The RFP Questions That Miss the Point
Most risk adjustment RFPs evaluate vendors on a standard set of criteria: chart processing volume, NLP accuracy rates, integration capabilities, implementation timeline, and pricing. These questions are necessary. They’re also insufficient. They evaluate whether the system works. They don’t evaluate whether the system produces output that survives federal scrutiny. The RFPs that generate the best vendor selections in 2026 include a second layer of evaluation focused on defensibility. These questions don’t replace the operational ones. They supplement them with the compliance and audit-readiness criteria that determine whether the system’s output holds up when CMS examines it.
The Questions That Should Be in Every RFP
Does the system validate documentation against MEAT criteria before recommending a code, or does it identify codes and leave validation to the coder? This is the difference between evidence-first and find-first architecture. Systems that validate before recommending produce defensible output by design. Systems that find first and validate later produce output that depends on coder judgment under throughput pressure. Does the system identify codes to remove as well as codes to add? If the vendor can only demonstrate addition capability, the system enables add-only programs that OIG’s February 2026 guidance identified as high-risk. Ask for deletion metrics across the vendor’s client base. If they can’t produce them, the capability likely doesn’t exist in practice.
Is the AI explainable? Ask the vendor to demonstrate how the system produces a coding recommendation for a specific chart. You should see the clinical language identified, the MEAT elements mapped, the evidence quality assessed, and the reasoning chain that connects documentation to recommendation. If the vendor can only show the input (chart) and the output (code) without the reasoning in between, the AI is opaque. Can the system produce a defensibility score for individual codes, individual members, and the plan’s overall submitted population?
Defensibility scoring is what converts raw coding output into audit-readiness intelligence. Systems without this capability leave the plan unable to quantify its own RADV exposure. Does the system support concurrent RADV audit management? With quarterly audit cadence, the plan may manage three or more active cycles simultaneously. The system should track each cycle independently, manage separate enrollee samples and deadlines, and produce formatted submission packages for each without manual tracking.
Scoring the Responses
Weight the defensibility questions at least equally with operational questions. A system that processes charts 20% faster but can’t produce evidence trails or support two-way coding is optimizing a function that generates regulatory exposure. A system that’s slightly slower but produces audit-ready output with every coded chart is optimizing the function that actually determines whether the plan’s submissions survive scrutiny.
The RFP Upgrade
Plans issuing RFPs for a risk adjustment solution in 2026 should test whether the vendor’s technology is built for the enforcement environment that exists, not the one that existed when most risk adjustment tools were designed. MEAT validation, two-way coding, explainable AI, defensibility scoring, and concurrent audit management are the capabilities that separate solutions built for compliance from solutions built for revenue. The RFP should test for all of them.