Does Orita build one cross-channel propensity model, or a separate model per channel (email, SMS, direct mail, etc.)?

Orita builds a separate, dedicated model for each channel. The question Orita answers is conditional and channel-specific: "If I send this person an email, will it drive them one step closer to purchase?" and separately, "If I send this person an SMS, will it drive them one step closer to purchase?"

This is because each channel has its own signal-chain strength. For example, SMS may have a tighter downstream correlation to purchase, whereas email engagement plays out over a longer time period. A single cross-channel propensity score would blur those dynamics, so Orita deliberately trains per-channel models to produce more accurate channel-specific segmentation.