Service · Fixed scope

Quality Audit

Find out what your healthcare AI gets wrong, how often, and why.

Fixed scope · your own data

Your dataencounters, not vendor demos
Your cliniciansdecide what counts as an error
Every scorecarries a margin of error
Tests you keepre-run them after every change
The problem

You can't fix what you can't see.

Most teams judge their AI by a demo, a vendor benchmark or a few complaints. None of those tell you how often a note drops an allergy, or whether the same mistake repeats across thousands of visits.

Errors hide in the long tail of real encounters

Vendor scores are not measured on your patients

Complaints arrive after harm, not before

What has to happen

How it works.

Run the AI on your own encounters, have your own clinicians decide what right looks like, and measure every error with a stated margin. Then work out which part of the pipeline causes each one.

  1. 01

    Agree what matters

    We meet your clinical lead and pick the 4 to 6 parameters that matter most for your product.

  2. 02

    Run on your cases

    We replay a sample of your encounters, or realistic simulated ones, through the product.

  3. 03

    Clinicians review

    Your clinicians mark outputs pass or fail and say why. Their reactions become the written rubric.

  4. 04

    Measure and diagnose

    Calibrated graders score every output. We swap one part at a time to find which one causes each error.

Head to head

Every other way to check leaves something out.

Demos flatter. Spot checks miss the long tail. Generic testing tools don't know clinical meaning.

CapabilityPrecision Quality AuditFixed scope, your casesVendor demo or benchmarkChosen by the vendorInternal spot checksA few notes, read by handGeneric AI testing toolBuilt for any chatbot
Uses your own encountersYesNoA handfulOnly what you supply
Clinicians define an errorAlwaysRarelyInformallyNot built in
Errors ranked by harmSeverity-gradedSingle scoreAd hocNot clinical
Finds the likely causePart-swap experimentsNoGuessworkLog reading
Shows the grader's error rateBeside every scoreNoNot applicableRarely
Leaves you a test suiteYesNoNoSometimes

Columns describe typical positions of each option, not named vendors. Check current terms before deciding.

What you get

What lands on your desk.

Indicative: four to six weeks from kickoff to readout.

Severity-ranked failure report

Every error type, how often it happens and how much harm it could do.

Likely causes

Ranked causes with effect sizes, from experiments on the same input.

A starter test suite

Cases you can re-run after every model or prompt change.

A readout with your team

A working session with clinicians and engineers, not a PDF handed over.

Who it's for
  • AI scribe, voice, chat and coding vendors
  • Digital health teams shipping a new AI feature
  • Health-system AI governance offices
Pricing

Clear scope, clear price.

Fixed price per audit, by number of parameters. Indicative pricing in US dollars. Final scope and price are confirmed on a short call.

Starter

$4,999one-off

The biggest risks in one product, measured properly.

  • Up to 12 parameters
  • Unlimited test cases
  • Severity-ranked failure report
  • Likely causes
  • Readout with your team
Get a quote
Most chosen

Standard

$7,999one-off

The full loop across the parameters that matter to you.

  • Up to 30 parameters
  • Unlimited test cases
  • Causes from part-swap experiments
  • Subgroup and language splits
  • Starter regression suite
  • Readout workshop
Get a quote

Enterprise

Customscoped with you

Several products, specialties or sites. We scope it together.

  • Everything in Standard
  • More parameters and products
  • Your environment or ours
  • Ongoing support
Talk to us

Indicative pricing in US dollars. Final scope and price are confirmed on a short call. See all pricing

Do you need access to patient data?

Not by default. We can work on de-identified encounters, synthetic ones, or inside your own environment. The free Precision De-ID tool masks identifiers on your machine first.

Who decides what counts as an error?

Your clinicians. We propose criteria from their reactions, then show where our graders disagree with them.

What if the product is a vendor's, not ours?

That works too. Health systems use the audit to test a vendor's product on local cases before and after go-live.

Book a Quality Audit

Your AI, tested on your encounters and graded by your clinicians.