AI/ML Platform for Canada's Largest Pharmacy Network
Product design and delivery of a patient-centric health data and AI platform at Loblaw Companies
The Problem
Canada’s pharmacy landscape changed dramatically when provincial regulations expanded pharmacists’ scope of practice to include vaccinations, medication reviews, and treatment of minor ailments. This was a significant clinical and commercial opportunity — but the underlying IT infrastructure wasn’t built for it.
Loblaw’s 1,300+ pharmacy stores each operated decentralized, transactional dispensing systems. There was no unified patient view, no way to measure health outcomes across the network, and no platform to power AI-enabled services at scale. The key questions were:
- How do you unify data from thousands of independent pharmacy systems into a coherent, patient-centric view?
- How do you build an AI/ML platform that enables data scientists to ship clinical features quickly, safely, and repeatably?
- How do you do this under PIPEDA, pharmacy regulatory requirements, and enterprise data governance — without making it impossible to move fast?
Platform Architecture
What I Built and Owned
As Director, Health Data Products & Services, I was accountable for the product strategy, architecture decisions, team delivery, and outcomes. I led a 10-person cross-functional team (ML engineers, data scientists, product owners, clinical analysts) and owned the roadmap from discovery through production.
The platform’s foundation was a unified data architecture built on Google BigQuery, consolidating data from pharmacy transaction systems, public health sources, clinical device vendors, and patient survey platforms into a small set of clean, governed, reusable tables — including a patient identity resolution layer that linked records across stores without a common identifier.
AI Features Delivered on the Platform
The data platform was the foundation — the point was what it enabled. Features shipped on it included:
- NLP-based clinical trial matching — identifying eligible patients from pharmacy records for recruitment
- Medication adherence prediction — ML models surfacing patients at risk of non-adherence for pharmacist intervention
- Vaccination targeting — identifying patients eligible for vaccination by age, chronic condition, and history; queries that previously took hours now ran in minutes and were automated in the visualization platform
Each feature followed a consistent product discipline: discovery with pharmacy operations and clinical partners, explicit success criteria before build, evaluation protocols with defined performance thresholds, and post-launch monitoring.
Outcomes
The platform underpinned the national rollout of new pharmacy programs — Minor Ailments, Diabetes Screening, and Medication Reviews — with documented patient and commercial impact. It replaced thousands of columns of unstructured transactional data with a governed, scalable architecture that cut analytic cycle times dramatically and made previously impossible analyses (like linking lab values to prescription records without a unified patient ID) feasible for the first time.
Governance and Privacy
Sensitive fields are encrypted at rest; data access is governed through Google IAM policies and service accounts. The architecture was designed to meet PIPEDA requirements and enterprise data governance standards — privacy and security as product requirements, not afterthoughts.
One acknowledged limitation: the platform does not adhere to HL7 FHIR interoperability standards. Adopting standard clinical data formats at source systems remains an important future investment for the network.