RADV & Risk-Adjustment FCA Defense
Every submitted diagnosis code in scope, read against the clinical record.
We work for firms defending clients in risk-adjustment audits, and every code that carries risk-adjustment weight is reviewed by a person, with the findings coming back signed. AI reads every chart and sends each of those codes to a person with its read and the note evidence it found, including every code it reads as supported; a person, not the software, decides them and signs. Your firm retains us under its engagement, in a consulting-expert lane.
- A person, not the software, reviews every code and signs the findings
- Every supported or not-supported read cites the line in the note it rests on; a code sent for review says why
- Detected uncertainty resolves to review, never to “supported”; the software can still misread a code as supported, so every code it reads as supported goes to a person
Synthetic chart · Illustrative
Submitted code
C50.911
Malignant neoplasm of unspecified site of right female breast
Verbatim note evidence
History of malignant neoplasm of the right breast, status post mastectomy in 2019, no evidence of recurrence.
The unit of work
One submitted code, one documented answer.
Every supported or not-supported read points at the line in the note it rests on — negation, “resolved,” history-of language — and a code sent for review says why it was sent, so review time goes to deciding, not searching.
Someone Has to Read Every Chart, One Code at a Time
When a plan or provider group defends a RADV audit or a risk-adjustment False Claims matter, someone has to review thousands of charts and answer, one submitted diagnosis at a time: is this diagnosis supported by what the clinician actually documented? That is the same slow, manual, page-turning work regardless of firm size or matter type, and however the surrounding regulatory landscape shifts, that chart-level question never goes away. We do not make the review smaller. Our software reads every submitted diagnosis that carries risk-adjustment weight against its note, and every one of those codes reaches a person with whatever note evidence the software found already pulled. The reviewer's hours go to deciding, not searching.
- Every submitted diagnosis that carries risk-adjustment weight read against its note, with the note language pulled and cited where a read rests on it; codes with no payment category are listed as out of scope
- Negation, resolution, and "history-of" language surfaced for the reviewer
- The signature on every finding is a person's, never the software's
- Uncertainty always resolves to review, never to "supported" — and because the software can misread a code as supported without detecting any doubt, every code it reads as supported still goes to a person
The absence of a finding is not a finding. This instrument does not clear codes. Codes it did not surface have not been adjudicated and are not represented as supported.
Request a CallMatters We Support
Sold through your firm, under your engagement, in a consulting-expert lane — supporting privilege.
RADV Audit Response
CMS's Risk Adjustment Data Validation audit of a Medicare Advantage contract's diagnosis submissions. We build the review queue against the audit's own sample, so the review starts from the software's read of each note, not from raw charts.
Learn MoreRisk-Adjustment FCA & DOJ CID Matters
False Claims Act and qui tam matters over risk-adjustment submissions, and DOJ civil investigative demands. Every adverse finding ties to the exact line in the note it rests on.
Learn MoreOIG & UPIC/RAC Contractor Audits
Where an OIG, UPIC or RAC audit turns on whether the record supports a submitted diagnosis code, the question is the same one, one code at a time. Where a matter needs a statistician's sampling opinion, we say so and stay in our lane.
Learn MoreHow a Matter Runs
Six stages across three hands. What you send us, what we do with it, and what comes back signed.
-
Your firmYou hold the matter and the privilege.
-
Intake
You send the chart set and the claims manifest. We are retained as your subcontractor under your engagement, in a consulting-expert lane. We do not testify, and the engagement letter scopes testimony out.
Crossing to us: charts, and the codes actually submitted
-
-
Us, under your engagementConsulting expert. Non-testifying. Your privilege, not ours.
-
Chart setup
Charts arrive as PDFs in whatever layout your EHR or vendor produces. Layouts vary, so we build a reading template for your format before any code review starts.
-
Our software reads every submitted diagnosis in scope against its note
Our software reads every submitted code that carries risk-adjustment weight against the encounter it was billed from and, where the note has one, pulls the exact line that supports it, or undercuts it, verbatim. It paraphrases nothing, and it will not use a quote that fails a word-for-word recheck. Codes with no payment category are listed as out of scope and are not reviewed. Every code in scope then goes to a person with the software's read and the note evidence it found, including every code the software reads as supported.
-
Review and signature
A person, not the software, reviews every code against the note and signs the findings. Your engagement letter names who that is. The software decides nothing and signs nothing. Where the reviewer disagrees with the software's read, in either direction, the disagreement is recorded.
-
Gated release
By default, nothing leaves until the reviewer's sign-off is recorded, every submitted code is accounted for, and the rendered package passes its checks. A failed check stops the release unless a named person waives it and the reason is recorded on the run; a waiver is never silent.
Crossing back to you: a signed work product, with no dollar column
-
-
Your firmThe deliverable is a signed work product you own.
-
Delivery and invoice
The signed findings come back to you under your engagement. Every supported or not-supported read cites the line in the note it rests on; a code sent for review says why it was sent. On a first matter your own Kovel-retained coder countersigns. The fee was set at intake and does not move with what the review found.
-
Two AI Models Read Every Code in Scope, Separately. Neither One Decides.
Two different AI models each read the note for every code in scope, separately, and answer the same questions, quoting the note for every yes. Neither model decides. Fixed rules, written down and applied the same way on every chart, read a code as supported only when both readings agree and every check passes; any disagreement or doubt sends the code to review. One more automated check then looks again at each code read as supported or not supported. It can send a code to review, never the other way, and nothing it does can make a finding look more favorable than the record supports.
- Every submitted diagnosis that carries risk-adjustment weight gets a documentation-support read — each reading judges from the whole sentence, not from single words, whether the condition is present at this visit or is ruled out, resolved, historical, a family history or someone else's
- The questions, code lists and result wording live in one versioned configuration file, approved by named people before it is used; the fixed checks are the engine's own code. Both are applied the same way to every chart — so the read stays consistent and reviewable
- The final automated check runs on the same model as the second reading, so it is an internal check, not an independent one — and whatever the read, every code in scope then goes to a person
- Evidence is pulled verbatim, never paraphrased or inferred — a quote the system can't produce is a finding it doesn't make
See how the two-model check works →
See the Full MethodWhat You Receive
A signed work product — not a data dump, and never a dollar total.
- Per-chart findings: plain-language, per-code findings with verbatim evidence quotes and a signature block
- Batch overview: verdict distribution and an exposure register — relative risk-adjustment weight, never a dollar figure
- Analyst workbook: a case-management starter your team owns — assign, track status, add notes
- Review queue: every code that carries risk-adjustment weight, with the software's read and the note evidence it found already pulled
- Agreement record: where the reviewer agreed with the software's read or did not — reported directionally, never averaged into one number
Our software
-
Reads every submitted code that carries risk-adjustment weight against the note it was billed from.
-
Pulls the verbatim line the code rests on, where the note has one.
History of malignant neoplasm of the right breast, status post mastectomy in 2019, no evidence of recurrence.
-
Gives the code a read: supported, not supported, or needs review. It decides nothing and signs nothing.
In scope
A person
The code reaches the reviewer with the software's read and the note evidence it found, whatever the read, including supported.
The reviewer works it against the note and decides whether a finding issues.
The reviewer signs the finding.
Out of scope
Listed, not reviewed
The code carries no payment category, so it is listed as out of scope and no finding is made on it.
The absence of a finding is not a finding. This instrument does not clear codes. Codes it did not surface have not been adjudicated and are not represented as supported.
Exchange & Compliance
How We Work With Your Data
Exchange-agnostic, per matter: your governed platform where you have one, or a per-matter S3 folder where you don't.
- Every demo runs on synthetic charts — no patient data, no BAA, no security review required
- No real record is processed until business associate agreements covering every processor in the chain are in place. Part of that is enforced in the software: once real records are declared, it refuses to send them to an AI vendor that has no agreement on file
- Retained by your firm, under your engagement, as a consulting expert — not a testifying one
- Manual and matter-driven. A matter is opened, run, signed, released and closed, and there is no always-on platform to secure
- Your matter's records are processed on our own equipment for that matter, except that the note text is sent to our AI vendor to be read. They are kept apart from the made-up charts we build and test on. We never use them to train or tune any model; what the AI vendor may do with what it receives is governed by its terms and, for real records, the business associate agreement. They are destroyed on your instruction and schedule
- SOC 2 Type II is not in hand. Your vendor-risk review will take as long as it takes, and we plan that lead time into the pilot
Frequently Asked Questions
What's your error rate on real charts?
We do not quote an error rate. None has been measured on real charts, and we will not publish a number we have not measured.
How is this different from the consultants we already use?
We're not replacing your coder's or your consultant's judgment — we're doing the page-turning underneath it. The reviewer works from a pre-built queue instead of raw PDFs, so the hours go to deciding, not searching.
What about PHI, BAAs, and vendor risk?
No real record is processed until business associate agreements covering every processor in the chain are in place. Part of that is enforced in the software: once real records are declared, it refuses to send them to an AI vendor that has no agreement on file. Until then, every demo and every pilot runs on synthetic charts only: no patient data, no BAA, no security review required.
Does using software here affect privilege or discoverability?
We are retained by your firm, under your engagement, in a consulting-expert lane. We do not testify. Our selection criteria and evidence assembly are discoverable methodology, and we write them down. The method is developed and tested against a permanent, versioned synthetic corpus, so it can be examined on its own terms without your matter's records being involved. The Methodology page sets out how the two are kept apart.
Who signs the findings, and what's their credential?
A person, not the software, reviews every code and signs the findings. Your engagement letter names who that is. Our coding lead approves the clinical criteria the software applies before any version of them is used on a matter. She holds no coding certification today, and we do not describe her signature as a coder's signature or as a certified one. She is working toward the risk-adjustment coding credential (CRC), and this answer changes only when the certificate is in hand. She is also a co-founder with equity in this business, so any grade she gives our own engine is an internal number and we will not offer it as proof. We have no independent measurement of the engine's error to offer today. On a first engagement your firm's own Kovel-retained coder countersigns the work product.
What AI is actually doing the work?
Two different AI models each read the note for every code in scope, separately, and answer the same questions, quoting the note for every yes. Neither model decides. Fixed rules, written down and applied the same way on every chart, read a code as supported only when both readings agree and every check passes; any disagreement or doubt sends the code to review. One more automated check then looks again at each code read as supported or not supported. It can send a code to review, never the other way. It runs on the same model as the second reading, so it is an internal check, not an independent one. Whatever the read, every code in scope then goes to a person. See how the two-model check works.
What will you not do?
We stay in a defined lane and don't extrapolate beyond it. See what we will not sign for the specifics.
How does delivery work?
We're exchange-agnostic per matter: your governed platform where you have one, or a per-matter S3 folder where you don't.
How are engagements priced?
A flat fee per matter for RADV-boutique and BigLaw-scale work, or a per-chart tier for volume matters — priced to the matter, never a percentage of recovery, so our read stays unbiased. See Engagement & Fees for the structure.
Request a call
Thirty minutes. No patient data, no BAA, no security review.
A few optional questions on the next page tell us what to bring. Nothing on that form is privileged; keep it general.
Request a Call