per trip?” Market basket analysis can also reveal customer and household spending patterns, including by age, demographics, household location, and income level. All of this information is useful on a customer service level but can also be used in a company’s marketing and product development.

 

Other descriptive analytics techniques include pattern discovery and customer segmentation models, which separate customers into items purchased, which can help the marketing department hyper-focus customer offers.

 

 

Diagnostic Analytics

 

Diagnostic analytics analyzes data to find patterns and trends that answer the question, “Why did it happen?” It can detect fraud and find flaws in the manufacturing process. It can uncover waste or abuse in a company’s procedures and processes. Potential problems can be identified before they become large-scale issues. It can also help predict future events based on current trends. For example, by identifying areas where customers are not receiving the service they had expected, a company can improve upon its service and raise customer satisfaction.

 

The key to diagnosing a problem starts with understanding it, and diagnostic analytics can look for patterns in a company’s data that might indicate anomalies. For example, a financial institution could monitor a customer’s credit card spending for unusual behavior, such as purchasing more than usual at certain times of day. Risk management is notified once an anomaly is detected, and they can investigate further.

 

In addition, reviewing historical data can uncover how and, possibly, why problems are occurring so they can be prevented in the future. This is artificial intelligence in operations (AIOps) in a nutshell. It uses diagnostic analytics to build a virtual understanding of a company’s IT system, then logs disruptive anomalies to develop a holistic view that keeps things operational in a proactive way.

 

 

Predictive Analytics

 

Moving up the analytics graph, predictive analytics uses data science, data mining, and statistical modeling to predict future events. The idea behind it is simple: with enough information about past events, accurate predictions about what will happen next can be ascertained. For example, with enough web browsing data, an eTailer can predict which users will abandon an ecommerce site; with plenty of individual customer buying data, future purchases can be predicted; with massive amounts of customer data, increasingly accurate stock predictions can even be made. Predictive analytics can be used in many different industries, including airlines, casinos, hospitality, manufacturing, marketing, gaming, healthcare, insurance, retail, finance, and more.

 

Predictive models take into account historical data points as well as other factors such as user behavior on a website or in a mobile app. Predictive analytics need both historical data points and a model that predicts future outcomes from those past observations. These models can be linear, logistic, and probit regression models, discrete choice models, Bayesian inference, time series models, survival or duration analysis, as well as machine learning and deep learning models like neural networks.

​​

Predictive analytics can provide insights into what activities drive conversions. Data about all relevant customer interactions are collected across multiple devices and this can help fill in the gaps that explain why a customer behaves in a particular way. Predictive models can help organizations make informed decisions about a customer’s buying behavior by predicting what they might purchase next. For example, building upon the diagnostic market basket analysis from above, logistic regression models can add a predictor component that helps a business understand what products get purchased together, which can both drive purchases as well as help with the location of an item throughout a store or even on a website. There’s even the possibility that all this detailed customer buying information can help optimize the company’s supply chain.

 

 

Prescriptive Analytics

 

Prescriptive analytics focuses on analyzing user behavior and intent rather than historical trends or other performance measures. Instead of focusing on what happened in the past, prescriptive analytics looks at how users interact with a company’s product today and then extrapolates how customers might use the product in the future. Put simply, prescriptive analytics asks the question, “What should happen?” Prescriptive analytics lets companies predict future customer behavior to understand how the customer moves through the marketing funnel and what the value proposition is for each customer. The hope is this will raise revenue by improving marketing conversion rates as well as increasing the average order value per customer.

 

 

Conclusion

 

According to archeologists, humans have been using analytics for tens of thousands of years. There’s just an innate need for humans to count, quantify, and evaluate things. Today, success with analytics is often what differentiates prosperous companies from struggling ones. Analytics can provide granular details about a company’s operation, then help streamline the businesses’ processes. It can provide important information about a customer base as well as help reduce labor costs.

 

Software, hardware, and cloud vendors are taking note, producing hardware dedicated to running highly complex machine learning and deep learning software simple enough they can be used by laymen. Companies like Microsoft, Google, Facebook, SAS, SAP, IBM, Hitachi, Teradata, and Hadoop are offering GUI-based solutions that allow companies of all sizes to use highly sophisticated analytics in their business.

 

From descriptive analytics models that create customer segmentation models to diagnostic analytics models that attempt to understand customer behaviors, to prescriptive analytical models that help with collection analysis, cross- and up-selling, reducing customer churn, to prescriptive analytical that set rates for a revenue management pricing model, analytics can help business in a myriad of ways. With a past as long and varied as analytics has had, there’s no telling what the future might hold for businesses that embrace its technologies and techniques. For businesses unwilling to join the analytics revolution, history’s judgment will probably not be kind.