8 tips on using predictive analytics in fundraising

Think about your main fundraising businesses (i.e., campaigns, major and principal donors, annual giving donors, events, etc.) and how donor intelligence could enhance your existing strategies.

Predictive model(s) help fundraisers prioritize a set of prospects likely to commit specific behaviors. For example, the foundation at an academic health system uses a variety of models focused on major gifts, grateful patient acquisition, and event attendee conversion. The major gift models below help their fundraisers create prospect portfolios:

• Donor Renewal Likelihood: Scores donors based on their likelihood to renew their giving year over year for five consecutive years
• $1K+ Upgrade Likelihood: Scores donors based on their likelihood to upgrade from a first gift below $1K to making a $1K+ gift
• Major Gift $50K+ likelihood: Scores donors based on their likelihood to make a $50K+ gift to the organization

While most fundraising predictive models focus on predicting groups of individuals who are most likely to make a planned gift, major gift, or annual gift, there is a lot of value in models that predict negative behaviors too. For example, understanding if a certain donor constituency is likely to lapse after 5-years will help you engage those donors sooner, and potentially prevent donor attrition.

Consider if any of your existing prospecting tools answer your questions today, and how your staff utilizes your tools.

Make a list of your tools and survey staff about their perceived value and utilization. Consider removing tools that are being underutilized. For tools with high staff ratings, consider asking the vendors of those tools if other tools exist to compliment them.

Evaluate current data sets and consider where there are gaps. Prioritize data elements that your organization has collected across multiple years—this will help your model find patterns. If you lack data for your grateful patient program, work with your organization’s IT team to develop a list of HIPAA compliant variables. If you use real estate values as wealth indicator, consider ways to weigh real estate differently in your model depending on location (hint: prospects with property in “hot markets” tend to receive overinflated capacity scores).

Examples of commonly collected data points:

• Wealth and giving history: Donor status, total giving, months since last gift, largest gift, consecutive annual gifts, 5-year giving capacity, and estimated net worth
• Biography and relationships: Geographic location, alumni status, board member status, employee status, volunteer status, number of connections with health system leaders or board members
• Health care participation data Number of visits with a specific physician, number of department visits, date of first visit, and date of most recent visit
• Philanthropy involvement: Fundraiser visits, event attendance, written off gifts, current participation in an existing pledge, and responses to philanthropy outreach

Assess data analytics expertise on your team as well as your access to your institution’s IT and data analytics team. Since staffing is a significant financial investment, it’s helpful to review staffing and budgeting together.

If you lack access to expertise, consider how to reallocate your budget for a new FTE or make the case for a data analyst. According to our recent survey of development leaders, the median organization deploys 1 FTE in a data analytics role. Investing in a data analyst helps teams deploy and review data models regularly.

If there is both a deficit in expertise and an inability to hire a dedicated FTE, consider partnering with an external vendor. Philanthropy consulting firms build custom models based on your needs and available datasets. For example, one foundation at an academic health system that we spoke with partnered with an external consulting firm to create five major models for their unique fundraising needs. These models are continually updated and improved by the external firm’s data experts.

Put processes in place to ensure your data’s integrity. To extract insights from your data, it needs to be complete, parallel, and current in your constituent relationship manager (CRM) software. At a minimum, make sure to constantly update addresses, contacts, demographic information, and deceased/active status.

To ensure your database collects accurate contact and gift information, run regular data integrity audits and limit record edit permissions to the main users of your CRM. Legacy Health runs 35 regular data integrity checks to quickly identify issues. As a result of these efforts, they’re able to import data from multiple external data sources and increase fundraisers’ confidence in their predictive analytics.

Collect a list of the major players in the industry for prospect research analytics and their offerings.

Review what pieces of data their models require and if you have access to that data. It’s helpful to request a sample file from each vendor of interest. Using your organization’s own data, the sample file can provide a quick glimpse into whether the data produced aligns with your expectations. Talk to the vendor’s service team about future goals and ask about their ongoing service model.

Provide insights based on extracted data and communicate actionable information to fundraisers such as cut-off score suggestions for prospect outreach. Communicate early successes from analytics projects to encourage other teams to use data analytics in their workflows.

The Dana-Farber Cancer Institute demonstrates the value of their internal data analytics function to their entire fundraising organization. The analytics team within prospect research develops various analytics related projects for several teams: events, annual giving, major gifts, and principal giving.

Following a project completion, they share how the predictive model performed in the following statistical terms:

• True positive rate: Percentage of donors who are correctly identified in the model
• True negative rate: Percentage of non-donors who are correctly identified in the model
• False positives: Percentage of unexpected donors who are incorrectly identified in the model
• False negative rate: Percentage of unexpected non-donors who are incorrectly identified in the model

As a result of their education efforts, nearly all frontline fundraising teams at Dana Farber use predictive models on at least an annual basis.

Tracking ROI is particularly difficult for predictive models since models (1) are used in addition to various strategies which makes quantifying the individual impact of a model challenging and (2) require a large sample size for results.

One way to measure ROI is by tracking how well utilized analytics scores are by different fundraising teams at your organization. The prospect research team at the Dana-Farber Cancer Institute tracks time spent on data analytics projects related to fundraising and annually reports on project completions to their leadership team.

ROI can also be measured in terms of conversion rate (from prospect to donor) for each distinct model. This is easier to track in predictive models that are targeting large populations. For example, as a result of testing their model, Dana Farber concluded there is a 3 to 15x greater response rate to their email campaigns targeting top scoring prospects (3% response rate) compared to baseline email outreach (0.2%-0.9% response rate).

Test your predictive model regularly. And since fundraisers are often the end users of a prospect score, it’s important to ask them for feedback on how helpful the model is at improving their jobs.

The development team at a large health system uses a custom scoring model based on 12 weighted variables which generates a letter grade. Each year, they retest the variables in the model using the most up-to-date data on their current donor population to analyze if those variables still produce strong correlation scores. Development leaders also solicit feedback from fundraisers on how the scoring system is performing. After receiving feedback from major gift officers who wanted separate information on wealth and affinity, they split the grade into letter grades for: (1) 5-year giving capacity and (2) overall affinity.