Opens in a new tab
Case studies/Insurance/RISK

How RISK keeps pricing and claims models accountable in real time

RISK monitors its AI models live, lets experts overrule decisions they disagree with, and logs every prediction for audit. Customers and supervisors can see why a decision was made.

About RISK

RISK is a strategic partner for insurance providers and advisors across financial services. RISK uses AI across their business labels to make faster, more intelligent decisions. These decisions affect customers, so RISK must be able to explain each one and have an oversight of all AI.

IndustryPensions and insurance
FocusMonitoring at runtime
The challenge

Challenges that stood in the way

Opaque decisions erode customer trust

Predictions must be explainable and with monitoring in place for internal teams.

Limited human control

Experts needed a way to review and overrule model decisions, and to see when models were drifting away from expert judgement.

Rising regulatory scrutiny

Insurance AI faces strict expectations on fairness, transparency and security, and the EU AI Act classifies life and health insurance pricing and risk assessment as high-risk.

The solution

How RISK and Deeploy work together

RISK uses Deeploy to run its AI models with oversight built in. Every prediction comes with an explanation that internal teams can read. Experts review decisions and overrule them when needed, and Deeploy tracks how often that happens. Monitoring, audit trails and documentation all sit in the same platform, so the risk function works from live information.

The platform

How the Deeploy platform is used

All insurance AI models sit in one central Deeploy platform, so teams can monitor every model reliably from a single place.

Explore AI InventoryAI Inventory in the Deeploy platform

Experts review and overrule AI predictions in Deeploy, a disagreement ratio flags when models need retraining and compliance templates keep every model aligned with the EU AI Act.

Explore Risk & Control FrameworksRisk and Control Frameworks in the Deeploy platform

Real-time alerts track model performance and drift, while Deeploy automatically logs every prediction, decision and change, so audit reports are quick to generate.

Explore Monitoring & Audit ReadinessMonitoring and Audit Readiness in the Deeploy platform
Outcomes

What changed for risk and compliance

A measurable signal for model risk

Deeploy tracks the disagreement ratio: how often experts overrule a model. A rising ratio shows that a model is moving away from expert judgement, before that shows up in losses or complaints. Risk officers get an early warning they can act on, and a clear trigger for retraining.

Human oversight with a record

Underwriters and claims handlers can review any model decision and overrule it. Each override is logged with the reviewer, the reason and the model version. Oversight becomes something RISK can demonstrate, not just describe in a policy.

Decisions customers can understand

Each pricing or claims outcome comes with the factors that drove it. Internal teams use those explanations to answer complaints and challenges.

Problems caught in real time

Performance and drift are monitored continuously for every model, with alerts when thresholds are crossed. A pricing model that starts to behave differently is flagged immediately, not at the next quarterly review.

Audit reports straight from the platform

Every prediction, decision and model change is logged automatically. Standard and custom compliance templates turn those logs into the documentation auditors and supervisors ask for. The same records support RISK's preparation for the EU AI Act, including for any use cases that fall into its high-risk category.

Technical details

How a pricing or claims decision is overseen

The model, deployed through Deeploy, produces a prediction such as a premium or a claims recommendation.

Deeploy attaches an explanation of the factors behind it, using standard or custom explainers.

An expert reviews the outcome and accepts or overrules it. The decision is logged against the prediction and model version.

Overrules feed the disagreement ratio and are collected as feedback for retraining.

Monitoring tracks performance and drift and raises an alert when thresholds are crossed.

Logs flow into compliance templates, producing audit reports and model documentation.

Who uses it
Data scientistsDeploy, monitor and retrain models
Underwriters and claims handlersReview and overrule model decisions
Risk and complianceTrack the disagreement ratio, review alerts, pull audit reports
Customer serviceUse explanations to answer customer questions
Case studies

More success stories

Cybersecurity

Hadrian scales AI and MLOps, deploying models 80% faster

See case study
Banking

Bunq governs AI-driven banking with 40–60% fewer false positives

See case study
Manufacturing

Vlisco keeps factory AI accurate, explainable and owned

See case study
Hospitals

Healthplus.ai earns clinicians' trust in infection prediction AI

See case study
Pensions

PGGM gives risk and compliance teams one view of every live AI

See case study
Mental healthcare

NiceDay keeps therapists in charge of every AI recommendation

See case study

Stay in control
after go-live

Monitor, trace and explain every AI system, and keep the evidence ready for review.