Article
One Point Two Seconds to Decide
A claims-flagging system and a claims-denial system are supposed to be two different things, connected by a physician's genuine review. It is easy for the second to quietly swallow the first: the doctor's signature still appears on the file, but nothing about the process checks whether that signatur
- Published
- August 18, 2026
- Updated
- August 19, 2026
- Reading time
- 10 min
The Global Signal
A March 25, 2023 ProPublica investigation, based on internal company data, found that Cigna's PXDX system allowed medical directors to deny claims that did not match Cigna's own preset criteria for a given diagnosis without opening patient files, and that over a two-month period in 2022, Cigna doctors denied more than 300,000 requests for payment at an average pace of 1.2 seconds per case. One medical director named in the reporting, Dr. Cheryl Dopke, is reported to have personally rejected roughly 121,000 claims in the first two months of 2022 alone. Cigna disputed the characterization in a public statement, saying its technology "verifies that the codes on some of the most common, low-cost procedures are submitted correctly" and that the process "does not result in any denials of care." Subsequent litigation, filed in federal court in California and reported by CBS News in July 2023, alleges the review Cigna describes did not meaningfully occur at the pace the internal data shows. The case remains in litigation; the claims-denial pace itself is drawn from Cigna's own internal data as reported by ProPublica, not from an independent audit conducted after the fact.
Average time per claim across more than 300,000 denials over two months, per internal data
Directly confirmed via ProPublica's March 25, 2023 investigation; Cigna disputes that PXDX makes clinical determinations without genuine physician review, stating the process 'does not result in any denials of care,' and the underlying dispute remains in litigation.
The physician's review exists on paper. Whether it exists in practice depends entirely on whether anyone ever checked how long that review actually took.
What changes when review time is measured, not assumed
A sign-off requirement can look identical on paper whether it takes a genuine review or a fraction of a second.
Measuring and reporting actual review duration per flagged claim, with an explicit minimum, closes the gap between a review that exists in policy and one that exists in practice.
Why the Visible Metric Misleads
A claims-processing system tracks throughput, denials issued, claims closed, average handling time, because those are the numbers that describe whether the operation is running efficiently. None of them describe whether the "review" a denial is supposed to represent actually took place. A system can report excellent throughput and full compliance with its own sign-off requirement while that requirement has, in practice, become indistinguishable from no requirement at all, because throughput and true review depth look identical from the outside once the review step no longer takes measurable time. The more revealing measure sits one layer beneath the sign-off itself: how long, on average, a reviewing physician actually spends on a flagged claim before a denial is issued, tracked and reported as its own number, not inferred from whether a signature field was completed.
The Leadership Move
The right move is not to eliminate automated flagging of claims that do not match expected treatment patterns, which is a legitimate tool for catching genuine mismatches at scale. It is to track, as a standing operational metric, the actual time a reviewing physician spends per flagged claim before a denial, and to set an explicit minimum below which a denial cannot be finalized without a documented reason.
- Ownership
Claims operations typically owns the throughput target the flagging system is built to serve. Medical or clinical leadership owns whether the sign-off requirement attached to a flagged denial reflects a genuine review. When no one owns the specific question of how long a review actually takes, throughput becomes the only thing anyone is accountable for, and a sign-off requirement erodes into a formality with nobody deciding that it should.
- Tradeoff
Requiring a documented minimum review time per flagged claim slows processing and costs real operational capacity at scale, across the volume the ProPublica and Capitol Forum reporting describes. The alternative, a review requirement that exists in policy but not in practice, carries a different and larger cost, both in the specific claims wrongly denied and in the exposure a company faces once the actual pace of its own "reviews" becomes public.
- Human consequence
A patient whose claim is denied experiences that denial as a considered medical judgment about their care, not as one of several hundred thousand decisions issued over two months at a pace of roughly a second each. The denial letter looks the same either way; only the internal data shows the difference.
Implication for Operators
Any organization with a review or sign-off requirement attached to an automated flagging system should assume that requirement is only as real as the time actually spent fulfilling it, and that time is not visible from throughput or compliance metrics alone. The practical shift is measuring and reporting actual review duration as its own number, on a standing basis, rather than discovering years later, as in this case, that a review requirement had quietly become a formality nobody was tracking.
PXDX's design intent, flag claims that do not match an expected pattern for physician review, is not, on its own, the problem this case describes. The problem is that nothing in the system's design ever measured whether the review actually took place at a pace consistent with a real review, so a requirement that looked identical on paper for years could erode into a formality without anyone being positioned to notice.
The real story here is not an insurer denying claims. It is a review requirement with no measurement attached to whether the review was actually happening, until reporters, not the organization's own systems, supplied that measurement.
Can your organization show whether AI improved the customer outcome, not only the internal process or employee task?
FAQ
Does Cigna agree that PXDX makes clinical determinations without physician review?
No. Cigna has disputed that characterization, describing PXDX as an administrative tool that flags claims not matching preset criteria for a physician's review, not a system that denies claims autonomously. The dispute over whether that review meaningfully occurred at the reported pace is the subject of ongoing litigation.
Where does the 1.2-second figure come from?
From Cigna's own internal data, as reported by ProPublica's March 25, 2023 investigation, describing an average pace across more than 300,000 claims denied over a two-month period in 2022. It has not been independently re-audited outside that reporting and the subsequent litigation.
Was there a documented earlier warning about this pattern before the investigation?
No independently documented earlier warning sign was found predating the March 2023 reporting. This is not a case where a red flag was raised and ignored; it appears to be a case where no standing mechanism existed that would have surfaced the actual review pace before an external investigation did.
How would tracking actual review time have changed anything?
If Cigna's own systems had reported average time-per-claim spent on flagged denials as a routine operating metric, the pattern the investigation later found would have been visible internally on an ongoing basis, rather than surfacing only once outside reporters obtained and analyzed the underlying data.
Does this mean automated claims-flagging systems are inherently unsound?
No. The distinction is between a flagging system whose attached review requirement is measured and enforced, and one where the requirement exists only as an unmeasured formality that erodes without anyone deciding it should.
Related intelligence
Article
The Algorithm and the Randomized Trial
Two education systems built algorithms to help make decisions about individual students at scale. One imposed its algorithm nationally, on results day, with no pilot and no override. The other tested its tool for years, on a smaller population, before trusting it broadly, and built it to route stude
Article
The Report That Audited Robots and Hallucinated Its Own Citations
A firm whose entire service is verification can still fail to verify its own output, if nobody explicitly owns checking AI-generated content before a client sees it. The irony sharpens considerably when the report in question is reviewing a government's own automated penalty system.
Article
The Playbook Before the Rollout
Two governments built automated systems that made consequential decisions about their citizens. One never wrote down, before deployment, what pre-launch testing an AI system had to pass. The other built a public assurance playbook years earlier, and has kept updating that playbook as the technology