Healthcare AI conversations gravitate toward clinical decision support — diagnostic assistance, imaging review, triage. The marketing decks are full of those examples. We think those examples are wrong as a starting point.
Clinical decision support is the highest-stakes possible use of AI in healthcare. The regulatory bar is appropriately high. The change-management requirement is enormous. The clinical staff have valid reasons to be skeptical. The deployment timeline is measured in years, not weeks. None of that is a reason to avoid clinical AI eventually — but it's a strong reason not to start there.
The largest, lowest-risk operational lift in healthcare is in the administrative layer that surrounds care. Intake summarization. Prior-authorization drafting. Internal documentation. Scheduling reconciliation. Records-request response. The volume of email that pulls clinical staff away from clinical work. This is the layer where AI delivers measurable value within weeks instead of years.
It's also the layer where the data sensitivity is the same as the clinical data — it's all PHI — but the accuracy bar is meaningfully different. A summarization that needs review by a human is acceptable. A diagnostic suggestion that needs review by a human is the entire problem. The first is operational; the second is clinical. The mistake is treating them as the same engagement.
When we deploy in healthcare, we are deliberate about the boundary. The PH22 handles operational AI work that surrounds clinical care. Clinical decision support is not where we start. It may be where some customers go eventually — with the appropriate clinical, regulatory, and change-management investment. But the operational lift is the work that justifies the deployment, and it's the work where we can show measurable impact in the first quarter.