Pharmacy software's lost decade

Every adjacent industry got rebuilt around AI. Dispensing didn't.

Posted 17 May 2026 · By a Qiri staff contributor · 5 min read

The dispensing system the pharmacist at your local Australian community pharmacy used this morning was architected before broadband internet was standard. The interface paradigms were designed when graphical user interfaces were still a novelty in this category. Menu structures, keystroke patterns, and database schemas have not been substantially reworked in two decades.

This is not an accusation. It is a description.

Australia's major community-pharmacy software vendors trace their origins to the 1980s and 1990s. They built rigorous, robust, regulator-compliant systems on top of the technology of the time. They kept those systems running through two decades of operating-system shifts, two decades of regulatory reform, and a complete rewrite of how PBS claims, e-prescribing, and pharmacist scope-of-practice worked.

What they did not do — what no community-pharmacy software vendor in the world has yet done at scale — is rebuild the core. The clinical reasoning layer most pharmacists rely on today is a stack of alert pop-ups bolted on top of an architecture that predates the public web. Approximately 9 in 10 of those alerts get overridden.¹ That is not a failure of pharmacists. It is what happens when "clinical decision support" is implemented as a layer of warnings on top of a 1990s data model.

What every adjacent industry did instead

The contrast with adjacent industries is hard to miss.

• Retail point-of-sale. Square launched in 2009. Shopify Retail and Lightspeed rewrote what a cash register is. Today an independent café operator can run a million-dollar business on a $200 iPad. The category was rebuilt from scratch.
• Banking and payments. Stripe launched in 2010. Wise and Revolut rebuilt the rails of cross-border transactions. The clearing house was around long before any of them existed. They didn't replace it. They sat alongside it and made it usable.
• Legal. Harvey, Casetext, Lex Machina. Generative AI was not a serious commercial tool four years ago. Today a large share of contract review and case research at AmLaw 100 firms passes through an AI layer.
• Clinical scribe. Abridge, Nuance DAX Copilot, Suki, and from Australia, Heidi Health. Ambient AI listening to a clinical encounter and producing a structured note was a research idea five years ago. Today major US health systems and a growing share of Australian GP clinics treat it as standard-of-care.
• Medical imaging. Aidoc, Viz.ai. Algorithms reading CT scans for stroke and pulmonary embolism at radiology-grade quality, in production, in tens of thousands of hospitals worldwide.

Pharmacy was on this same trajectory at one point. By 2015 every adjacent industry had kept rebuilding. Pharmacy quietly didn't.

And this is not a geography problem. Australian engineering has produced category-defining companies in just about every adjacent vertical. Canva rebuilt design tooling from Sydney into a $40-billion-plus product. Atlassian made Jira and Confluence the global default for software teams. Airwallex rebuilt cross-border payments. SafetyCulture took workplace audit out of clipboards and into every Bunnings, Coles, and BHP site. Heidi Health is right now competing with the biggest names in global ambient clinical scribe and winning. The Australian talent base has rebuilt almost every category it has touched. It just has not touched dispensing.

Why pharmacy was left behind

Three reasons, in order of how much they matter.

1. Switching cost. A dispensing system is the operating system of a community pharmacy. If it goes down, the bench stops. Every pharmacy owner we have spoken to about Qiri has said some version of: "I cannot afford to rip out Fred." Reasonable. The cost of being wrong about a replacement is the pharmacy not opening on Monday. Incumbents are protected not because they are good, but because they are integrated.
2. The claims moat. PBS Online integration, authority workflows, safety-net calculations — these are decades of accumulated edge cases that no greenfield vendor wants to rebuild. Whoever owns the claims pipe owns the relationship.
3. Clinical decision support was treated as a feature, not a paradigm. When the incumbents added drug-interaction alerts in the 1990s and 2000s, they bolted them on as pop-ups over the existing dispensing flow. The architecture never changed. Twenty years later, with override rates above 90%, no one rebuilt the thing. They added more alerts to the existing pop-up system. Each additional alert made the override problem worse.

The net effect: pharmacy got a lot of bolted-on additions and no real rebuild.

What's different now

The cost-of-rebuild has changed. Three reasons.

• You no longer need to rip out the dispensing system to add a real reasoning layer. Open standards — FHIR R4, HL7 v2, and NCPDP SCRIPT in the US — let a new clinical layer sit alongside the existing dispenser and do its job without disturbing the foundation. The same architectural move Stripe made on the bank clearing house, and Abridge made on the EHR, works here.
• AI is finally capable enough to reason about a script, not just flag it. A clinical reasoning engine reads the patient history, drug knowledge, and clinical guideline, and produces a traced decision — not just an alert. The 9-in-10 override problem reshapes when the system surfaces something the pharmacist has not already considered five hundred times this week.
• The audit trail becomes the product, not a byproduct. Indemnity insurers, the Pharmacy Board of Australia,² and pharmacy owners themselves want documented clinical reasoning. The same architecture that produces the reasoning also produces the record.

The dispensing system pharmacies already run keeps running. The reasoning layer is what was missing.

What comes next

Every figure in this article is footnoted to its peer-reviewed source on qiri.ai/sources. If you run an Australian community pharmacy and want to talk about the founding-partner program, reach us here.