In today’s tech-enabled world, there a lot of terms being thrown around about predictive models and algorithms. We want to set the record straight on those terms and give you tips on what to think about in terms of selecting a vendor for predictive modeling.
Term 1: Algorithms
The term “algorithms” is a popular word and basically, is a sorting program. They take large amounts of data and create sort order based on certain attributes within an individual record compared to other records. For example, some data helps you rank prospects by wealth, investments, etc. This provides a good way to sort data but doesn’t really compare that data to specific anticipated outcomes.
Term 2: Predictive models
Predictive models are built based on historical data which is used to predict a future outcome. A true predictive model is built to predict a specific outcome based on historical data and provides a score to the data outputs. Scores help emulate your workflow and automate some of the workflow by creating a useable rank which is calibrated to your market and your historical results. For example, a model will show you prospects willing to donate based on historical data and other behavioral indicators. Then, it will rank and prioritize those prospects for you so that your staff is effective with their outreach.
Tips in selecting a vendor:
- Determine if the solution is an algorithm or a true model? A model is preferable since it will provide the most useful information. Ensure that the model is using the correct type of data and if it was built specifically for your industry because one size does not fit all with predictive models. What works for a healthcare foundation doesn’t work for other foundations outside of healthcare.
- Ensure that the solution is calibrated to your historical results and market. There are big differences in data and financial estimates by regional markets so it is important to take those into account when looking for a solution.
- Determine if the vendor provides you with an “odds table” that shows their scores based on prospects that donated and those that were solicited/did not donate. This is key to developing your workflow strategy and program design. It will also prove ROI from the onset so changing vendors or strategies does not become so risky.
- Find out if the solution is a generic score or algorithm and how many other non-profits in your market are using the same data. By finding this out, you will be able to determine how many other non-profits are chasing the same prospects. If everybody is using the same solution (and data), it really becomes a race to pitch a cause in hopes you get there first and trigger interest with prospective donors.
- Ask the vendor what provisions are in place to adjust the solution as results are accumulated and what is done to improve the model over time. Predictive models need to be calibrated over time so it’s important that the vendor stays innovative in this respect.
- Determine the flexibility of the solution in that it can be used across any fundraising software or is it captive to that specific vendor. Having integration between systems is critical. Even though you have a donor management solution in place, having a predictive model that overlays it and is integrated can make the donor management solution even more robust with the ability to provide more positive outcomes.
In today’s world, there’s a “solution for everything” but it’s important to select one intelligently that meets your needs. The real approach is to remain open-minded and nimble, and look for solutions that are specific to your mission and market.