Member Health uses your association’s membership history to generate reliable retention insights, tested on historical data. To keep those insights accurate and fair, we only build association-specific models when there is sufficient history to learn from. This helps avoid “false confidence,” where a model that appears precise does not have enough information to make reliable predictions.
We use three levels of modeling, depending on the data available:
General models: These are built using data across all associations. They provide solid baseline predictions even when an individual association does not yet have enough history for a tailored model and gain a broad understanding of member behavior.
Association models: These are built using your association’s membership history so the predictions better reflect your unique patterns.
Member type models: These are built within your association for specific member types. They allow even more tailored predictions when there is enough data within each group.
Minimum Requirements for Association-Specific Modeling
To train an association-specific model (and the member type models under it) responsibly, we need a minimum amount of membership outcome history.
At least 20 expired members: We need enough examples of members who did not renew so the model can learn what “at risk” looks like.
At least 20 members who have renewed (Current members with at least one renewal): This is important: we do not count brand-new members who are simply “current” but have not yet renewed. We only count current members who have renewed at least once, because that gives renewal behavior examples (not just members who haven’t reached renewal time yet).
A balanced mix of outcomes: We also need the expired vs. renewing mix to be reasonably balanced. If nearly everyone renews (or nearly everyone expires), the model can’t learn meaningful differences.
If the numbers are too lopsided, predictions will become less trustworthy.
What happens if you do not meet these requirements?
If your association does not meet the minimum requirements above:
Member Health will be turned off for your association by default.
We will not train an association-specific model or member type models for your association yet.
If you prefer, we can turn on the feature for you, and your Member Health predictions will be created, but will come from the general models only, which are trained on all associations’ data and will not reflect association-specific behavior.
This means you could still receive Member Health predictions, but they will be based on broader patterns across associations rather than patterns unique to your organization.
Why do these requirements exist?
These minimums are in place to protect you from misleading results. With too few expired or renewing examples, a tailored model can “overfit,” meaning it learns quirks in small samples rather than real signals. The result can look confident while being inaccurate.
What happens when your association becomes applicable?
As more membership history accumulates, your association can become eligible for association-level and member-type modeling. At that point, predictions can better reflect your organization and your membership structure. Member Health requirements are reviewed at the beginning of every month when new predictions are made, and Member Health will be turned on automatically when applicable.
If you want help understanding where your association stands, contact support and we can review your current eligibility and what would be needed to qualify.