Key Points
- Leading plans start by comparing each member’s true risk to what was actually submitted, splitting the population into understated-risk and overstated-risk groups from one shared analysis.
- NLP-driven coding reads 100% of charts with documentation support, achieving up to 93% accuracy in mature programs, letting coders focus only on flagged opportunities in 10- to 15-minute passes for a 5-10x productivity gain.
- The most effective programs run analytics, coding, and audit discipline as one accountable engagement rather than stitching together separate tools and vendors.
In our previous blog, we discussed how CMS’ new V28 model and annual audits requirement rewrote the revenue rules for health plans. So, what are health plans doing to address these major challenges?
The health plans pulling ahead in 2026 share one habit: they stopped treating revenue capture and audit defense as opposites. The conventional wisdom says push harder on capture and you raise audit risk; pull back to stay safe and you leave money behind. The best programs have shown that the trade-off is an artifact of how the work is organized, not a law. The same analysis that recovers earned revenue also retires indefensible risk.
Step 1: They Start with True Risk, Not History
A leading program does not open by reviewing medical charts. It opens by comparing each member’s true risk — modeled at the county, member, and diagnosis level — against what the plan actually submitted. That comparison immediately splits the population into two groups: members with understated risk, where supportable revenue is being missed, and members with overstated risk, where submitted Hierarchical Condition Categories (HCCs)may not survive an audit.
That single view is the unlock. Capture and compliance are no longer separate workstreams competing for budget. They are two outputs of one analysis, drawn from the same source of truth.
It also changes the internal conversation. When capture and compliance live in different teams with different tools, they argue. The revenue team wants to submit more; the compliance team wants to submit less; each suspects the other of creating its problems. When both teams read from the same true-risk analysis, the argument dissolves, because they are looking at the same members and can see exactly which ones represent opportunity and which represent exposure. Alignment stops being a meeting and starts being a property of the data.
Step 2: They Read Every Chart, Then Spend Coders Wisely
Sampling cannot find what it does not look at, so leading plans move to full coverage. NLP-driven coding reads 100% of medical charts with MEAT documentation support and, in mature programs, 93% accuracy. But total coverage does not mean burying medical coders in every chart. The engine surfaces opportunities with evidence attached; coders review only the flagged charts with new HCC opportunities, in focused 10 to 15 minute passes.
That targeting is what produces a 5 to 10x coder productivity gain. It is not people working faster under pressure. It is expert judgment applied only where it changes the outcome, on encounters that are already documented and evidence-linked.
Step 3: They Build the Audit Trail as They Go
The health plans that sleep well during an audit do not assemble documentation after submission. They capture it as the work happens. Every submitted HCC carries its supporting evidence and a full trail, so Risk Adjustment Data Validation (RADV) audit readiness is a property of the program, not a fire drill triggered by a memo.
Just as important, they decide what not to submit. A dedicated “delete bucket” gives coders an explicit, tracked way to drop HCCs that documentation will not support. Retiring weak risk on your own terms is far cheaper than having an auditor find it and extrapolate it across the contract.
The discipline here is cultural as much as technical. In many programs, declining to submit a borderline code feels like leaving money behind, and the incentives quietly push toward submitting it anyway. Leading plans flip that instinct. They treat a tracked deletion as a win, because it converts a future liability into a present, controlled decision. Over time, the delete bucket also teaches the analytics engine where documentation tends to fall short, so the next cycle flags those patterns earlier and the program gets sharper on its own.
Step 4: They Run It as One Program, Not a Toolkit
The common failure mode is buying the pieces — an analytics tool here, a coding vendor there, a compliance review somewhere else — and asking an already-stretched team to integrate them. Leading plans instead run the whole model as one accountable program. That is the design behind Sutherland’s Healthcare Risk Adjustment Solution for Payers: analytics calibrated to V28, AI coding across every chart, expert coders focused on new opportunities, and audit discipline built into each step, delivered as a single outcome-based engagement.
None of this requires heroics from an already-stretched team. The point of running the model as one program is that the integration burden, the calibration to V28, and the audit discipline are carried by the engagement, not added to a department’s to-do list. Leading plans get the outcome without having to become systems integrators on top of their day jobs. The result is a program where each cycle sharpens the next, and the gap between true risk and submitted risk narrows from both directions at once.
Continue Reading
Approach is one thing; proof is another. In the final post, we look at what this model actually delivers, in numbers.
