Use 1 consumer or patient risk model per audience for best results

Use one predictive risk model per audience the same way you build an audience for a single channel (for example, an email list). Why? A single risk model keeps targeting focused, platform performance fast, and outcomes measurable.

Learn more about this core rule — one risk model per audience — and how to handle multi-goal campaigns, combine consumer and patient audiences, balance audience sizes, work with models, and avoid common pitfalls.

1. Guiding principle: one model per audience
   
   1. Keep it focused. Select a single risk model that aligns with the audience’s primary goal (e.g., patient risk for behavioral health, or a consumer risk for retention). Avoid selecting many overlapping models for the same audience because that dilutes signal and obscures impact.
   2. Performance benefits. Single risk model audiences run faster in-platform, are simpler to analyze, and are more transparent from a data science perspective (each audience represents one clear objective).
       
2. When campaigns have multiple goals...
   
   1. Union audiences, not models. If your campaign covers distinct goals, create separate audiences for each goal and union them where appropriate. For example, run Audience A (Model A) for retention and Audience B (Model B) for acquisition, and then union audiences in campaign targeting if you need combined reach.
   2. Prefer a general service-line model over many narrow ones. If goals are closely related (e.g., multiple services within a line), it’s often better to use a single, slightly broader model that covers the whole service line than to stitch together multiple narrow models that overlap.
       
3. Combining Consumer and Patient audiences
   
   1. You can and often should combine consumer and patient audiences for the same goal/topic. Pooling consumer and patient segments is valid and recommended when the modeling objective overlaps (for example, awareness or screening campaigns).
   2. Balance by regional market share. Patient models will generally cover more of the primary market areas because of available patient-level data; Consumer models tend to pick up secondary and tertiary areas. Adjust the relative size of each audience to reflect real market share and campaign objectives.
   3. Adjust score thresholds. Use score ranges to tune the mix — increase or decrease the score cutoffs to make one audience larger or smaller relative to the other.
       
4. Fully mastered vs. unmastered records
   
   1. Fully mastered records (learn more)  take precedence. When combining fully mastered and unmastered records (e.g., email address only), the platform will prioritize fully mastered records and dedupe on shared emails/phones.
   2. Verify applicable criteria. When you multi-select model types or audience sources, check the platform’s “applicable criteria” panel. If you see an audience size of zero, some selected criteria may be incompatible.
   3. Keep selection rules consistent. If you intentionally collect looser identifiers (email/phone) for broader reach, recognize the trade-offs in match rate and deduplication behavior.