CPCU CPCU 550Maximizing Value with Data and Technology

CPCU

2 hours·85 multiple-choice (a handful may be unscored pretest items)·≈65 study hours

Overview

CPCU 550 introduces data and technology as a competitive tool: the data analytics lifecycle (define the problem, collect/clean data, analyze, communicate, deploy, monitor), the descriptive/predictive/prescriptive analytics continuum, big data sources including telematics, IoT sensors, and unstructured text/image data, an accessible introduction to machine learning (supervised vs. unsupervised, classification vs. regression, overfitting/underfitting), model evaluation metrics, and the ethical and privacy issues in using data and AI across underwriting, claims, and marketing.

Duration: 2 hours
Questions: 85 multiple-choice (a handful may be unscored pretest items)
Style: Computer-based, 4 answer choices, scenario-based stems
Credit: One of the CPCU program's course requirements
Passing: Scaled score; approximately 60/100 is the pass threshold set by The Institutes

Syllabus map

Data analytics lifecycle & data governance

20%

Actuarial Data Quality and ASOP 23

Descriptive, predictive & prescriptive analytics

20%

GLMs and Predictive Modeling in Pricing

Big data sources & emerging technology (telematics, IoT, AI)

20%

Machine learning fundamentals (supervised/unsupervised)

20%

GLMs and Predictive Modeling in Pricing

Model evaluation metrics

10%

Statistics and Estimation for Actuaries

Data ethics & privacy

10%

Key formulas

Confusion-matrix metrics for a binary classifier: $\text{Accuracy}=\frac{TP+TN}{TP+TN+FP+FN},\quad \text{Precision}=\frac{TP}{TP+FP},\quad \text{Recall (sensitivity)}=\frac{TP}{TP+FN}$

Lift: how much better a model identifies positives than a random selection of the same size: $\displaystyle \text{Lift}=\frac{\text{Response rate in targeted decile}}{\text{Overall response rate}}$ .

Overfitting intuition: as model complexity rises, in-sample error keeps falling but out-of-sample (holdout) error eventually rises — the gap between the two is the overfitting signal.

Bias-variance framing (conceptual): $\text{Expected error}\approx \text{Bias}^2+\text{Variance}+\text{Irreducible error}$ ; simple models are high-bias/low-variance, complex models are low-bias/high-variance.

Study strategy

Learn the stages of the analytics lifecycle in order — sequencing questions are common.
Be able to classify a described technique (regression, decision tree, clustering) as supervised or unsupervised and as classification or regression.
Memorize the confusion-matrix metrics (accuracy, precision, recall) and be ready to compute one from given TP/FP/TN/FN counts.
Understand overfitting conceptually: a model that fits training data almost perfectly but performs poorly on new data.
Review telematics and IoT examples (usage-based auto insurance, connected home sensors, wearables) and how each changes underwriting or claims.
Know the core data-ethics concerns: proxy discrimination, explainability, consent, and data security/privacy regulation.

Common traps

Confusing supervised learning (labeled outcome, e.g., predicting claim cost) with unsupervised learning (no labeled outcome, e.g., customer segmentation/clustering).
Treating a high in-sample accuracy as proof a model will generalize well — ignoring holdout/out-of-sample validation.
Mixing up precision (of predicted positives, how many are correct) and recall (of actual positives, how many were found).
Assuming more data automatically means better decisions without addressing data quality and governance.
Overlooking that a facially neutral variable correlated with a protected class can still create proxy discrimination.

← CPCU CPCU 540 CPCU CPCU 551 →