Risk and model risk teams
Show that automated document decisions are tested, tracked and reviewed.
QA for AIConfidence in every automated decision.
QAAI is quality assurance for the AI that reads your documents. We test it against verified correct answers, track every update, and give you evidence your auditors and regulators can read.
Free self-assessment. About 4 minutes. No sales call.
Claim email, water damageSample, synthetic data
Source document
Hello, I'm writing to submit a claim for water damage at my home.
Policy POL-216739.
The loss happened on March 4, 2026.
The repair estimate is $14,928.25.
Insured: Maya Kowalski, Manitoba.
| Field | AI extracted | Verified answer | Status |
|---|---|---|---|
| Policy number | POL-216739 | POL-216739 | Verified |
| Insured | Maya Kowalski | Maya Kowalski | Verified |
| Loss date | 04/03/2026 | 2026-03-04 | Flagged: ambiguous date format, flagged for review |
| Loss type | property_water | property_water | Verified |
| Claimed amount | $14,928.25 | 14928.25 | Verified |
| Province | MB | MB | Verified |
5 of 6 fields verified. Ambiguous date format, flagged for review.
Who it is for
Claims and underwriting operations
Know before each update goes live whether it still reads your documents correctly.
Compliance and audit
Get a report that maps testing evidence to the guidance you are accountable to.
Built for insurers and fintechs in Canada and beyond.
Three steps from your AI's output to proof.
Define what your AI should extract.
List the fields, such as policy number, loss date and amount, and mark which ones are critical.
Add the verified answers.
Upload a test set where every correct value is confirmed. Each version is locked, so results stay comparable.
Run, compare and share.
Upload what your AI produced. See accuracy by field, what changed since the last version, and a pass or fail decision against your own standard.
Evidence your auditors can read.
Every test run produces a report: what was tested, how it scored with a confidence range, how it compares with the last version, and a sign-off block for your independent reviewer.
Take the free assessmentSample report, synthetic data
Documents fully correct
88%
Example figure from synthetic data
95% range 78 to 94%
Policy number98%
Insured94%
Loss date88%
Claimed amount96%
Release gate: review needed
How we test
Method notes available on request.
You give us documents and the verified correct answers.
We score every field, not just the document overall.
We show a confidence range with each result, so small samples are not over-read.
We compare each new version with the last and flag anything that got worse.
Built around what regulators ask for.
QAAI's tests and reports are designed to support the testing and documentation practices described in these frameworks.
- OSFI Guideline E-23Model risk management, effective May 1, 2027
- Quebec AMF guideline on the use of artificial intelligenceEffective May 1, 2027
- NAIC Model Bulletin on the use of AI systems by insurersAdopted by many US states
Supporting evidence for these practices does not by itself demonstrate compliance.
Where does your AI stand today?
Answer 17 questions about your AI system and get a readiness result matched to its risk level.
Built by a QA professional
Sai Krishna Selvaraju
Mississauga, Ontario
I spent my career in software quality assurance, where the question is always the same: how do you know it works? Insurers now rely on AI to read emails and applications, and that deserves the same discipline. QAAI is how I am bringing it. If you run or oversee this kind of AI, I would value a conversation.
Talk to the founderFrequently asked questions
What does QAAI test?
AI that reads documents such as emails and applications and passes fields into your systems. We check each field against verified answers.
Do you need access to our AI system?
Not for the first version. You upload the documents, the verified answers and the AI's outputs. A direct connection is planned later.
Where does our data go?
The first version works from files you upload. We recommend de-identified or synthetic data while we confirm data residency options. We will state our hosting and retention terms in writing before any pilot.
Does a passing result mean we are compliant?
No. QAAI gives you testing evidence. Compliance is a decision for your organisation and your regulators.
Which guidance does the evidence map to?
OSFI E-23 (effective May 1, 2027), the Quebec AMF guideline on AI (effective May 1, 2027) and the NAIC Model Bulletin. We show the mapping so your team can review it.
How much does it cost?
Early access is free while we shape the product with a small group.
Join early access
Be among the first insurers and fintechs to run ground-truth tests with QAAI.
Data handling
Early access runs on files you upload. Use de-identified or synthetic documents. Written hosting and retention terms before any pilot.