Pharmacists were doing detective work the system could do for them. We moved detection to the machines and the judgment back to the people.
Adverse drug events in the weeks after a hospital discharge are one of the largest preventable causes of readmission in Australian health care. Hospital pharmacists are the safeguard. The problem isn't competence. It's that the system loads them with information work they don't have time to do thoroughly.
A discharge medication review requires reconciling a patient's history across multiple data systems — what was prescribed in hospital, what was being taken at home, what's been dispensed, what's been changed, what's been stopped. Each of those answers lives in a different place. None of them is integrated.
So pharmacists do detective work. They search dispensing records. They reconcile changes. They check for interactions. They look for patterns that suggest non-adherence. And they do it under ward time pressure, while also delivering clinical advice, patient counselling, and coordination with prescribers — the work only they can do.
The result is predictable: the patients who need the deepest review are the same patients who are most likely to slip through it. High-risk discharges are missed not because pharmacists aren't capable, but because the system asks them to be detectives first and clinicians second.
The Sandpit method asks where each of the four operations should sit. For AutoMedic, the answer was: keep risk-carrying and the clinical call exactly where they are — with the pharmacist. Move noticing and the first layer of recognition to the system that already has the data.
Pharmacists noticed potential issues by manually reviewing patient charts during ward rounds. With high caseloads, many risks weren't flagged until after a problem had escalated.
AI continuously scans dispensing data, prescribing records and EMR context. High-risk patients are surfaced for review before the pharmacist begins their list — every shift, every ward.
Pattern recognition relied on individual experience, working across fragmented data. Cross-referencing took minutes per patient — and was often skipped under time pressure.
The system surfaces a structured patient timeline with risk flags. The pharmacist sees the integrated picture in seconds: what changed, what's missing, what's risky.
Time pressure forced informal triage. Highest-risk patients sometimes received less attention than those who happened to be flagged earlier in the round.
Risk-stratified workflow follows the evidence. Pharmacist attention concentrates where the algorithmic detection shows it should — supporting, not replacing, clinical judgment.
Pharmacists made the call on incomplete information, under time pressure. Many issues weren't escalated until after the patient had been readmitted.
The clinical decision stays with the pharmacist — that doesn't move. But it lands earlier, with better evidence, on more patients per shift, and before discharge complications occur.
AutoMedic is deployed across six South Australian public hospitals as part of an MRFF national infrastructure investment. The platform integrates with state EMR systems and dispensing data feeds, surfacing prioritised review lists directly into the pharmacist's workflow.
Every component operates under formal information governance and clinical safety frameworks. The system supports the pharmacist's existing clinical authority — it does not generate clinical recommendations, and it does not act without human review.
The numbers below are projected at full pilot scale across six SA public hospitals — not yet finalised in published evaluation. Where outcomes are projected rather than verified, we say so.
The same configuration logic that worked upstream — moving knowledge to consumers in EndoZone — works inside a hospital ward. Different domain. Different actors. Different workflow. Same method.
Each Sandpit configuration applies the same four-operation decomposition to a different capacity problem. What stays constant is the method.
Bring it to the Sandpit. We'll diagnose where capacity is lost, configure what should change, and test it in a governed live setting before you commit at scale.