Some of the most important redesigns are invisible. The object looks the same from the outside, but its purpose has been quietly inverted. A tool built to push in one direction gets rebuilt to protect instead. Healthcare is in the middle of one of these reversals right now, and it is worth watching.
The tool is the way private health plans review patient records after a visit. For years it had one job. Now it is being redesigned to do almost the opposite.
What the tool was built to do
Private plans that cover older Americans get paid more for sicker patients. So the records of how sick those patients are drive real money. To capture that, plans built a process to comb back through old charts and find conditions that were missed.
This is retrospective risk adjustment, and for a long time it ran in one direction only. The instruction was simple: find more. Add conditions, lift the score, increase the payment. The entire design pointed at growth.
Why one direction stopped working
A tool that only adds has a blind spot, and in 2026 that blind spot became expensive. Federal auditors reviewed a group of private plans and found that 80 to 91 percent of the diagnoses they sampled were not fully supported by the medical record. The most common error was a condition pulled from old notes that no longer matched the patient’s current health.
A federal advisory panel told Congress the broader pattern adds up to about 22 billion dollars in excess payments a year. One major insurer paid 117.7 million dollars to settle claims tied to a system that only ever added conditions and never removed the ones that did not hold up. The one-directional design, once the point of the tool, had become its biggest liability.
The redesign: from offense to defense
Here is the inversion. The same retrospective review is being rebuilt as a two-way process. It still finds conditions that are genuinely documented but were missed. It now also removes conditions that are no longer supported. Add and remove, not just add.
That sounds like a small change. It is actually a different machine. A growth engine asks one question: what can we add? A safety system asks two: what is missing, and what does not belong? The second question is the one that survives an audit, and it is the one the old design could not answer.
The Technology that makes the reversal possible
Removing a code is harder than adding one. To delete a diagnosis with confidence, you have to prove the evidence is not there, across hundreds of pages of records. Doing that by hand at scale is close to impossible.
This is where a specific kind of AI earns its place. The approach is called Neuro-Symbolic AI. It pairs pattern recognition with explicit clinical rules, so for every diagnosis it can point to the exact line in the chart that supports it, or flag the absence of support so a human can remove it. The clinician or coder still makes the final call. The machine makes that call faster and shows its reasoning, which is exactly what an auditor wants to see.
A better shape for the tool
Step back and the redesign is almost philosophical. An instrument built to maximize a number is being rebuilt to protect the truth of a record. Its job is no longer to grow the score. Its job is to make sure the score is real.
Retrospective review is not going away. It is becoming a safety layer instead of a growth lever, the thing that catches errors in both directions and keeps the record honest. The forward-looking version of this work is not a flashier algorithm. It is the same familiar tool, redesigned to do the harder and more durable job. That is usually what real progress looks like.
