For more insights from the Kellogg Data and Analytics Club, check out their recent post on Medium — where they “helped compile an initial summary of data-driven research to aid communal education on racial disparities in the United States” as part of a larger effort “to engrain equity into the core of how we operate as a club and community.”
By Michael Ng and Patrick Donovan (both 2Y 2021)
Co-Presidents, Kellogg Data and Analytics Club
Companies across industries are increasingly operating in a world of big data, analytics and AI. The most profitable and productive firms have learned to harness these methods to improve how their business operates. A recent McKinsey study found that companies with the highest growth in revenue and earnings do so by leveraging data and analytics; these firms are three times more likely than other firms to have data and analytics contribute at least 20% to their EBIT. Yet, at the same time, many companies and organizations still struggle to build organizations around data science in practical, targeted ways. Here are five ways that organizations can use data science to improve their business.
Reporting and visualization
The most obvious and common use of data is in reporting, often classified as part of Business Intelligence. Though it might not seem very sexy, creating a regular and robust set of benchmarks and Key Performance Indicators (KPIs) is essential for monitoring all verticals of the business. Many reports are generated automatically and can be viewed through interactive dashboards through platforms like Tableu. Data Visualization in particular has become increasingly important in communicating trends and insights in the business to senior managers and clients across organizations. Innovative companies such as Narrative Science even allow for automatic text generation and machine-generated storytelling to accompany regular reporting metrics through a process known as Natural Language Generation (NLG). It’s become increasingly important for managers to be versed in analytics — even if they’re non-technical — in order to be able to contextualize and track changes in KPIs across the business.
One of the most fundamental ways of using data analytics in any business, particularly in software, is to constantly run tests. The purpose of a test is to understand whether a change you make to the business causally impacts your firm’s KPIs. That is, it allows you to provide data-driven evidence that a proposed change actually works. The way A/B testing works (similar to a Randomized Control Trial in social science) is to test a random sample of users with some kind of ‘treatment’; a treatment could be anything from the color of your website navbar, the copy used in an e-mail marketing campaign or your company’s process for signing up new users. You then compare the outcomes in the test group to the outcomes in an equally sized, randomly selected control group who did not receive the treatment. If the KPI of interest (conversion rate, sales, click-through-rate, add-to-cart rate, etc.) is statistically different in the test group than the control group, then you can attribute that shift in behavior directly to the change you made to the business. Testing provides validated, data-driven evidence that changes to a business process or product have reasonable expectations of being effective before being rolled out en masse. Companies such as Netflix, Wayfair and Amazon are constantly running tests to understand how changes to their platforms drive performance. Testing is also foundational to Lean methodologies that have become popular in software development. Platforms such as Optimizely provide platforms to easily structure and run tests.
Companies utilize data collected from business operations to automate various processes. This involves taking some form of relatively predictable work done manually by a human and replacing it, often with an algorithm. Take, for example, the case of fraud detection. Companies such as Capital One use advanced machine learning algorithms to classify whether a transaction might be fraudulent. Its system then automatically sends an alert to customers, allowing for a timely and quick response that would be much more difficult to do if transactions and communications were manually monitored and handled by a human employee. There are many other examples in manufacturing whereby automated machine learning methods can identify product defects and predict equipment failures that would be monotonous or difficult to do accurately by a human. Notably, this type of automation doesn’t necessary remove the need for human workers, but just shifts their focus towards more difficult or creative tasks. In the previous example, a worker can spend more time actually fixing a part as opposed to trying to figure out when it will break. Automation using data is best applied to very targeted and specific processes where you have to predict some discrete outcome that occurs regularly or classify some observation into groups (ex. is this machine part faulty or not?).
A close neighbor of automation, customer targeting uses data to deliver some type of offer (marketing e-mails, ads, product recommendations, etc.) to customers that are most likely to act on it. Facebook and Google are famous for their use of data to provide targeted advertising to consumers for whom it is most relevant. However, you don’t have to be a huge tech company to use data to target customers. RFM Analysis is a time-tested method of targeted marketing that doesn’t require any sophisticated algorithm or programming. More advanced forms of targeting might involve using statistical models to predict the probability that a customer will respond to a coupon, for example, based on their past behavior. You could then send the coupon to those customers with the highest probability of purchase (known as propensity targeting). An even better method would be to use uplift targeting to send offers to customers whose predicted probability of purchasing changes the most in response to an offer; this helps you avoid sending coupons to users who probably would’ve bought from you anyway. Though they sound sophisticated (and they are), many off-the-shelf machine learning algorithms are actually easily implementable using open-source software such as R or Python. The Customer Analytics and AI course at Kellogg covers many of these topics directly.
Perhaps one of the more important but more nuanced forms of analytics is the use of causal inference. In its purest form, A/B tests provide the best form of causal inference by randomly introducing some business or process change to a randomly selected treatment and control group. But when tests aren’t possible, it’s important to work with observational data to understand behavioral changes in a firm’s customer base or shifts in its operations. Such analysis breaks down drivers of KPIs by controlling for several covariates across groups. The most common way to implement such analysis is through multivariate regression. For example, regressing churn on exposure to a certain marketing campaign controlling for various customer characteristics (demographic info, purchase frequency, engagement, etc.) and covariates might provide a sense if the marketing campaign worked. For example, companies like Uber use causal inference to measure how delays to food deliveries in the Uber Eats business affects customer churn using a variety of methods. This is not a process they can test (you wouldn’t want to intentionally delay deliveries just to see how bad the effect is), so they use causal inference. Other companies like LinkedIn use causal inference to study the nature of user engagement. Causal inference can provide key insights into customer behavior, operational changes and other processes to inform strategy, often in broad ways. What’s important for adopting causal inference is to create a culture around data science such that managers know how to ask sophisticated questions about the business by using techniques such as causal inference.
To learn more about how data analytics is used in business across industries, check out content and events from the Kellogg Data Analytics Club.