What is predictive modeling?

Need a review? Check out our previous post, which defines predictive modeling and features some examples.

Does my organization have enough data?

The term big data has been everywhere in recent years, so it’s natural to think that you’d need vast amounts of information to reveal actionable patterns and trends. And while it is helpful to have your own data on potential customers/clients/donors and actual customers/clients/donors, the fact is that there are many third-party data sources that can be used to supplement internal data. Geodemographic data can be particularly useful when it comes to building models aimed at predicting human behavior.

Do we really need someone with a strong statistical background to create these models – can’t we just feed all of the data in and let the model figure out what is important?

Most people have heard the adage that “with great power comes great responsibility”. Its origins may be fuzzy (Voltaire, Winston Churchill, and Spiderman’s dear old Uncle Ben have all received attribution), but its truth is absolute. Yes, today’s use of algorithms and machine learning greatly scales the ability to discern signs in data that can be used to improve business outcomes. But software doesn’t know what your specific goals are; or how to spot problems and identify opportunities; and (perhaps most importantly) if it is has drifted into an erroneous and potentially dangerous place based on invisible, built-in biases.

Do you remember the movie Minority Report, in which murders were eliminated via the anticipatory intervention of police? Pre-crime technology is actually being used in law enforcement today, and it comes courtesy of predictive modeling. For example, the PredPol algorithm is used by several states to predict when and where crimes take place. But New Scientist reports that in 2016, the Human Rights Data Analysis Group found that the software could lead police to unfairly target certain neighborhoods. The takeaway is that machines and models cannot be left unattended, or even in the hands of untrained analysts. Constant review and adjustment by experienced statisticians is the best way to optimize your results and avoid missteps.

How can predictive modeling help us increase our effectiveness while at the same time cutting costs?

Knowing who to target and with which messages can greatly improve efficiencies in an organization. Say, for example, a Development Office has a database containing 25,000 potential donors, but the reality is that only a very small portion of those (e.g., 10%) will ever actually make a donation. A predictive model or machine-learning algorithm can identify which leads are most likely to donate and which are least likely. If we can eliminate 20,000 leads who have almost zero probability of becoming donors, we now have a much more manageable pool of 5,000 to cultivate. Devoting all staff resources to the new list of 5,000 can further increase the likelihood that these leads will donate.

How will we apply the findings to our particular business?

Back in Question 1, we posited that the first and most important step in deriving benefits from predictive models is defining your business goals. That recommendation comes full circle here as we answer the question of how you’ll apply findings to your particular business by saying that it depends on what you want to achieve.

There is almost no industry that would not benefit from the insights predictive modeling can bring. Whether you work in retail and need to better plan inventory so that you can meet demands while still avoiding storing product that will never move out of the warehouse; or you are a marketing company looking to demonstrate ROI to current and future clients; or you work in Enrollment Management at a college or university and want to know how many students you can expect to enroll out of your admit pool or what you can expect to spend on enrolling them, predictive modeling can work for you.