The Pitch Deck vs The Terminal
We've built or worked on seven HealthTech products in the last three years — teleconsultation platforms, diagnostic aggregators, EHR systems, and AI-assisted clinical tools. We've seen the pitch decks that get these companies funded, and we've seen the terminal where the actual systems run. The gap is... significant. This isn't about any specific company — it's about systemic challenges in HealthTech that we keep encountering.
Interoperability Is a Fantasy
Every HealthTech pitch deck mentions "seamless interoperability" with existing hospital systems. The reality: most Indian hospitals run on one of a handful of HIS (Hospital Information Systems) — TCS iON, Practo, eHospital, or custom-built Access/Excel-based systems. None of them have reliable APIs. The ones that do have APIs use HL7v2, a messaging standard from 1987 that transmits data in pipe-delimited text strings. Integrating with these systems is an exercise in parsing fragile, poorly documented message formats and handling the inevitable inconsistencies between hospitals.
FHIR (Fast Healthcare Interoperability Resources) is the modern standard that should solve this, and ABDM (Ayushman Bharat Digital Mission) mandates FHIR for health data exchange in India. But adoption is nascent. Of the hospitals we've integrated with, exactly two supported FHIR endpoints, and both had significant deviations from the spec. We now maintain a hospital-specific adapter layer with custom mapping logic for each integration. It's ugly, it's expensive, and it's the only thing that works.
The Data Quality Problem
AI in healthcare is only as good as the data it trains on. And Indian healthcare data quality is, frankly, terrible for ML purposes. We've encountered: handwritten prescriptions scanned as images (OCR accuracy on doctor handwriting: about 60%), diagnostic reports with inconsistent units (mg/dL vs mmol/L for the same test at different labs), patient records with incorrect demographic information (wrong age, wrong gender — we once found a male patient record with obstetric history), and duplicate patient records because there's no universal patient identifier in India.
ABHA (Ayushman Bharat Health Account) is supposed to be the universal identifier, but adoption is still low. For our clinical AI products, we spend 40-60% of our data engineering budget on data cleaning and normalization. This is rarely mentioned in investor presentations, but it's often the difference between a model that works and one that's dangerous.
Regulatory Risk Is Underpriced
We wrote a separate post about AI in HealthTech regulation, but the business risk deserves mention here too. India's regulatory framework for digital health is evolving rapidly. The DPDP Act, the Digital Health regulations, CDSCO's stance on Software as Medical Device — all of these are works in progress. A product feature that's compliant today might require a license tomorrow. We've seen companies build entire products around features (like AI-based diagnosis) that are one regulatory notification away from needing pre-market approval.
Our approach: build every health AI feature as "decision support" rather than "decision making," maintain audit trails for every AI inference, and design the architecture so AI features can be toggled off per jurisdiction without breaking the product. It's defensive engineering, but in healthcare, the downside of regulatory non-compliance is existential.
What Actually Works in Indian HealthTech
Despite these challenges, we've seen genuinely impactful HealthTech deployments. The products that work share common traits: they solve a specific workflow problem (not "healthcare platform" — more like "reduce lab report turnaround time"), they work with existing systems rather than trying to replace them, they have a clear monetization path that aligns with how Indian healthcare payments work (OPD packages, insurance claims, or out-of-pocket), and they're designed for the low-bandwidth, multilingual reality of Indian clinics, not for a San Francisco conference demo.
The biggest opportunity we see right now is in the digitization of Tier 2-3 city diagnostics. India has over 100,000 diagnostic labs, and the majority still use paper-based workflows. The winner in this space won't be the one with the best AI — it'll be the one that builds the simplest, most reliable software for lab technicians who've never used anything other than WhatsApp and Excel.