We hear a lot about data-driven fundraising throughout the industry. That’s because using data to make smarter decisions has proven to lead to better investment outcomes.
Over the years, people have been obsessed with such jargon as “predictive modeling,” “machine learning” and “artificial intelligence” to describe an analytic process that works like a crystal ball to predict the future. Not surprisingly, while these buzzwords have helped spread this new technology, they have also created a lot of confusion.
As a data scientist in the field, I would like to help clear up the common misconceptions with a simple Q&A.
1. What is predictive modeling?
Modeling is an ensemble of math-powered techniques that help humans predict outcomes by using large-scale data to explore and capture complex relationships. Nonprofit organizations use modeling to help assess and make strategic moves that would otherwise be too risky to attempt.
2. What is machine learning?
Machine learning is essentially modeling—they are two sides of the same coin. The term “modeling” describes the purpose, while “machine learning” emphasizes the process.
By a strict technical definition, modeling overlaps with machine learning by about 70%. But in most cases, the terms can be used interchangeably when it comes to fundraising and marketing.
3. What is artificial intelligence?
Artificial intelligence (AI) is the umbrella term to describe any part of decision-making not involving human brains.
Modeling and machine leaning constitute the fundamentals of AI. In fact, modeling, machine learning, and artificial intelligence are all roughly equivalent in today’s nonprofit marketing terminology.
4. Is there a "best" modeling technique?
Given the right data, today’s prevailing modeling techniques should yield similar results. The performance difference lies more in how that data is qualified into a model—i.e., the “craftsmanship” made of the modeler’s data philosophy, business acumen and technical expertise.
Simply put: A model is only as good as the modeler who created it. That is, the devil is in the details.
5. Is modeling a black box?
In other words, do you just input data into a model and let it spit out predictions and forecasts? Yes and no.
A model consists of a set of mathematical equations, the intricacies of which can go beyond the layperson’s appreciation. However, we at RKD believe in making sense of every model to grasp its strengths while being prepared for its potential pitfalls.
6. When should modeling be used?
Modeling tends to be most helpful when used to predict something that is unlikely to occur. Computer algorithms can aid in reaching a decision that is difficult for the human brain to discover.
In the context of direct response fundraising, modeling comes in most handy in developing “new” donor relationships—think acquisition, lapsed reactivation or even major gift upgrades.
In each case, we are trying to push donors from one category to another, anticipating inertia to be high. Thus, modeling can help us identify a penetrating point.
7. What are the limitations of modeling?
Because a model is built on past records, we need to be careful when the history of the data does not represent the future.
Modeling means following. If following in a wrong direction, a technically sound model can turn out to be futile.
Modeling also works within boundaries. If the goal is to push those boundaries, however, the model will be ill-fated.
8. What should we avoid in modeling?
Frankly, there seems to be a tendency to over-model in certain areas.
For example, the proliferation of co-op lists backed by modeled databases has driven increasingly older donors into many organizations. These donors are picked up by models as more responsive prospects and, therefore, shared among more and more organizations. It is a self-reinforcing process that can quickly wear out donor motivation.
This heated competition for a common donor population has contributed to a noticeable decline in direct mail acquisition performance across the industry.
When we use a model to make a choice on behalf of a donor, all subsequent decisions we make about the donor can be biased. We must be aware of the potential risk of stereotyping in any model.
9. Can AI replace human intelligence?
Only partially. Knowing the pros and cons of modeling, we can conclude that human judgment plays a crucial role in modeling success. Artificial intelligence does not replace human intelligence—AI extends it. The stronger the human intelligence is, the greater the combined intelligence will be.
An updated model entails updated human guidance. At RKD, we see full-lifecycle model management as integral to a successful modeling project.
I hope the information above helps provide some context and understanding for the next time you’re discussing how nonprofits can use modeling, machine learning and AI. Best of luck in all your fundraising ventures!