Actuarial Labs introduces Pia
The modeling platform
for insurance pricing.
Pia turns raw policy and claims extracts into filing‑ready GLMs and GAMs in minutes, with every number traced and every decision on record. No data‑engineering queue, no data‑science backlog.
- CanadaFSRA · AMF · AIRB
- US StatesNAIC · State DOIs
- United KingdomFCA · PRA
Who does what
The hand‑offs disappear.
A pricing review used to wait on a data engineer, then a data scientist. Pia now does both, and the actuary works on the part that needs judgment.
Handled by Pia
Data engineering
- Map columns to rolesTarget, exposure and 9 rating factors found
- Join policy and claims filesClaims summed per policy, exposure counted once
- Clean what extracts really look likeSentinels, currency typed as text, Excel dates
- Split and freeze the holdoutHalf of it stays sealed until the champion is chosen
Handled by Pia
Data science
- Explore every factorOne-way frequency, exposure and credibility intervals
- Tune splines and penaltiesSmoothing selected by REML
- Search for interactionsGBM-suggested pairs, kept only if the holdout improves
- Benchmark against a black boxGBM benchmark on the same frozen holdout
Stays with your team
The actuary
- Review relativities and credibilityThin levels are flagged for you
- Set trend and large-loss capsApplied to experience and recorded with the model
- Override and document whyYour reason goes into the audit trail
- Approve the model and the filingNothing is filed without your sign-off
How it works
From raw extract
to filed rate.
Pia does the data engineering and the data science. Your actuaries make the calls and sign the filing.
Step 01
Drop in raw extracts. Pia finds the schema.
Policy, claims and exposure files exactly as your admin system exports them. Pia recognises the target, the exposure and every rating factor from the data itself, reconciles the files and cleans what real extracts look like.
| IDpol | Unique on every row | Identifier |
| ClaimNb | Count, mostly zero | Target |
| Exposure | Fraction of a year | Exposure |
| DrivAge | Integer, 18 to 100 | Rating factor |
| BonusMalus | Integer, 50 to 230 | Rating factor |
| Region | 22 levels | Rating factor |
Step 02
Pia proposes. You approve.
Safe fixes are applied and logged. Anything that changes what the data means is proposed with its evidence and waits for an actuary. Nothing touches the data until you say so.
Step 03
Every model family, ranked on one holdout.
GLMs, shape-constrained GAMs, frequency × severity and a GBM benchmark, tuned and ranked in minutes. You see how much lift is left on the table, and what it costs to file.
Step 04
Relativities you can stand behind.
One-way experience beside every relativity, credibility-weighted where data is thin. Override a level and your reason lands in the audit trail, with your name on it.
Step 05
The filing writes itself.
Actuarial memorandum, exhibits and rate pages, drafted from the model with every number traced to its source. Edit any section; Pia keeps the numbers tied.
Selected relativities are credibility-weighted between indicated experience and current factors, and capped at ±10%. The selected relativity for Territory 08 is 1.180, for an overall indicated change of +6.2%.
Step 06
Reviewed before the regulator does.
Pia reads the filing as the examiner will, against the rules of the jurisdiction it is filed in. Every check shows the rule, its source, and what your filing says.
Why Pia
Built for the actuary whose name goes on the filing.
Move dramatically faster, without taking on more professional or regulatory risk.
A portion of the holdout stays sealed until model selection is complete.
Actual vs. expected performance, checked in every decile of risk.
Every run, override and approval is logged with who, when and why.
Customer-isolated, hosted in Canada, and never used to train shared AI models.
The benchmark library
57 books to test Pia on.
Every use case pricing teams meet, from frequency to large losses, each with the trap that catches real models. Open one and watch Pia reach a filable model.
Open the library
Bring one book.
Leave with a filing.
Pilots run on one line of business and one jurisdiction, with your actuaries in the loop from day one.