Healthcare documents · 29 of 31 · Evidence · 2 minutes
Healthcare products. From problem to production.
Mistakes & lessons
Claims automation, provider matching and a legacy claims integration replaced — explored through the documents and decisions.
01
Lessons across automation, identity and migration
Lessons I can defend with the analysis behind each one.
Product decision
Do not let a local success hide an end-to-end failure.
- Document type
- Mistakes & lessons
- Lifecycle stage
- Operate & learn · 21 / Operate & learn
- Contents
- 3 sections · 1 table · about 1 min
- Status
- Sample format · headline results from my healthcare claims work
- Product decision
- Do not let a local success hide an end-to-end failure.
Portfolio sample from my healthcare product experience. Requirements, thresholds and review records are illustrative; they are not employer documents or historical sign-offs.
Experience-based synthesis
Document conversion is only one part of claims automation; provider completeness and representation decide identity; a platform replacement needs continuity as well as capability; and a confident estimate with no data behind it is a risk, not a plan.
- Document conversion is only one part of claims automation
- provider completeness and representation decide identity
- a platform replacement needs continuity as well as capability
- a confident estimate with no data behind it is a risk, not a plan
Tempting choices and corrections
| Tempting choice | What it would have cost | Correction |
|---|---|---|
| Treating conversion rate as the outcome | Review work hidden behind a rising rate | Report net automation with the manual share beside it |
| Dropping the provider match rule | Tens of thousands of extra manual fallouts | Measure a month with and without; fix identity at the source |
| Assuming streaming throughput | A design built for six claims a second that handled one | Demand the proof of concept |
| Patching a legacy engine | Years of work still short of the volume | Replace limited legacy EDI shredding with homegrown mapping into a normalized datamart before adjudication. |
| Counting processed claims as received | 300,000 claims and about $10M unseen | Daily reconciliation against what was received |
What changed in my framing
I assess the complete operating path: eligible automation, trustworthy identity, visible state and recoverability. Manual review is a deliberate capability for uncertain evidence, and every rate is reported with its denominator.
- eligible automation
- trustworthy identity
- visible state
- recoverability