MedIQ
Feature · Governed AI

Managed by MedIQ. Audited end-to-end.

We don't ask tenants to pick models or hold AI keys. MedIQ centrally vets each model for HIPAA eligibility and accuracy, then assigns the right one to each task. You get the capability — we own the cost, the compliance, and the upgrade path.

Talk to engineeringTrust Center →
Why managed

Three reasons MedIQ governs every model centrally.

Centralized model governance isn't a limitation — in healthcare RCM, it's the only defensible posture. Here's how we operate it, and why it matters.

Compliance is not a checkbox
Every model we use is reviewed for HIPAA eligibility, data handling, and BAA coverage. We can't extend that review to a model you wired in last Tuesday.
Accuracy drifts; we'll catch it
Models change. We monitor scrubber and coding accuracy across the fleet and roll forward — or back — without any action from your team.
Your auditor needs one story
When OCR or a payer asks how AI touched the chart, the answer is the same across every tenant: managed by MedIQ, logged with input hash, model id, and decision.
What you get

Governed AI, in seven properties.

Task-by-task assignment

Each AI task — scrubber risk pass, appeal letter, coding suggestion — runs on the model MedIQ has approved for that task.

HIPAA-eligible only

We only enable models whose provider terms support our BAA posture. Non-eligible models cannot be assigned.

Confidence threshold

Per-tenant slider — Conservative, Balanced, Aggressive. Auto-apply only fires above your threshold.

Reversible patches

Every AI-applied change is a reversible patch with the original preserved. One click to undo, log preserved.

Field-level audit

Every AI decision writes input hash, decision, rationale, and confidence — never PHI.

De-identified prompts

PHI stripped before any model call. Logs record the structured payload, not patient identifiers.

No keys to manage

No model keys to provision, rotate, or invoice. No metered surprise. No per-inference billing.

Roll-forward, roll-back

We monitor accuracy and latency across the model fleet and adjust assignments. You don't need to do anything.

Auto-apply

From suggestion to safe, reversible patch.

The same pipeline runs on every claim. Suggestions that don't meet your tenant's confidence threshold flow to a biller for review; everything else applies cleanly and writes itself into the audit log.

  1. 1
    Deterministic findings
    Rules execute first; AI never overrides a rule.
  2. 2
    AI suggests patches
    Governed model proposes a reversible change with a confidence score and rationale.
  3. 3
    Threshold check
    If confidence ≥ tenant threshold, the patch is queued for auto-apply.
  4. 4
    Apply + audit
    Patch applied; original value preserved; audit row written with input hash + decision.
  5. 5
    Biller review (optional)
    Anything under threshold lands in the work queue with the proposed diff.
  6. 6
    Reversal
    One-click undo restores the original and logs the reversal.
FAQ

The questions auditors actually ask.

No. MedIQ assigns models per task based on platform-wide compliance review and accuracy benchmarks. This is a security posture choice — your auditor gets one consistent answer, and we own roll-forward / roll-back across the fleet.

Want the governance brief?

We can send the model review checklist, audit-log schema, and accuracy-monitoring approach.

Request the briefTrust Center