CPCU 550 flashcards

32 cards. Read the term and the definition here. The deck at the top of this page is optional.

Structured data

Data organized in predefined rows and columns, such as policy and claims tables, that can be queried directly with standard database tools.

Unstructured data

Data without a predefined model, including free text, images, audio, and scanned documents, which requires extraction or natural language processing before analysis.

Data quality dimensions

The attributes used to judge fitness of data: accuracy, completeness, consistency, timeliness, validity, and uniqueness. Profiling against these dimensions precedes modeling.

Data governance

The framework assigning data ownership, stewardship, definitions, and access rights so that data is managed as a controlled organizational asset.

Metadata

Data that describes other data, such as source system, format, refresh date, and owner, enabling users to interpret and trust a dataset.

Data lineage

The documented path of data from its origin through each transformation to its final report, supporting audit, troubleshooting, and regulatory review.

ETL versus ELT

ETL transforms data before loading it into a warehouse; ELT loads raw data first and transforms it inside the target platform, common in cloud architectures.

System of record

The authoritative application for a data element, such as policy administration for coverage terms, which other systems must reference rather than duplicate.

Predictive model lifecycle

The sequence of defining the target, assembling and cleaning data, engineering features, splitting samples, training, validating, deploying, and monitoring the model.

Overfitting

A modeling failure in which the model captures noise in training data, producing strong training results but poor performance on validation, test, or new data.

Supervised learning

Modeling with labeled outcomes, such as known fraud or claim amounts, so the algorithm learns a mapping from inputs to the target.

Unsupervised learning

Modeling without labeled outcomes to find structure, such as clustering similar risks or flagging anomalies that differ from normal patterns.

Generalized linear model

A regression framework specifying a response distribution and link function, widely used in frequency-severity pricing for Poisson, gamma, and Tweedie targets.

AUC

Area under the receiver operating characteristic curve; it measures how well a model ranks positive cases above negative cases, with 0.5 equal to random.

Model calibration

The agreement between predicted and observed outcomes across score bands; a calibrated model's predicted loss levels match actual experience.

Model drift

Degradation of model performance after deployment caused by changes in the underlying population, behavior, or input data relationships.

Model risk management

Governance covering documented development, independent validation, periodic review, and controlled approval of overrides, changes, and retirement of models.

Feature engineering

Creating or transforming input variables, such as loss ratios or territory groupings, to improve a model's predictive power and interpretability.

Telematics data

Vehicle sensor data on mileage, speed, braking, and time of day used for usage-based insurance pricing, subject to consent and privacy constraints.

Embedded insurance

Distributing coverage at the point of a related transaction, such as a rental or purchase, through partner platforms and API integration with the insurer.

Smart contract

Self-executing code on a distributed ledger that triggers actions, such as payment, when predefined conditions are verified, supporting parametric coverage.

Robotic process automation

Software that performs repetitive, rule-based tasks such as data entry across systems; it follows defined rules and does not learn from data.

API

An interface letting separate systems exchange data and requests in real time, enabling quoting, binding, and partner integrations without manual rekeying.

Straight-through processing

Automated end-to-end handling of a transaction from submission to issuance without manual intervention, dependent on clean data and codified rules.

Confidentiality, integrity, availability

The core security objectives: limiting data access to authorized users, keeping data accurate and unaltered, and ensuring systems remain accessible when needed.

Multi-factor authentication

Access control requiring two or more independent credential types, drawn from knowledge, possession, and inherence, reducing risk from stolen passwords.

Least privilege

A security principle granting users only the access necessary for their role, limiting damage from error, misuse, or compromised accounts.

Defense in depth

Layering multiple independent controls, such as firewalls, segmentation, endpoint protection, encryption, and monitoring, so no single failure exposes data.

Ransomware

Malicious software that encrypts or locks systems and data and demands payment; recovery depends on immutable backups and a tested incident response plan.

Third-party risk management

Assessing and monitoring vendors with access to data or systems through pre-contract review, security clauses, and ongoing oversight of their controls.

FCRA adverse action notice

A notice required when a consumer report contributes to an unfavorable underwriting or credit decision, informing the individual of the reason and their rights.

NAIC Data Security Model Law

A model requiring licensed insurers to maintain an information security program, investigate breaches, and notify the insurance commissioner within set timelines.