How Predictive Analytics in Fundraising Fails to Deliver Accurate Caseloads

Greg Warner

July 12, 2023

Every industry loves to predict the future. We predict stock markets, sports outcomes, educational outcomes, climate change, weather, and performance in just about any business sector. And as Big Data has grown, so too has the attraction to predictive data analytics, and fundraising is no exception. So how well does predictive analytics in fundraising really work?

Does it help nonprofits and gift officers effectively and efficiently qualify major gift prospects for outreach?

As you’re about to see, the answer is a resounding ‘no.’ And the reasons why aren’t hard to understand.

What Is Predictive Analytics in Fundraising?

Predictive analytics attempts to analyze pre-existing data and then use the results from that analysis to predict future behavior.

In fundraising, it aspires to exceed the capabilities of other problematic methods of filling gift officer caseloads (also known as portfolios), such as RFM (recency, frequency, monetary value) and wealth screening.

Is such a feat even possible? In theory, yes.

For example, a Harvard Business Review article reported on the successful use of predictive analytics in anticipating when airplane engines need to be replaced. Seems like a worthy goal. And in this case, the ability of the technology to forecast when engines will begin to break down – but before that breakdown causes any loss of life – has proven successful.

So predictive analytics as a general concept can succeed, given the right conditions.

And therein lies the problem. Fundraising – particularly major gift fundraising – does not contain the right conditions.

Biggest Flaw in Using Predictive Analytics to Identify Major Donors

If you take time to study the flaws of RFM and wealth screening, the problem with predictive analytics becomes immediately apparent:

It relies on the same data as those!

For example, one company that offers predictive analytics to nonprofits lists these eight fundraising metrics as good sources of data:

• Donation volume, within a chosen time frame
• Average gift size – for specific individuals or across the board
• Gift recency
• Gift frequency
• Demographics such as age, gender, location
• Wealth ‘markers’ such as job, real estate ownership, stocks (although these might generate false positives, especially if someone is upside down on their mortgage or house poor)
• Affinity markers such as past donation history, volunteerism
• Return on investment for particular campaigns or initiatives

Right away, you can see the RFM data in that list, as well as typical wealth screening data. This is the data they use to perform predictive analytics.

Problems with the Data

As the saying goes, garbage in, garbage out. If you have poor quality data, whatever you’re using that data to predict or recommend will not go well. Let’s look at just a few of many reasons why the above data is not only insufficient, but unreliable if you want your predictive analytics to deliver useful recommendations for prioritizing major gifts prospects to engage more intimately.

Outdated

The sources for wealth data used for RFM analyses constantly struggle to keep their data current. They are always playing catchup. People move, change jobs, gain family members, lose family members, get promoted, retire, get laid off, receive an inheritance, blow the inheritance – changes happen perpetually in everyone’s lives, and many of these changes affect their finances.

It is simply not possible for the entities attempting to keep current financial data on people to succeed in the task – to the degree necessary for accurate predictive analytics for major gifts fundraising.

Inaccurate and Untrustworthy

Wealth data is inherently inaccurate for two primary reasons.

First, many wealthy people proactively attempt to conceal their assets, precisely because of companies using their data for things like predictive analytics, as well as taxes. Whether you agree with this behavior or not, its reality cannot be disputed. And since most wealthy people own most of their wealth in assets instead of cash, concealing any of those assets will present a very different picture of their wealth from the reality.

Second, all this data looks to the past. That means it will fail to identify people who have newly acquired wealth, such as via a big job promotion, huge business success, or an inheritance. It will also fail to identify people who used to be wealthy but have lost a large share of it for some reason.

Thus, you’ll be relying on predicted behaviors that, for some donors, have zero chance of coming true.

Unreliable

Suppose you receive some new predictive analytics data on a set of ten donors who have all donated at least four times per year for the past five years, and have donated at least $2,500 each of those years.

Are these good prospects to consider adding to your major gifts caseloads?

This sort of information, by itself, simply cannot answer that question to the degree that is worth the time it takes for a gift officer to reach out and follow up with these prospects. Yes, you will have more information than just this, in some cases. Hopefully. But the point remains – this simply isn’t enough information.

One donor may be retiring and planning to stop donating altogether. Another may have donated because they knew someone at your organization, but that person no longer works there. Another may have just been laid off. They might have taken on medical debt. There are all kinds of reasons why these donors may stop giving. They might already be giving at their capacity, but the wealth screening data misleads you or doesn’t give the full picture.

And, some may indeed be capable of giving much more.

But that leads to the next problem with predictive analytics data.

Only Quantitative

Donors don’t give just because the numbers say they should or will. They give because of non-math related reasons. Donors feel emotions. They are drawn to causes. They have interests, values, beliefs, communities they care about. These are very personal motivations for giving to the causes they’ve chosen.

Essentially they want to feel like a hero to themselves and others. You need to know why before adding them to a gift officer’s caseload. Otherwise that gift officer will be wasting their precious time following up on leads who have zero chance of making a major gift because the hero story they want doesn’t align well enough with your cause.

And that kind of personal, qualitative information simply doesn’t exist in the quantitative data being relied upon for predictive analytics.

Small Sample Sizes

This isn’t necessarily true for organizations with very large donor databases. But for some nonprofits, you might not have enough supporters for predictive analytics to generate statistically valid predictions from your database.

Predictive analytics requires large sample sizes to have any hope of delivering truly helpful recommendations for major gifts prospects.

Predictive Analytics Fails in Almost Every Industry

This is a fun topic if you like to poke fun at people for making zany predictions.

For example, go look up articles from January 2020 on how the stock market was predicted to perform that year. It will give you a good laugh. Why? Because covid arrived a month later and blew up every single economic prediction. All the predictive analytics in the world meant absolutely nothing. And that’s not really an anomaly. It’s just an easy one to pick on.

Let’s look at COVID itself.

Early in 2020, scientists used predictive analytics to forecast the spread of covid, the anticipated caseloads and death rates, and the capacity we would need for hospitals.

Scientists with prestigious credentials predicted 100 million covid cases in just four weeks. They projected infections doubling every three days. They anticipated needing 140,000 hospital beds in New York state.

This article from the International Journal of Forecasting goes into great and depressing detail of just how wrong pretty much every covid prediction turned out to be.

Why did this happen? Here’s a damning quote from the abstract of the study:

“Poor data input, wrong modeling assumptions, high sensitivity of estimates, lack of incorporation of epidemiological features, poor past evidence on effects of available interventions, lack of transparency, errors, lack of determinacy, consideration of only one or a few dimensions of the problem at hand, lack of expertise in crucial disciplines, groupthink and bandwagon effects, and selective reporting are some of the causes of these failures.”

For example, one source of data informing the scientists’ predictive analytics was the Spanish flu. That disease killed 50 million people, with an average age of 28. Covid’s average age of death was 80, and even three years later, total fatalities aren’t even close to the Spanish flu, in a world with a much larger population.

Now – the point here is not to get political. This is just data.

The point is – this is the data they used in their predictive analytics modeling!

For predictive analytics to work, you need:

• Updated and accurate data – a very costly and ongoing endeavor
• Valid assumptions guiding your work
• Accurate math formulas constrained by reality
• An awareness of all the relevant variables, and a valid way to incorporate them

All four of those features are very, very difficult to achieve, and very expensive. When formulas involve exponents, just tiny variations in the data will lead to drastically disparate long term projections.

The Bottom Line on Predictive Analytics in Fundraising

Airplane engine breakdown can be predicted, because we have tons and tons of data, with known variables, and the ability to document everything due to global industry standards. But they still have to work very hard, and spend a lot of time and money, making sure their data is accurate and their math is correct.

But fundraising data is fraught with all the same problems as the stock market, communicable diseases, and many other sectors where predictive analytics largely fails.

There are too many unknowns. Too many variables. The data isn’t updated or accurate. And it relies purely on quantitative metrics that, even if accurate, would be insufficient to predict donor behavior and interest in giving major gifts.

A Better Approach

If you want to fill your major donor pipelines with more qualified prospects, you need a different set of data, and a different approach to using it.

You need qualitative data.

You need to know more about a donor’s actual life, real wealth capacity, interests, values, beliefs, and passions. You need to know the timing – for them – when making a big gift is most feasible.

None of that will show up in any predictive analytics data. But the good news is, you can glean all of it and much more using MarketSmart’s software.

We invented our software to solve this exact problem – to help nonprofits effectively and accurately pre-qualify real major gifts prospects.

And our system works so well that we offer a 10:1 ROI guarantee – you will generate donations worth at least ten times the cost of our software. We guarantee it. Schedule a free demo to see how we do it.

Related Resources:

• Why Wealth Screening Data Fails to Identify Your Most Generous Donors
• How to help your donors score a victory and make an impact
• 5 Questions to Help Identify Your Best (and Potentially Most Transformative) Major Donors in a Sea of Prospects
• How to put marketing automation to work for your nonprofit’s fundraising

LIKE THIS BLOG POST? LEAVE YOUR COMMENTS BELOW AND/OR SHARE IT WITH YOUR PEERS!

Subscribe

Login

Notify of

new follow-up comments new replies to my comments

Label

{} [+]

First Name*

Email*

Website

Δ

Label

{} [+]

First Name*

Email*

Website

Δ

0 Comments

Inline Feedbacks

View all comments