Every digital transformation deck contains the same promise: we will automate the finding of value. Revenue recovery, cost anomalies, missed opportunities, underpriced risk. Point the algorithms at the data, surface what humans missed, and book the upside. It is the most fundable sentence in enterprise technology, and one of the largest industries in America just demonstrated, at a cost of hundreds of millions of dollars, the clause that deck always omits: an automation trap that only finds value in your favor is not an analytics program. It is a liability with a dashboard.
The case study deserves a place in every board packet because the trap it exposes is architectural, industry-agnostic, and almost always invisible until a regulator makes it visible.
Table of Contents
The setup: automation with a preferred direction
The industry is US healthcare, specifically Medicare Advantage, where private insurers covering more than thirty million older adults are paid according to how ill their members’ documented diagnoses show them to be. More recorded illness, higher monthly payments. To ensure no illness went unrecorded, insurers deployed exactly the automation the decks promise: teams and AI pipelines re-reading years of clinical charts, surfacing diagnoses that were supported by the records but never submitted for payment.
As technology, the systems were genuinely impressive. As governance, they had one property that would eventually anchor a Department of Justice case: they corrected in only one direction. Missed diagnoses, which increased revenue, were found and added by the thousand. Unsupported diagnoses, which would have reduced revenue to remove, were structurally invisible; the pipelines were simply never pointed at them.
In March 2026, a major insurer settled federal claims for 117.7 million dollars over precisely this pattern. The same season, government auditors reported that at three plans, 81 to 91 percent of certain sampled high-risk codes lacked adequate documentation, errors that years of “accuracy programs” had somehow never surfaced, because every one of them pointed the profitable way. Regulators now audit the industry with roughly two thousand certified coders on a quarterly cycle, extrapolating sample error rates across entire contracts.
Why does the trap catch good companies?
The uncomfortable truth for executives is that no one needs to intend the asymmetry. It assembles itself from ordinary incentives. The business case for the automation was written in recovered revenue, so recovered revenue is what got built, measured, and celebrated. Finding errors in the other direction had no sponsor, no KPI, and no dashboard. Every individual output of the system could be defended; the indefensible thing was the distribution of outputs, visible only when someone asked the question prosecutors eventually asked: across all your corrections, which way do they point?
That question is now a standard tool of enforcement, and not only in healthcare. Financial regulator asks for it from pricing models. Tax authorities ask for expense classification engines. The asymmetry test is devastating because it requires no proof of intent; the ratio is the confession.
The governance pattern that survives
The healthcare industry’s forced rebuild offers the corrective pattern, and its centerpiece is a discipline the field calls two-way retrospective risk adjustment: the same automated pass that surfaces missed revenue now also flags recorded items the evidence cannot support, with both streams routed through human validation and full audit trails. Removals are treated as first-class outputs, budgeted, staffed, and reported, not as an embarrassing byproduct.
The pattern generalizes into four governance moves that any executive can mandate this quarter.
Instrument the direction of corrections. For every system that modifies consequential data, report the add-to-remove ratio to the risk committee. A ratio near one hundred to zero is not a performance metric; it is the first exhibit in someone else’s case.
Give the unprofitable direction a sponsor. Symmetry does not survive on policy alone. If finding errors against your interest has no owner, headcount, or KPI, deadline pressure will quietly restore the asymmetry within two quarters.
Attach evidence at the point of output. The systems that survive audits emit conclusions with receipts: source data, rule applied, human who confirmed. Retrofitting that trail after the subpoena is archaeology at billable rates.
Price the trap into the business case. The one-way automation trap looked cheaper for fifteen years. Then a single settlement consumed a decade of its returns, before counting the quarterly audits, the remediation programs, and the vendor replacement cycle now underway. Symmetric systems cost more per quarter and less per decade, and the decade is the horizon that boards are paid to see.
The closing question for your next review
Digital transformation is not slowing down, nor should it. But the healthcare precedent hands every leadership team a question worth institutionalizing: for each automation we run, if a hostile examiner audited the full history of its corrections, what story would the direction tell?
Companies that can answer “both ways, with receipts” have built analytics. Companies that cannot have built exposure, and are currently enjoying the interval, always temporary, before someone with subpoena power runs the query for them. The interval in healthcare lasted about fifteen years. The invoice, when it arrived, had nine figures on it. Boards that read case studies get to pay less.
