Figura Analytics
Predictive underwriting that actuaries trust — and regulators can audit.
Figura's actuarial team had decades of underwriting expertise, but the company's growth was outpacing the speed at which that expertise could reach individual decisions. Qubiz built the platform that turned actuarial knowledge into a live underwriting capability.
The challenge
The expertise existed. It just couldn't reach the underwriter on the line.
Figura's loss ratios were drifting in the wrong direction across two specific lines of business. Internal analysis had identified the early warning patterns, but communicating them back to underwriters as actionable guidance — consistently, on every quote, in the moment of decision — required something the existing underwriting platform could not do. Any solution would have to satisfy the model-risk expectations of an insurance regulator before it touched a binding decision.
Translate actuarial expertise into a model the underwriter never has to question.
Three phases, each gated by actuarial sign-off and model-risk review.
- 01
Hypothesis design
Worked with the actuarial team to articulate the underwriting hypotheses worth modelling. Without that step, the model wouldn't be answering the right question.
- 02
Model construction
An ensemble trained on twelve years of policy and claims data, calibrated and stress-tested across line-of-business segments before any binding decision touched it.
- 03
Workflow embedding
The model lives inside the existing underwriting platform with an explicit rationale on every score. Recommendations, not replacements.
- 04
Continuous governance
Drift monitoring, scheduled retraining, and a documented escalation path — aligned with regulator expectations from day one.
WHAT WE BUILT
Underwriting hypothesis design
Worked with Figura's actuarial leads to articulate the underwriting hypotheses worth modelling — before writing a line of training code. The model targets the questions actuaries already ask, not the ones a textbook says to ask.
Predictive risk model
An ensemble model trained on twelve years of policy and claims data. Calibrated against held-out cohorts and stress-tested across line-of-business segments before any underwriting decision touched it.
Underwriter-in-the-loop
The model surfaces inside the existing underwriting workflow with an explicit rationale — the underwriter sees what drove the score, not just the score. Recommendations, not replacements.
Model risk framework
Full model risk documentation aligned with insurance regulator expectations. Versioning, monitoring, drift alerts, and a clear escalation path the moment performance degrades.
The outcome
Loss ratios moving in the right direction. Actuaries still in charge.
22%
Reduction in loss ratio across the two targeted lines of business.
1.8x
Faster quote turnaround through embedded model-driven guidance.
12 yrs
Of historical claims data feeding the production model.
100%
Of binding decisions covered by full model-risk documentation.
Most AI vendors arrive with a model and a deck. Qubiz arrived with questions for our actuarial team. That's why what they built is the model we still use.
Real-time warehouse intelligence across 12 automated sites.
Unified sensor, WMS and ERP data into a single operational platform — eliminating manual reporting and enabling predictive maintenance at scale.
Automated donation processing at scale — with full compliance.
Built an end-to-end intake and compliance pipeline that processes millions of transactions per year with zero manual intervention.