Why off-the-shelf retention tools fail mission-driven organizations
A few months ago, the company I worked for evaluated an enterprise retention platform. Quoted price: $135,000 a year. In the proof-of-concept, it hit 57.7% recall, meaning it correctly flagged just over half of the customers who were actually about to leave.
My own model, built in-house over the previous few years, was already running at 89.7%. Same company, same customers, same problem. The executive team turned the vendor down on performance alone.
I don't tell that story to brag about the model. I tell it because the gap wasn't really about talent. It was about fit.
The platform wasn't wrong. It was generic.
Enterprise retention tools are built to work reasonably well across thousands of customers with different businesses, different data, and different definitions of "at risk." That's the whole value proposition: buy once, deploy everywhere. To do that, they have to make broad assumptions about what churn looks like.
My model didn't have to make those assumptions. It was built on that one company's actual data, tuned to that company's actual definition of a customer worth saving, and updated as their business changed. It had no other customers to be generic about.
That's not a special case. It's what happens whenever a general tool meets a specific problem.
Nonprofits have the same gap, just less visibility into it
A membership nonprofit's "at risk" donor doesn't look like a SaaS company's "at risk" customer, and it doesn't look like another nonprofit's either. Recurring givers, one-time donors converted from an event, lapsed members who still open every email: they all decay differently, and a generic scoring model tends to flatten them into one curve that fits none of them well.
The difference is that a SaaS company can usually tell when its churn model is underperforming, because churn shows up in monthly revenue within a quarter. A nonprofit's donor attrition is quieter. It shows up eighteen months later as a smaller gala, a flat annual fund, or a program that gets cut because the money that used to be there isn't anymore. By the time it's visible, the people who could have been re-engaged are long gone.
What "custom" actually buys you
Not a fancier model. A model that was built to answer your actual question (who is about to lapse, and is there still time to reach them) instead of a generic version of that question averaged across a few thousand other organizations that aren't you.
That, and someone who can explain exactly why the model flagged who it flagged, in a sentence you could repeat to your board.
If you want to see what this looks like against your own numbers, book 20 minutes and I'll walk through it live.