Building Pack: A Drug Pricing Benchmark for the Pharmacy Industry
Built a machine learning benchmark to forecast drug price movements after noticing a blind spot in pharmacy pricing data.
- Predictive Analytics
- Healthcare Data
- Pricing Models
- Machine Learning
Anupam Dubey
Director at AKDesign
Overview
Anupam co-built Pack, a machine learning benchmark that forecasts drug price movements, after spotting a blind spot in pharmacy pricing data during a consulting engagement. The model now underpins pricing decisions for the largest players in US pharmacy benefits.
The Story
Anupam came into a consulting engagement with a pharmacy benefits manager working on maximum allowable cost optimization. He and a colleague noticed loose correlations between publicly available pricing data points, including AAC, SMAC, WAC, AWP, and NADAC figures, and suspected there was an unaddressed opportunity to model actual drug costs more accurately.
Drawing on a data science background built at FICO, where he worked on credit scoring and fraud detection models, he began building out variables from utilization data, demand-supply patterns, and co-dispensing relationships between drugs. He tested the model repeatedly, gathered more data, and refined the benchmark until it held up against real pricing behavior.
He and his co-founder took the model to their PBM client, who validated its value, and expanded from there. The resulting product, Pack, is now used daily by 19 of the 20 largest PBMs, 9 of the 10 largest pharmacies, health plans, wholesalers, and pharmaceutical manufacturers, with plans covering over 100 million lives relying on it.
The nation's largest PBM uses Pack for automated claim adjudication, meaning the model is not only informing pricing decisions but executing them directly.
Ownership Snapshot
Broad role
Data scientist and co-builder of a pharmacy pricing benchmark product.
Goal
Bring transparency to opaque drug price movements using machine learning.
Direct ownership
Built the predictive model itself, selecting variables and validating the benchmark against real client data.
Team execution
Worked directly with the co-founder to test, refine, and bring the model to market with PBM clients.