Which drugs reach the preferred list decides both what a health system spends and how well its patients do. We built the analysis that had not been possible before: every therapy option weighed against the complete patient, assembled from pharmacy, clinical records and claims, on infrastructure that never leaves their walls.
cut from drug spend by selecting for patient outcome
agreement with the reviewing pharmacist on the top-ranked selection
the entire AI stack, inside their own infrastructure
THE PROBLEM
The highest-leverage decision, made on the least evidence
Formulary and preferred drug selection sets what a health system will pay for and, in practice, what most patients will receive. It is among the highest-leverage decisions in pharmacy, and it was being made against a fraction of what was known.
What a patient had been prescribed lived in pharmacy. Their diagnoses, history and outcomes lived in the clinical record. What had already been tried, and paid for, lived in claims. Nothing could read all three together, so selection fell back on unit price, which is the one attribute that says nothing about whether a drug works for the person taking it.
THE ANALYSIS
Every therapy weighed against the complete patient
Unifying pharmacy, clinical records and claims made a comprehensive patient analysis possible for the first time. Therapy options are scored against the full longitudinal picture, so the question shifts from what a drug costs to what it does across a patient journey.
The output is a ranked, evidence-linked view of the options a given patient can actually take, with the supporting record one click away. Across reviewed cases the top-ranked selection agreed with the reviewing pharmacist roughly 95 percent of the time. Alternative drug recommendation runs on the same foundation, which is why a recommendation arrives with its reasoning rather than as an unexplained suggestion.
THE CONSTRAINT
None of it leaves the building
Patient data at this scale cannot be shipped to a vendor to be made analysable, so the entire AI stack was built and deployed on premise, in their environment, on their hardware. That shaped every decision in the build rather than being bolted on afterwards.
It is what makes the rest of it possible: the system can read everything about a patient precisely because nothing about that patient ever leaves.
THE RESULT
Ten to fifteen percent, taken by choosing better
Drug spend came down 10 to 15 percent, and not by buying cheaper. The saving comes from selecting therapies against the full clinical picture, where unit price alone had been the deciding attribute. The analysis that finds the saving is the same one a pharmacist can check, line by line, before anything is acted on.
The analysis behind it had not been possible in their environment before. It is now routine, and every selection carries the evidence that produced it.
Also delivered at Kaiser Permanente
Infrastructure, integration and application, delivered together.
- Full on-premise AI infrastructure, built for their own environment
- Pharmacy, EHR and insurance claims integration into one unified portal
- AI search across the unified patient record
- Comprehensive patient analysis across the longitudinal record
- Formulary and preferred drug selection scored on patient outcome
- Alternative drug recommendation with the supporting evidence attached
- Data handling designed so patient information never leaves their environment
Discuss your next system.
Start with a free, one-hour engineering review. Leave with a practical next step.