Healthcare documents · 3 of 31 · Evidence · 7 minutes
Healthcare products. From problem to production.
Claims automation, provider matching and a legacy claims integration replaced — explored through the documents and decisions.
01
A press release, customer FAQ and hard internal questions about value, sourcing, AI and operating risk.
Product decision
Promise less manual handling while preserving evidence and review.
Portfolio sample from my healthcare product experience. Requirements, thresholds and review records are illustrative; they are not employer documents or historical sign-offs.
Claims teams gain a supported intake path from CMS-1500 and UB-04 documents to validated electronic handoff. Operators can inspect the source, correct uncertain fields with authority, understand provider resolution and see the claim’s actual downstream state. Eligible work advances without avoidable re-entry; unresolved work has a reason and an accountable queue. This release narrative describes the reconstructed product commitment. It does not announce that the separate provider and platform projects shipped together.
One conversion. Two connected outputs.
saved per claim
in manual processing
claims per month
of manual claims avoided
Select any step to explore. Form layouts are simplified.
Incoming paper claims include CMS-1500/UB-04 forms and a primary-payer Explanation of Benefits (EOB). CMS-1500 professional claims convert to EDI 837P; UB-04 institutional claims convert to EDI 837I. The primary-payer EOB is an input, distinct from the EOB generated after adjudication. The simplified forms contain no patient data.
Process explanation: Confirm the form type and legibility before extraction.
OCR and AI extract the information from the paper forms for conversion.
Process explanation: Check incomplete or uncertain values against the source image before applying conversion rules.
Custom conversion rules create EDI files alongside image and index files.
Process explanation: Validate required values, apply the corresponding 837 format and route generation errors for review.
The image and index share a common claim ID in OnBase. Stakeholders can retrieve the original claim for appeals or re-review. Storage supported the project's stated 10-year retention requirement.
Process explanation: Keep the image and index associated, and surface a failed lookup or storage handoff.
Professional claims use EDI 837P; institutional claims use EDI 837I. The files go to the claims processor for adjudication.
Process explanation: Confirm receipt and investigate missing or rejected acknowledgements. File acceptance is distinct from adjudication.
The claims processor adjudicates the claim, followed by payment, invoice and Explanation of Benefits (EOB) outputs. The generated EOB is distinct from the primary-payer EOB received with the incoming paper claim.
Process explanation: Processing and its downstream outputs follow the receiving workflow. This conversion project is separate from the claims-engine replacement.
The benefit is not simply a document converted to structured data. It is a claim that reaches the next legitimate state with less manual work and enough evidence to recover when something fails. Claims operations gains a usable exception path. Customer care gains status language it can explain. Provider-data teams gain actionable mismatch reasons. Integration consumers gain valid transactions and attributable acknowledgements. My claims experience converted more than 400,000 paper claims to EDI 837, redeployed an 80-person team and saved about $1M a year; the provider API improved latency/reporting qualitatively. No combined business result is implied.
No combined business result is implied
Answers describe the release boundary.
| Question | Answer |
|---|---|
| Which claims are supported? | Defined CMS-1500 professional and UB-04 institutional cohorts, producing the applicable 837P or 837I. |
| Will OCR approve a claim? | No. Extraction and validation do not decide coverage, clinical necessity or final adjudication. |
| What if the provider cannot be identified? | Return missing, no-match, ambiguous or stale status and route authorized review. |
| What happens after submission? | Report submission, acknowledgement and final disposition separately. |
| Can I correct an error? | Authorized corrections retain source history and re-enter validation. |
Why partner for extraction? Supported-form expertise and document-processing capability can reduce initial implementation work, provided the vendor contract exposes uncertainty, evidence and recoverable errors. Why build the provider interface? A central resolution contract can make normalization and failure semantics consistent for consumers. Why not replace the integration platform at the same time? A platform build changes operational ownership and consumer behavior; it needs its own cost, continuity and migration decision. The 0→1 replacement experience informs that decision without being treated as an OCR dependency.
OCR assistance proposes extracted values; it cannot supply missing clinical facts or choose an ambiguous provider merely to increase throughput. A component’s output must be attributable and testable. Model quality, form coverage and review workload affect total value. A more sophisticated extraction model is not the default remedy for address normalization, stale reference data or an invalid receiver contract. Deterministic checks and authorized judgment remain explicit even when extraction uses learned models.
A more sophisticated extraction model is not the default remedy for address normalization, stale reference data or an invalid receiver contract
The economic unit is completed medical claims, including extraction, compute, support and manual correction cost. Track pending backlog so cheaper completed work does not hide costly unresolved claims. Principal risks are false identity, rejected transactions, duplicates, lost inventory and inappropriate access. Release evidence must cover Q1–Q12, the p95 ≤500 ms at 50 requests/second sample API contract, operator acceptance and recovery ownership. Operations, care, legal/compliance and relevant patient-safety partners assess the boundaries they own. High automation volume cannot waive an unexplained identity or inventory defect.
High automation volume cannot waive an unexplained identity or inventory defect
If extraction corrections dominate, inspect source quality and field review before changing infrastructure. If provider exceptions dominate, prioritize evidence and normalization rather than a more elaborate extraction model. If acknowledgements stall, inspect the receiver path. If replay disagrees, stop migration promotion. The weekly business review must connect each signal to an owner and a follow-up decision.