What are KIT Predictions?
KIT is a technology that uses organizations' unique data to provide insights into donor behavior, make predictions about future donor behavior, track fundrasing success, and guide fundraising strategies.
Predictions look at your historical data, pick up on donors’ behavior patterns, and communicate the potential for donors to give; up to and including how they are likely to give, the amount they are likely to give, and when they are likely to give.
A few examples of KIT’s predictions include:
- Major Donor Potential
- Likelihood to Recur
- Time of Year
- Likelihood to Lapse
- Second Gift Likelihood
See the full list of Predictions here.
How do Predictions work?
While artificial intelligence (AI) can be a daunting term, our job is to simplify it so you can better understand how AI interacts with your unique dataset to generate predictions.
To generate predictions, KIT uses machine learning models. These models look at all of your historical data points and look for top indicators that lead to a particular donor behavior. The indicators that the model identifies are weighted, meaning the indicators found to be the most predictive of a certain donor behavior have a heavier influence when KIT makes predictions.
As an example, let's look at ‘Recuring Donor Potential’, the model is going to look at your current recurring donor base, and identify commonalities amongst those donors. Those overlapping characteristics are translated into indicators of who is more likely to become a recurring donor. In the case of weighing indicators, if we find that one of the most prominent indicators for ‘Recurring Donors’ is a donor's frequency of donations, the model places more weight on that indicator than it will on the others; it will do this for each indicator. Overall, the more a donor's past behavior reflects the model's indicators, the more likely they are to be associated with that prediction.
Keep in mind that each prediction has its own unique model that generates specific indicators for each prediction based on your dataset. The machine learning model continues to learn over time which indicators are most important for each prediction and will continue to adjust them as your database grows.
For example, perhaps for your organization, one of the indicators of 'Major Donor' potential might be demographic area. However, as you gain more Major Donors, the model might find interactions as playing a significant role in Major Donor behavior, thus, the model adjusts.
It’s important we capture these nuances and patterns to uncover potential in donors that may have otherwise been overlooked if we were only to look at more obvious and default categories such as giving amount. So while some of the indicators used to make predictions might surprise you, keep in mind that the purpose of artificial intelligence is to pick up the patterns and nuances (i.e. indicators) that might not be so obvious or intuitive from the human perspective, presenting you with the most possible opportunities. Sounds like a fundraiser's dream, right?
Which indicators are used to make our organization’s unique Predictions?
Great question! Given that model will determine which indicators amongst your donor base are the top indicators for each prediction, you might be curious about what the model found. For this very reason, KIT has a Predictions Report that defines each prediction while showing you which indicators were found to be the most indicative for each prediction.
How accurate are our Predictions?
To determine the accuracy of the predictions KIT generates, we test the models that are used to make predictions. How do we do this? Well, we take a chunk of historical data from a set period of time and allow the model to make predictions on that data set. We then look at the later outcomes of that historical data to see if the predictions that were projected were accurate. The model then learns from how accurate its predictions ended up being and improves itself to make smarter predictions. These models are trained on an ongoing basis to continue to improve their accuracy. Below you will find the current accuracy score for KIT’s prediction models:
- Time of year - 92%
- Donor Readiness - 91%
- Likelihood to Recur - 96%
- Major Donor Potential - 79%
- Repeat Donor Lapse - 92%
- Recurring Lapse Potential - 92%
- Second Gift Likelihood - 75%
How do I use Predictions?
Keep in mind that when looking at a prediction, ‘Major Donor Potential’ for example, it is going to pull up anyone in your donor base that has behavior that reflects the indicators that have been identified as influencing someone's likelihood to become a Major Donor. What this means is that when you look at a prediction, you are looking at the potential for someone to become a Major Donor, even if that potential falls into a lower range. KIT’s models give you a large range of results to make sure you have access to the widest range of possibilities within your data set - giving you access to the most opportunities possible.
With that said, we emphasize the importance of filters and exclusions. Filters and exclusions are how you narrow down contacts in a predictions list that you believe might not be the best candidates for a particular campaign. This is what will create a balance between an opportunity list: the raw predictions list that KIT generates, and a strategic list: the additional filters and exclusions you add to the list, to create more strategic and targetted solicitation based on your campaign goals.
For example, if KIT’s model for my unique data identified demographic data as one of the highest indicators for Major Donor Potential, there may be donors in my ‘Major Donor Potential’ list that come from an affluent area but haven’t donated much in the past. They are on this list as they have the potential capacity to give. However, if I know I only want to target donors who have given within a certain range in the past, I would want to filter for donors who have historically given above a certain threshold (i.e. exclude donors who have given less than $10,000+).
We understand that filters and exclusions can be tricky, how do you take a large list of opportunities and determine who to filter out to produce a more targeted list? To guide you, visit Building Strategic Lists Using Predictions & Filters for recommended filters to consider for each prediction list. This will help guide you in building more strategic lists based on your campaign audience and goals.