Why Healthcare AI Initiatives Fail – The Data Engineering Problem Few Talk About
This five-part blog series from Jeffrey Eyestone, Chief Strategy and AI Officer at P-n-T Data Corp., explores why healthcare’s data sharing infrastructure is inadequate and how a new, agent-based model is helping providers, payers, and partners move beyond security and scalability limitations in order to improve key data-centric workflows like claims and prior authorization adjudication, AI and analytics, and more. Drawing on real-world insights from AI deployments and next-gen platform design, the series explores the link between healthcare data sharing challenges and improved data engineering.
In my last post, we highlighted high-value healthcare workflows that are only possible when you move beyond legacy data exchange solutions. But there’s another reality most people won’t say out loud: many healthcare AI initiatives fail because of poor data engineering. Healthcare AI obviously holds tremendous promise, but we are also hearing about numerous failed AI initiatives and viable AI models being shelved due to poor data access, quality, transformation, and security.
The Unstructured Data Trap
One of the biggest roadblocks is the Healthcare AI industry’s fixation on unstructured data. Processing unstructured data like faxes, PDFs, and EMR output (charts and more) is expensive and error-prone. Adjudication systems and AI & Analytics solutions perform much better with structured data like EDI to meet data quality standards.
For example, consider prior authorization AI tools that aim to help summarize PAs and help with adjudication, generate denial letters or requests for info, and provide data to downstream utilization management process (UM) or appeals and grievances (A&G): if these AI solutions are built on unstructured data, they regularly underperform or outright fail. Structured-data approaches such as the 278 + 275 transaction set for auths and attachments are rising in popularity. Spending energy on this structured data approach is also a smart move given regulatory requirements like CMS-0057, payer-specific requirements (larger payers are starting to eliminate faxes and more), and even individual state requirements for prior auths.
The Data Engineering Foundation AI Actually Needs
Successful healthcare AI isn’t just about better models. It’s about better data pipelines and data engineering. Post-n-Track Gen 3 delivers real-time, structured, and secure data without long-term residency — the exact conditions AI needs to thrive:
- Scalable Data Access and Delivery: Real-time availability without storage exposure
- Data Quality Management: Intelligent edits, validation agents, and rules engines
- Data Transformation: Universal format support and standardization across systems
- Data Governance: Full audit trails and compliance automation
- Intelligent Orchestration: Agent-based capabilities that optimize workflows automatically
AI Use Cases That Actually Work
When AI is paired with the right data engineering architecture, the potential is enormous:
- Adjudication and reconciliation AI use cases benefit from more structured data. Get rid of faxes!
- Federated learning networks that don’t require central data pooling or risk exposure
- Real-time predictive analytics with streaming data
- Clinical decision support embedded directly into provider workflows
- Payment integrity AI with advanced fraud detection, duplication checks, and real-time scoring
- Population health analytics orchestrated without long-term data residency
Building AI on Bedrock, Not Sand
It’s tempting to think AI success is about more data scientists or more powerful models. But without an agent-based, zero-residency data sharing platform as the foundation, AI in healthcare is like building a skyscraper on sand. Instead of burning calories on training AI to help with unstructured data, devote energy to getting the data you need via a structured data sharing platform with exceptional data engineering capabilities.
Post-n-Track Gen 3 enables AI to finally deliver on its promise in healthcare.
Next in the series, we’ll look at how organizations are using Gen 3 to gain competitive advantage in data sharing, compliance, and innovation.
Jeffrey Eyestone is Chief Strategy and AI Officer at P-n-T Data Corp.