Why healthcare demos succeed but production fails
- A social thread argued demos work on clean data but production fails because EHR workflows are opaque and bespoke. - The thread emphasised embedding engineers to support undocumented routes like SharePoint queues and hidden queues. - That highlights the need for demos that surface real-world exceptions, routing, and data hygiene rather than toy datasets (x.com)
Healthcare demos often work because they run on tidy test data; production fails when hospital work actually moves through custom Electronic Health Record, or EHR, routes staff built over years. (healthit.gov) The federal government’s main health information technology rule, HTI-1, took effect on February 8, 2024, and added new transparency requirements for predictive algorithms inside certified health software. It also moved the baseline interoperability data standard to United States Core Data for Interoperability Version 3 on January 1, 2026. (federalregister.gov, healthit.gov) Those rules cover certified software, but they do not erase the local workarounds inside hospitals: nursing workflow studies still describe EHR screens as fragmented and disorganized, with documentation spread across fields, flowsheets, and duplicate steps. In one 2022 quality-improvement study, redesigning a nursing reassessment workflow cut EHR time by 18.5% and reduced documentation steps by 88% to 97%. (pmc.ncbi.nlm.nih.gov) That gap helps explain why buyers now ask harder questions about workflow, not just model quality. Bain and KLAS said on October 9, 2025 that 70% of providers and 80% of payers had an artificial intelligence strategy in place or in development, while providers were prioritizing revenue cycle management and clinical workflow with demonstrable return on investment. (bain.com) Vendors and platform companies are also describing the same bottleneck in plainer terms: the hard part is not generating an answer, but getting data in and actions back out across each hospital’s mix of systems. Microsoft said in a June 26, 2025 engineering post that connecting advanced AI systems to EHRs involves “significant challenges” because technical environments, data formats, and interoperability support vary by institution. (techcommunity.microsoft.com) In practice, that means a polished demo can skip the inbox no one documented, the SharePoint list one department still uses, or the hidden work queue that decides whether an order, referral, or prior authorization gets touched at all. Those routes may sit outside the clean Fast Healthcare Interoperability Resources, or FHIR, path that product teams show on stage. (techcommunity.microsoft.com, healthit.gov) The social thread that sparked the discussion framed the fix as embedding engineers with operations teams until those routes are mapped, not assuming the EHR is the whole workflow. The thread pointed to undocumented queues and side systems as the place where live deployments break even when demos look finished. (x.com) A stronger demo, then, is less polished and more forensic: it shows missing fields, misrouted tasks, duplicate records, and the handoff back to staff when confidence is low. In healthcare, the product that survives production is usually the one that exposes the mess before go-live, not after it. (pmc.ncbi.nlm.nih.gov, bain.com, x.com)
