Tesora: Frontier AI for Actuaries

Complete complex actuarial work in half the time.

Precise actuarial work, finally trusted to AI

Get actuarial work done fast, reproducibly, and at a fraction of the cost.

Foundation Capital

Y Combinator

What we ship

We built the infrastructure around real actuarial workflows.

Integrations

Works with what you already run.

Built-in checks

AI that can’t hallucinate

Checks run between the AI model and your desk. If a number cannot be traced to its source page, you never see it.

Every agent must pass these checks before its answer counts as finished. In actuarial work, a confidently wrong number ends up in a rate filing, a reserve opinion, or testimony.

Built for enterprise security

Isolated in an environment of its own, so no one else sees your data

We follow state-of-the-art security protocols and are SOC 2 Type II audited. Your data stays on a virtual machine of its own, and never joins a shared index or passes through anyone else’s context window.

FAQ

Common questions.

From chief actuaries, pricing teams, and the IT and risk officers who review us.

Bring us your hardest problem

A tangled spreadsheet, an outdated memo, or a rater your team can't get off Excel.

Customers

Start with one team. Keep building.

Single-tenant

A virtual machine of your own, with its own service account and data disk. Agents and workflows run in hardened sandboxes with web search and fetch switched off.

No training on your data

We do not train any model on your data, ours or anyone else’s. Not your rate plans, filings, books or submissions, not under any contract, and not at any tier.

SOC 2 Type II, SOX aligned

Continuously monitored by Rippling, with three separate environments, separated roles, and a full audit trail on every promotion.

ASOP-aligned

Built to the CAS syllabus and ASOPs 23, 41, and 56.

Tesora Workbench

Specialist AI agents do the work, read-only audit agents check it, and every figure cites its source.

Ingestion & analytics

Loss runs, policy data, filings and ledgers, read where they already live: Excel and SharePoint, Databricks, Snowflake, or your cloud storage. Anomalies are flagged on the way in, and every value keeps the cell or page it came from.

Comparative raters

Rebuild any rater, yours or a competitor’s, from Excel or SERFF filing PDFs in hours, with every factor cited to its source. Then run your book through them and see whether you will grow or bleed policies at each rate action.

Pricing & rater deployment

Build a rater by describing it in plain English, then test and refine it until the numbers are right. Every change is versioned and auditable, and you deploy without waiting on IT, actuarial, or underwriting. Your existing Excel raters import directly.

Reserving & Close

Most of the quarterly close, done. Anomalies flagged upon ingestion. IBNR recommendations with every value traced back to the source.

a month, saved

Per actuary, measured on our own workflows.

the output per actuary

Same headcount and standards, twice the completed work.

traceable to source

Every value Tesora populates cites the cell or page it came from.

How is this different from Claude, GPT or Copilot?

Those are models, and we route to them. The difference is everything built around them: your data wired in and vaulted for later recall, every figure citing where it came from, and a platform finally conducive to defensible, reproducible actuarial work. Tesora’s audit agents automatically flag issues like censored data and mismatched aggregations.

Can we import our existing rating models?

Yes, and it is usually the first thing we do. Send us your Excel rater and we rebuild it as a versioned model that runs your existing logic exactly as it does today. It must pass a test suite before anyone can deploy it, so you can check it against your own numbers first.

What integrations do you support?

Microsoft 365 (including Excel and SharePoint), Databricks, Snowflake, and the major cloud storage providers connect directly today. Policy and claims systems vary by carrier, so we map yours during onboarding at no extra cost. Everything else connects through our REST API. For a system we have not seen before, we will tell you once we have looked at it.

How does role separation work for SOX?

There are three separate environments: actuaries build in dev, QA validates in test, and underwriters use prod. An actuary cannot promote their own work to prod, which is the control your auditor cares about. Every promotion records who made it and when, and access follows the same split, so no one holds two roles at once.

Do you train on our data?

No. We do not train any model on your data, ours or anyone else’s, and every model provider we route to is bound by terms that forbid training on it. Two of them, Wafer and TypeSafe, keep nothing once they return a result. The others keep requests for a limited period under their own terms. We can send those terms to your security team before you sign.

Is confidential data protected?

Yes. You get your own single-tenant environment, a virtual machine and data disk of its own, and your team signs in through your own SSO. Your book is never pooled with anyone else’s data. If you prefer, we can deploy entirely inside your own infrastructure.

Extraction

Reads SERFF filings, Excel raters, and scripts, and cites every value to the cell it came from.

Pricing

Trends and on-levels your own history, then fits severity and frequency separately.

Rating

Turns the selected rates into a tested rater you can call from underwriting.

Reserving

Checks triangles against the last period, flags anomalies, fits the tail, and traces every IBNR value back to the source.