Your systems may all be working. Your health plan can still be losing money.
Health plans have spent years buying strong technology for individual functions.
Risk adjustment identifies opportunities. Coding platforms validate diagnoses. Encounter systems submit them. Quality systems measure performance.
Individually, each system may be doing exactly what it was designed to do.
And the plan can still lose money.
Because some of the biggest failures in payer operations do not happen inside these systems.
They happen between them.
A condition identified by analytics may never make it into the medical record. A diagnosis validated by a coder may sit behind a rejected encounter. A correction may be resubmitted but never reconciled. A condition captured last year may disappear this year without anyone understanding why.
Every vendor can report good performance while the health plan loses value on the same member.
That is the problem I believe our industry now needs to solve.
We connected healthcare. We did not make it agree.
I have spent more than two decades working with technology that moves information between payers, providers and healthcare systems.
We have made tremendous progress in connectivity.
But connectivity is not the same as intelligence.
For one member and one condition, a health plan should be able to answer a very simple set of questions:
What did we suspect? What evidence supported it? Was it validated? Was it submitted? Was it accepted? And if not, why not?
Today, those answers often live across multiple applications, vendors and teams.
That is not just a technology problem.
It is a revenue integrity, operating-cost and audit problem.
Ultimately, it becomes an EBITDA problem.
One member. One condition. One longitudinal truth.
This is the principle I have challenged our team at Invent Health to build around.
Health plans already understand the idea of maintaining an authoritative member record. I believe the same discipline needs to be applied to the condition.
The journey should be visible end to end:
Signal → Evidence → Validation → Submission → Acceptance → Reconciliation
Once those events are connected, the exceptions become obvious.
A condition we expected but never documented becomes a provider-engagement opportunity.
A supported diagnosis that was never submitted becomes an operational issue.
A submitted diagnosis that never reached the expected downstream outcome becomes a reconciliation issue.
A previously supported condition that disappears becomes something worth investigating early, not at year-end.
This is where the value is.
The next opportunity is not just finding more diagnoses.
For years, the industry has focused heavily on suspecting.
Who can find more conditions?
That will continue to matter.
But I believe the more important question is becoming:
Of everything we already identified, documented and coded, how much actually made it all the way through?
Sometimes deteriorating risk performance is not a suspecting problem at all.
It is a reconciliation problem.
Buying another algorithm does not solve that.
Hiring more coders does not solve that.
Adding another dashboard does not solve that.
Sometimes the highest-return opportunity is simply making sure the work you already paid for does not disappear between systems.
CMS is making this more important.
The regulatory direction is also becoming clearer.
For CY2027, CMS finalized the exclusion of diagnoses from unlinked chart review records from Medicare Advantage risk-score calculations, with a limited exception for members switching MA organizations. CMS also finalized the exclusion of diagnoses from audio-only encounters. At the same time, RADV audits are expanding across eligible MA contracts.
The implication is straightforward:
Evidence, coding, encounter submission and acceptance can no longer be treated as separate events.
The lineage between them matters.
Where AI belongs
AI can help enormously here.
It can retrieve evidence, connect historical information, prioritize exceptions and reduce the manual work required to reconcile millions of clinical and operational events.
But in healthcare, intelligence without traceability is not enough.
Every recommendation should point back to evidence.
Material decisions should retain human oversight.
And when an auditor asks, “How did this diagnosis get here?” the answer cannot simply be, “the AI found it.”
There has to be a traceable path.
The record should belong to the health plan.
I also believe the longitudinal history of the member and condition should belong to the plan, not become trapped inside a vendor application.
Health plans change vendors. Members move between products. Regulations change.
The intelligence accumulated around the member should survive all of those things.
I would rather earn the next contract than make it difficult for a client to leave the current one.
The question I would ask every health plan
Pick one member.
Pick one meaningful condition.
Then ask your team to show you, in one place:
What identified it → what supported it → who validated it → how it was submitted → whether it was accepted → what happened next.
If answering that requires four systems, several spreadsheets and a meeting between vendors, the problem is probably not that you need more technology.
You need the technology you already have to operate as one system.
That is what we mean at Invent Health when we talk about a Payer Intelligence Platform.
One member. One condition. One longitudinal truth.
From evidence to acceptance.
See how it works
We will walk your team through the platform end to end, follow a single condition from evidence to acceptance, and show you where the money leaks in a book like yours.
Frequently Asked Questions
What is payer intelligence?
Payer intelligence is the practice of running risk adjustment, coding, encounter submission and quality against one shared record of the member and the diagnosis. Instead of four systems reporting four results, the plan holds one position per condition and works the differences between what was expected, coded, submitted and accepted.
Why do risk, coding and encounter systems disagree about the same member?
Most plans bought the strongest product for each function, one at a time, over many years. Each vendor is measured on its own number, whether that is suspects delivered, charts coded or files accepted, and none of them is responsible for what happens between systems. Disagreement collects in that space.
Do unlinked chart review diagnoses count toward CY2027 risk scores?
No. In the CY2027 Rate Announcement finalized on April 6, 2026, CMS excluded diagnoses from unlinked chart review records from Medicare Advantage risk score calculation, with a narrow exception for members switching from one MA organization to another. CMS estimated the payment impact at about 1.53 percent.
What happens to diagnoses from audio only visits in 2027?
CMS also excluded diagnoses captured on audio only services, identified by modifier 93 or FQ, from the CY2027 risk score calculation. Plans that lean on telephonic assessments for condition capture should size that exposure now and move those conditions onto visits that will still count.
Why do coded diagnoses never show up in accepted risk scores?
Usually the diagnosis was coded correctly and then lost between systems. Rejected or orphaned encounters, acknowledgments that were never reconciled at the diagnosis level, corrections that were never resubmitted, and prior year deletes all end the same way, with coded work that never becomes accepted risk.
Should a health plan own its own diagnosis history?
Yes. That history is the plan’s asset. When it sits inside a vendor’s application in a form the plan cannot export, changing vendors means losing years of context and audit history. Ownership protects the operation, keeps condition history intact when members change products, and strengthens the plan at renewal.
Can agentic AI be used safely in risk adjustment?
Yes, when every recommendation is traceable. That means citing the source page, keeping a coder in the decision on anything with payment or audit consequence, generating no diagnoses automatically, naming the person or agent behind each entry, and running agents in shadow mode before they touch live work.
Can one platform support Medicare Advantage, Medicaid and ACA?
Yes, and members who move between programs are the main reason it matters, because condition history usually fragments when someone changes products. A shared record preserves that history, supports CMS-HCC, HHS-HCC and CDPS models, and handles both Medicare and ACA submission reconciliation in the same workflow.
