Beyond Guesswork: Predicting Donor Behavior with Data Analysis

?

Client question: How can software predict when a donor is likely to give?

My response: The same way Amazon’s recommendation system predicts what products you might find interesting. Or how Netflix predicts which movies and shows you are likely to enjoy based on your viewing history.

Of course, I did not, could not, stop there. The client was curious if this approach would enable their organization to customize marketing efforts to learn the optimal outreach time for each of their specific donors. The open secret behind these recommendations lies in the power of machine learning.

I promise this will not get technical!

?

Learning from Examples

Today, let us focus on a predictive model that purports to predict which donors are likely to donate and when.

For that, let us imagine you are a new employee in the grants department of a relatively small non-profit. You want to write successful grant proposals but have no access to any manual or instructions for success. Hesitating to ask your super busy colleagues, you decide to learn from examples of past grant proposals submitted by your team.

You gather a collection of different grant proposals submitted by your organization. Then, you scrutinize the components of these proposals (need statements, project plans, budgets). You start noticing patterns. Awarded proposals exhibited certain characteristics:

· Proposals with strong data-supported needs assessments tend to score higher

· Proposals with clear, measurable objectives and evaluation plans are often favored

· Proposals with realistic budgets and evidence of institutional support seem to do better

Studying these patterns from past successful examples, you begin to understand what makes for a compelling, fundable grant application. During this process, you are training yourself on what works or does not work by examining the ingredients and methods used in previously awarded and unsuccessful grant proposals. You begin to extract the core underlying patterns and characteristics that can guide you to developing new winning grant applications.

And this is the essence of AI's machine learning!

?

Recommendation Systems

Much like how you are learning from data (past grant applications), the recommendation system (like Amazon's) learns from your past purchases and browsing history, combined with those of similar customers for relevant product suggestions. ??As your grasp of what makes a successful grant application grows when you study more and more past grant applications, the recommendation system becomes more accurate with more data about your preferences.?

Predicting Donor Behavior

Let us extend this logic to the model for predicting donor behavior. The model needs to analyze a significant number of records containing donor demographics and past giving patterns to build its predictive power. The more quality data used in training, the better the model's predictive capabilities. Note several factors influence how much data is needed for each use case but let us set that aside for another day!

For non-profits, leveraging machine learning models for donor behavior prediction can lead to more efficient fundraising efforts and better allocation of resources. ?Organizations can tailor their outreach strategies and optimize their campaigns knowing when to reach out to different donors.

?

Real-World Applications

Just like recommendation systems, machine learning models are used in various sectors.?? Your email spam filters use this approach.?? Banks use them to detect fraudulent transactions.???? Manufacturers leverage machine learning for predictive maintenance, identifying potential equipment failures before they occur.

It all comes down to identifying patterns in historic data to label a new email as spam, a new transaction as fraudulent or an anomaly in equipment performance as a precursor to potential breakdown.

?

A Word of Caution

Limited data can handicap a predictive model’s accuracy (Amazon has no problem on this front!).? Similarly, responsible data collection and usage are crucial for the ethical application of machine learning models.? Biases can easily slip into the analysis and affect predictions.

?

In Conclusion

Humans can learn and improve their skills through studying examples. So can machine learning that enables systems to identify patterns in data and make accurate predictions or recommendations.

Remember this analogy the next time someone mentions predictive modeling or machine learning. You will feel more confident in understanding how it works at a high level. You’ve got this!

#nonprofits #philanthropy #techforgood #ai

*************************************************************************************

Indu Sambandam ( Agaya Consulting Inc. ) is dedicated to helping mission-oriented/social impact organizations struggling to unlock the full potential of their data.