health · digital · family: the model was never trained on this
faxes to approve a scan
Automated extraction and matching for healthcare prior authorization
Problem statement
Healthcare prior authorization — the process by which insurance companies approve medical treatments before they are delivered — requires extracting clinical information from unstructured documents (physician notes, faxes, scanned PDFs, lab reports) and matching it against complex, insurer-specific rule sets that change frequently. Olive AI raised $902 million and, by its own count, was deployed in more than 900 hospitals to automate this and related revenue-cycle processes, but a 2022 Axios investigation found the company inflated its capabilities and in many cases delivered only a fraction of the savings it pledged. The company announced in October 2023 that it was winding down, selling its clearinghouse and patient access business to Waystar and its prior authorization business to Humata Health. The problem persists: prior authorization consumes the equivalent of about 12 hours of physician and staff time per physician per week (AMA physician survey, 2024) — and current approaches still depend heavily on human labor.
Why this matters
Administrative costs account for an estimated 15–30% of US health spending — roughly $950 billion per year by one 2021 estimate (Health Affairs policy brief, 2022). Prior authorization alone is estimated to account for $35 billion of US healthcare administrative spending per year (Sahni et al., Health Affairs Scholar, 2024). The American Medical Association's 2024 survey reports that 94% of physicians say prior authorization delays access to necessary care, and 24% report that prior authorization has led to a serious adverse event for a patient. Medicare Advantage insurers alone made nearly 53 million prior authorization determinations in 2024 (KFF), so even modest automation improvements could free billions in healthcare spending and reduce treatment delays. The problem is not shrinking: in Sahni et al.'s 2023 survey, 65% of provider respondents and 79% of private payer respondents reported growth in prior authorization volumes over the previous three years, and KFF reports that Medicare Advantage requests rose steadily from 2021 to 2024, although the number of requests per enrollee stayed roughly constant.
What’s been tried and why it hasn’t worked
Current and former employees told Axios that Olive's pharmacy and supply-chain automation relied on screen scraping — bots that pull information from a PDF or webpage — rather than AI or machine learning, and that the only automation work that meaningfully involved AI was prior authorization, a capability Olive acquired when it bought Verata Health in 2020; the company said it used robotic process automation (RPA) to reach some payer portals when APIs were not available. Prior authorization resists automation for several reasons: (1) clinical documentation is highly heterogeneous — physician notes use inconsistent terminology, abbreviations, and formatting across health systems; (2) insurer rule sets are complex, ambiguous, and change often, meaning the matching logic requires constant updating; (3) many critical clinical details are buried in unstructured free-text notes, scanned faxes, or handwritten annotations that defy reliable extraction; (4) edge cases are common, not rare — prior authorization decisions often hinge on nuanced clinical judgment that rule-based systems can't capture. Sahni et al. note that electronic prior authorization can shorten decision times but has not produced the expected reduction in provider burden or form-filling cost. Providers' intended adoption of AI for the task is low: in their 2023 survey, 11% of provider respondents said their organizations planned to incorporate AI into the process in the next three to five years (against 65% of private payer respondents), with lack of budget and lack of trust in the technology the leading barriers on the provider side.
What would unlock progress
Progress requires advances on two fronts: (1) reliable information extraction from messy clinical documents — a system that can parse physician notes, fax images, lab reports, and scanned PDFs with high enough accuracy that a human doesn't need to verify every extraction; and (2) a computable, standardized representation of insurer authorization criteria that can be updated automatically when rules change, rather than requiring manual reprogramming. The FHIR (Fast Healthcare Interoperability Resources) standard and CDS Hooks specification provide some infrastructure for structured clinical data exchange. Adjacent fields: legal document analysis (extracting structured facts from unstructured legal text), regulatory compliance automation (matching documents against changing rule sets), and claims processing in property/casualty insurance (similar document heterogeneity challenges).
Entry points for student teams
A student team could: (1) build a prototype clinical document parser that extracts structured prior-authorization-relevant fields (diagnosis, procedure, clinical justification, supporting evidence) from a corpus of de-identified physician notes, measuring extraction accuracy by document type; (2) design a computable representation of prior authorization criteria for a specific insurer and procedure category, demonstrating how rule changes could be encoded and applied automatically; (3) develop a clinical NLP pipeline that matches extracted document information against authorization criteria and flags cases requiring human review, with the goal of maximizing the percentage of cases that can be auto-adjudicated. Relevant disciplines include natural language processing, health informatics, software engineering, and human-computer interaction.
Genes
Structural cousins: stuck the same way, in other fields
Sources
"Olive AI is shutting down," Axios Pro, Oct 31 2023, "Health AI startup Olive to shut down," Healthcare Dive, Nov 1 2023, "$4 billion health tech startup Olive overpromises and underdelivers," Axios, Apr 5 2022, AMA, "AMA survey indicates prior authorization wreaks havoc on patient care," Jun 18 2024, Sahni et al., "Perceptions of prior authorization burden and solutions," Health Affairs Scholar 2(9), 2024, Accessed 2026-08-20. go to source 1 ↗ go to source 2 ↗ go to source 3 ↗ go to source 4 ↗ go to source 5 ↗
Working notes
Notes made while writing and checking this brief, shown as written. They use working shorthand.
- Olive AI ($902M raised since 2012, $4B peak valuation — both per the company's July 2021 funding announcement) is among the most expensive failures in healthcare automation. Waystar acquired the clearinghouse and patient access business; Humata Health (led by Jeremy Friese, Olive's former president) acquired the prior authorization technology, suggesting the underlying IP had some value.
- The failure is best characterized as `failure:ignored-context` — the system was designed for structured, predictable inputs but deployed against messy, heterogeneous real-world documents. Also `failure:disciplinary-silo` — the NLP/AI capabilities were disconnected from deep clinical workflow understanding.
- Related to healthcare administrative cost problems broadly. A federal rule that took effect in April 2024 is intended to improve the use of electronic prior authorization and applies to Medicare Advantage and certain other insurers, with requirements phasing in from January 2026 (KFF); it may create both new infrastructure and new data standards that change the problem landscape.
Reconciliation 2026-08-20: The original Source line cited two SEO-blog venues ("SunsetHQ" and "OyeLabs") for the Olive AI shutdown facts; these were replaced with verified trade/wire coverage — Axios (which broke the wind-down story, Oct 31 2023), Healthcare Dive (Nov 1 2023), and the April 2022 Axios investigation. Corrections made against verified sources: (1) "34 hours per physician per week" on prior authorization was wrong — the AMA's 2024 physician survey (released Jun 18 2024) reports the equivalent of 12 hours of physician and staff time per physician per week (~two business days), 43 prior authorizations per physician per week; body corrected. (2) AMA percentages corrected to the 2024 survey: 94% report care delays (was "93%"), 24% report a serious adverse event (was "34%"). (3) "$902M raised / 900+ hospitals / $4B valuation" verified against Olive's July 1 2021 funding announcement (PR Newswire, https://www.prnewswire.com/news-releases/olive-hits-a-4-billion-valuation-with-400-million-of-capital-led-by-vista-equity-partners-301323850.html) — the 900-hospitals figure is Olive's own claim, now attributed as such. (4) "Administrative costs ~30% / $1 trillion" softened to the verified Health Affairs range (15–30% of health spending; ~$950B in Sahni et al.'s 2021 estimate; https://www.healthaffairs.org/do/10.1377/hpb20220909.830296/). (5) "$35 billion" prior-auth cost verified to Sahni et al., Health Affairs Scholar 2024, and now cited. (6) "over 2 billion prior authorization requests annually" could not be sourced; replaced with the verified KFF figure of nearly 53 million Medicare Advantage prior authorization determinations in 2024 (https://www.kff.org/medicare/nearly-50-million-prior-authorization-requests-were-sent-to-medicare-advantage-insurers-in-2023/). (7) The CommonSpirit-terminated-its-Olive-contract bullet traced only to the removed SEO blogs and could not be verified in real coverage; removed. (8) "Investigations revealed the AI required extensive manual intervention" restated to what the verified Axios investigation actually found (inflated capabilities, fraction of pledged savings). Humata Health leadership (Jeremy Friese, ex-Olive president) verified via Axios/Fierce Healthcare, Jun 2024.
Reconciliation 2026-09-30: Fact repair after reading the whole write-up against every source it names, all fetched on 2026-09-30; a second reader went over the repairs on the same day and this note describes the final state. Read in full: the Healthcare Dive report of Nov 1 2023 (https://www.healthcaredive.com/news/olive-ai-shuts-down/698455/); the AMA news release of Jun 18 2024 (https://www.ama-assn.org/press-center/ama-press-releases/ama-survey-indicates-prior-authorization-wreaks-havoc-patient-care); Sahni et al. 2024, read as the open full text at Europe PMC (PMC11425057; the publisher page itself returned a robot check); the Health Affairs policy brief of Oct 6 2022, read from the Internet Archive copy of May 25 2024 (the publisher blocks automated access); the KFF analysis (the link now redirects to KFF's Jan 28 2026 update, "Medicare Advantage Insurers Made Nearly 53 Million Prior Authorization Determinations in 2024"); Olive's July 1 2021 funding announcement on PR Newswire; and the two Axios Pro articles (Oct 31 2023 and Apr 5 2022), which are paywalled on screen but whose full article text is carried in the Internet Archive copies of the pages, where the second reader read it. (1) Display title changed from "Thirty Faxes to Approve a Scan" to "Faxes to Approve a Scan": no source gives a count of faxes per approval. (2) "The company wound down in October 2023" now reads that the company announced in October 2023 that it was winding down, which is what Axios (Oct 31 2023) and Healthcare Dive report. (3) "roughly two business days" removed: the AMA release gives 12 hours of physician and staff time per physician per week and does not convert it to days. (4) The sentence that the problem is worsening because insurers keep adding requirements was not in any source; it now gives what the sources say: 65% of provider and 79% of private payer respondents in Sahni et al. reported growth in prior authorization volumes over the previous three years, and KFF reports Medicare Advantage requests rising steadily from 2021 to 2024 while requests per enrollee stayed roughly constant. (5) "change quarterly" for insurer rule sets softened to "change often"; no source gives a cadence. (6) The sentences naming R1 RCM, UiPath, Waystar, Epic and Cerner as competitors whose products need human intervention or lack trusted accuracy were not in any source; replaced with Sahni et al.'s findings that electronic prior authorization has not produced the expected reduction in provider burden, and that 11% of provider respondents (against 65% of private payer respondents) said their organizations planned to incorporate AI into the process in the next three to five years, with lack of budget and lack of trust the leading provider barriers. (7) "but most prior authorization still happens outside these standards" removed; no source says so. (8) The Source Notes bullet named the CMS Interoperability and Prior Authorization Final Rule (CMS-0057-F, effective 2026); no source gives that name or number. It now says what the KFF analysis says about the rule: effective April 8 2024, intended to improve the use of electronic prior authorization, applying to Medicare Advantage and certain other insurers, with requirements starting in January 2026 and January 2027. (9) The description of Olive's technology (robotic process automation combined with machine learning classifiers that worked for simple cases but failed on complex ones) is not what the April 2022 Axios investigation reports. It now gives what Axios reports: employees said the pharmacy and supply-chain automation relied on screen scraping rather than AI or machine learning, the only work that meaningfully involved AI was prior authorization (acquired with Verata Health in 2020), and the company said it used RPA to reach some payer portals when APIs were not available. The four numbered reasons that follow are the write-up's own analysis and are no longer presented as findings about Olive's system. Confirmed in the sources and left as written: $902 million raised since 2012, more than 900 hospitals and the $4 billion valuation (PR Newswire, Jul 1 2021); the Axios investigation's finding that Olive inflated its capabilities and in many cases generated only a fraction of the savings it pledged (Axios, Apr 5 2022); Waystar's acquisition of the clearinghouse and patient access business, and Humata Health, led by former Olive president Jeremy Friese, who had co-founded and run Verata Health, the prior authorization company Olive bought in 2020 (Axios, Oct 31 2023); the 12 hours per physician per week, 94% reporting delays and 24% reporting a serious adverse event (AMA, 2024); 15–30 percent of health spending and the $950 billion Sahni et al. 2021 estimate (Health Affairs brief); $35 billion for prior authorization (Sahni et al. 2024); nearly 53 million Medicare Advantage determinations in 2024 (KFF); and the Source line citations. Not stated in any source and left as written: that Olive is "among the most expensive failures in healthcare automation" and that the sale suggests the underlying technology had value (both are the write-up's own characterizations); Axios describes Humata's purchase only through a former employee's remark that Verata's former chief executive bought back his company, and separately reports that Olive's payer-facing prior authorization business went to Availity in February 2023. Genome tags untouched.