Independent assurance for insurers and lenders
Find the premium you’re losing. Prove every price matches what you filed.
Tensyl checks every quote and decision against what you filed and approved, and gives your reviewers a signed record an examiner can re-run without us.
What Tensyl isAn independent check for the people who sign off, delivered like an auditor’s report.
- Not a testing tool
- Not monitoring
- Not an experimentation tool
Findings report
Small business · checked against the filed WA rate manual
- +$343
A surcharge the filing removed is still applied
RuleFiled $2,513 · charged $2,856
- +$528
Owner not found in vendor data, default filled in
Hand-offFiled $2,513 · charged $3,041
- −$647
Classifier missed the fryer, wrong business class
ModelFiled $2,513 · charged $1,866
1,630 checks built from the filing
No correct price flagged
Sign-off
- Pricing actuarysigned
- Model risksigned
- Compliancepending
Already in production: a public P&C rating platform pays us on outcomes.
The problem
The model is rarely where the decision goes wrong.
Every price comes out of a chain: the quote site, the data hand-off, your rate tables, a risk model, an underwriter. Each team checks the piece it owns. Nobody checks the price that comes out the end, and the premium leaks where the pieces hand off to each other.
- 237
- decision steps we mapped across 9 regulated use cases. The outcome was set by a rule, a table or a hand-off.
- 24.9%
- over-charge on one car insurance quote when a single field is lost on the way to the insurer.
- 21 months
- that Sompo kept charging its old rates after an approved 28.8% cut. A check against the filing catches that on day one.
One quote, six steps: every step passed, the price was still wrong
Regulators act on the rules, not on model accuracy. Discover’s rule was wrong for about 17 years, and Bank of America paid $225M for denying claims on a fraud filter alone. Yet the Federal Reserve’s model-risk guidance (SR 26-2) leaves fixed rules and generative AI out, and no one owns the final outcome.
How it works
We check the decision you sign, not just the parts behind it.
Tensyl reads your filed rates and approved rules, writes down the right answer for every case, and re-runs your quotes and decisions against it. Each finding names the rule or hand-off that caused it and the dollars involved, totaled across your book. Your owners set the thresholds before anyone sees a result.
- 01
Filing check
Every rate-table combination priced against the filed manual, so a missed rate change shows up on day one, not month 21.
- 02
Data feed check
What the shopper entered, what the quote site sent, and what your system filled in by default, with the dollars lost per field.
- 03
AI and vendor change approval
A sign-off pack for every generative AI step and vendor update, re-checked before the change reaches a decision.
- 04
Signed decision record
Extends the record you already keep to cover what other companies sent you, so every hand-off, and every error, has an owner.
One record, four sign-offs
- Chief underwriting officer
- Money lost by field and data source, and which fixes to approve.
- Pricing actuary
- Proof the system charges what was filed.
- Model risk
- Rules and generative AI brought under review, without building every check by hand.
- Compliance
- Exam-ready evidence before anyone asks for it.
Proof
Paid on results today. Checked against real filings.
In production
A public P&C rating platform pays us on outcomes.
In our prototype, on mock systems
- 7 of 7
- planted pricing errors found on MGT Insurance's filed small-business manual. A typical hand-built check found 2.
- 9 of 9
- planted errors found on Reg E card disputes. A typical hand-built check found 1.
- 0
- correct prices or decisions flagged as wrong, so your team doesn't chase noise.
- $313,534
- in overcharges surfaced from one 5,000-claim export, with no system access.
Errors were planted in mock systems modeled on real enforcement cases, not found in any company’s live system.
Why now
Models are getting cheap to build. Getting them approved is not.
Before a model touches an underwriting, claims or lending decision, the people who can say no have to agree it is right, fair and controlled. Today that proof is written by hand, per use case, per model, per update.
Generative AI arrives without the usual checks
No hand-checked cases, no fixed model version, and vendors update on their own schedule.
Vendors are bundling AI into decisions
The NAIC says 42% of homeowners and auto AI models come from third parties, and no vendor can certify itself.
Checking a whole filing is now affordable
AI can read a filed rate manual or a regulation and turn every rule into a check. Until now that was done by hand, filing by filing.
The pressure has moved to states
Colorado's AI law, California's CARS Act and the NAIC's AI review supplement. Reinsurers and litigators ask for the same evidence.
Trust
We never sign our own work.
No vendor cooperation needed
We check your vendors' outputs as they are. What went in and what came out is enough.
Your officer signs, not us
Your team sets the thresholds before any result is seen, and can plant known errors to check our work.
Your data stays under your control
An audit uses one de-identified export. A pilot runs read-only inside your cloud. We never train on your data.
Open format, no lock-in
Leave with every check and every record. Your validator can re-run them without us.
Your evidence is your IP
You keep every check, finding and signed record. We keep the engine and a library of failure types that grows with every customer.
How this differs from the tools you have: engineering tools tell you the software changed or a model improved. Tensyl tells you a price or decision is wrong, which rule did it, and what it cost.
Free audit
Send one export. Get a findings report.
Start where there’s no security review and no system access: one de-identified file of past decisions. A spreadsheet is fine. We check it against your own rulebook and send back what we found, with dollars attached.
Email only here. We’ll reply with a secure way to send the file, so please don’t include data in this form.
- 01
What we need
One de-identified export of past decisions, such as past quotes for one state and line with the inputs, rate version and premium charged. No access to your systems.
- 02
What you get
A findings report: every decision that doesn't match your filing or policy, the rule that caused it, and the dollars involved.
- 03
If it's useful
A 60 to 90 day pilot on one use case. Your team plants known errors, and your validator re-runs the evidence without us.