I85PREP
⌘ K
CPCU 550 study materials
CPCU 550 study resources

CPCU 550 practice questions

One question at a time. Build your CPCU 550 understanding.

First 50 of 200 questions. Each one lists the choices, the correct answer, and an explanation. The checker at the top of this page saves progress in this browser when it is available.

Continue in the practice tool above to load more questions, 50 at a time.

Question 1Data and Technology (content domain; official weight breakdown not published in source)

An underwriter uses a predictive model that scores commercial property risks. Which term describes the model output that estimates the probability of a large fire loss for a given submission?

  1. A legacy system
  2. An audit trail
  3. A data warehouse
  4. A predictive score Correct answer.

Explanation

A predictive score is the numeric output a model produces for a specific record, here the estimated likelihood of a large fire loss. It is generated from input variables such as construction class, occupancy, and protection class. The other terms describe infrastructure or records, not a model's risk estimate.

Question 2Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants to know whether a machine learning model's accuracy holds when applied to a new region's book of business. Which practice best tests this?

  1. Increasing the number of variables in the model
  2. Validating the model on a held-out sample from the new region Correct answer.
  3. Deleting records that contain missing values
  4. Reviewing the model's source code line by line

Explanation

Out-of-sample validation on data from the target region shows whether the model generalizes beyond its training population. Code review and adding variables do not measure performance on new business. Deleting incomplete records changes the sample but does not test regional accuracy.

Question 3Data and Technology (content domain; official weight breakdown not published in source)

A claims department stores scanned documents, photographs, and emails in a system designed to hold unstructured content. Which technology category does this system represent?

  1. A relational database
  2. A content management system Correct answer.
  3. A spreadsheet model
  4. A rules engine

Explanation

Content management systems are built to store and retrieve unstructured items such as images, scanned documents, and messages. Relational databases are structured around tables and defined fields. Spreadsheets and rules engines serve calculation and decision logic rather than document storage.

Question 4Data and Technology (content domain; official weight breakdown not published in source)

An insurer's analytics team joins policy data, claims data, and third-party geocoding data into a single repository for reporting. What is this repository called?

  1. A data warehouse Correct answer.
  2. A firewall
  3. An operating system
  4. A mobile application

Explanation

A data warehouse consolidates data from multiple internal and external sources to support reporting and analysis. Firewalls, mobile applications, and operating systems are unrelated to integrated analytical storage. The defining feature is the integration of varied sources into one analytical repository.

Question 5Data and Technology (content domain; official weight breakdown not published in source)

Before loading customer data into a marketing analytics platform, a team standardizes address formats and removes duplicate customer records. Which data management activity does this describe?

  1. Data cleansing Correct answer.
  2. Data archiving
  3. Data encryption
  4. Data mining

Explanation

Data cleansing, or data quality management, corrects inconsistencies and removes duplicates so downstream analysis is reliable. Data mining seeks patterns, encryption protects confidentiality, and archiving stores inactive records. Standardization and deduplication are hallmarks of cleansing.

Question 6Data and Technology (content domain; official weight breakdown not published in source)

A personal auto insurer applies a model that assigns each applicant a risk score, and the score determines whether the policy is issued at standard or preferred rates. Which concern is most directly raised by this use of the model?

  1. Whether the model generates paper policy forms
  2. Whether the model can process claims payments
  3. Whether the model reduces the need for reinsurance
  4. Whether the model's outputs are used fairly and comply with rating regulation Correct answer.

Explanation

When model outputs drive underwriting or pricing decisions, fairness, transparency, and compliance with insurance rating rules become central. Claims payment processing, reinsurance purchasing, and forms generation are separate operational functions. The key issue is the governance of automated decisions affecting applicants.

Question 7Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants to detect unusual patterns in claim submissions, such as repeated claimants appearing across unrelated policies, to identify possible fraud. Which analytical approach is most suited to this task?

  1. Straight-line forecasting of premium volume
  2. Network link analysis of relationships among claimants and policies Correct answer.
  3. Manual review of a single claim file
  4. Descriptive reporting of monthly claim counts

Explanation

Network or link analysis maps relationships among people, policies, and claims to reveal hidden connections that suggest organized fraud. Simple reporting and trend forecasting summarize aggregate activity without exposing relationships. Reviewing one file cannot reveal cross-policy patterns.

Question 8Data and Technology (content domain; official weight breakdown not published in source)

A company's data governance program assigns accountability for the accuracy of each critical data element. What is the primary purpose of this assignment?

  1. To establish ownership and accountability for data quality Correct answer.
  2. To eliminate the need for data backups
  3. To reduce the cost of hardware purchases
  4. To replace the need for cybersecurity controls

Explanation

Data governance designates data owners and stewards so that someone is accountable for accuracy, definitions, and appropriate use. It does not reduce hardware costs, remove backup needs, or substitute for security controls. Clear ownership is the mechanism that sustains data quality over time.

Question 9Data and Technology (content domain; official weight breakdown not published in source)

An insurer stores policyholder records in the cloud and must ensure that only authorized employees can view sensitive personal information. Which control most directly addresses this requirement?

  1. Role-based access control Correct answer.
  2. Batch processing
  3. Data compression
  4. Load balancing

Explanation

Role-based access control restricts data access according to job function, so only authorized employees can view sensitive records. Compression reduces file size, load balancing distributes traffic, and batch processing groups transactions. Only access control governs who may view protected information.

Question 10Data and Technology (content domain; official weight breakdown not published in source)

A model predicts claim severity, and its developers report that the model explains a high share of the variation in historical losses. Why should an underwriter still treat the model with caution?

  1. Historical data always contain no useful information
  2. High explanatory power on historical data does not guarantee accurate predictions on future risks Correct answer.
  3. Explanatory power measures the cost of running the model
  4. Models that fit historical data well can never be used in production

Explanation

A strong fit to historical data may reflect overfitting or a changing environment, so future accuracy is not assured. Models with good historical fit are commonly deployed after validation. Explanatory power is a statistical measure, not a cost measure, and historical data usually carry real signal.

Question 11Data and Technology (content domain; official weight breakdown not published in source)

An insurer replaces paper inspection reports with mobile devices that transmit field data directly to headquarters. Which benefit is most directly associated with this change?

  1. Elimination of the need for underwriting judgment
  2. Reduced data entry errors and faster availability of inspection results Correct answer.
  3. Guaranteed reduction in claim frequency
  4. Removal of all regulatory reporting duties

Explanation

Capturing field data digitally reduces manual transcription errors and shortens the time before results reach decision makers. Technology does not eliminate underwriting judgment, guarantee lower claim frequency, or remove regulatory obligations. The direct operational gain is accuracy and speed.

Question 12Data and Technology (content domain; official weight breakdown not published in source)

An analytics team wants to identify which combination of variables most strongly predicts whether a small business policy will lapse. Which technique is designed for this purpose?

  1. Network routing
  2. Version control
  3. Regression analysis Correct answer.
  4. Disk defragmentation

Explanation

Regression analysis estimates the relationship between predictor variables and an outcome such as policy lapse. Defragmentation, routing, and version control are information technology maintenance functions. Regression is the standard statistical tool for quantifying predictive relationships.

Question 13Data and Technology (content domain; official weight breakdown not published in source)

A carrier wants to segment its commercial auto book into groups with similar loss experience so that underwriters can tailor pricing. Which technique is most appropriate?

  1. Data purging
  2. Cluster analysis Correct answer.
  3. Screen scraping
  4. Password hashing

Explanation

Cluster analysis groups records with similar characteristics, which supports segmentation of risks by loss experience. Hashing protects passwords, purging deletes data, and screen scraping extracts displayed information. Only clustering performs the grouping task described.

Question 14Data and Technology (content domain; official weight breakdown not published in source)

A predictive model used in underwriting relies on a variable that is a proxy for a protected characteristic. What is the most significant risk of this practice?

  1. The model will run more slowly than models without proxies
  2. The model may produce unfairly discriminatory outcomes even without using the characteristic directly Correct answer.
  3. The model will require more storage space
  4. The model will be unable to produce numeric scores

Explanation

Proxy variables can carry the effect of a protected characteristic, producing discriminatory outcomes even when that characteristic is excluded. Performance, storage, and output format are technical matters unrelated to the fairness concern. Regulators focus on disparate impact arising from such proxies.

Question 15Data and Technology (content domain; official weight breakdown not published in source)

An insurer maintains a formal inventory describing what data it holds, where the data originate, and who may use them. What is this inventory called?

  1. A premium ledger
  2. A claims register
  3. A data catalog Correct answer.
  4. A loss run

Explanation

A data catalog documents data assets, their sources, definitions, and permitted uses, supporting governance and discovery. Claims registers, premium ledgers, and loss runs are operational or financial records about policies and claims. The catalog is the metadata inventory of the organization's data.

Question 16Data and Technology (content domain; official weight breakdown not published in source)

A company plans to use customer data for a new analytics initiative. Which action best reflects responsible data use?

  1. Sharing the data broadly with vendors before any agreements are signed
  2. Using all available data because it is already stored in company systems
  3. Confirming that the intended use is consistent with disclosed purposes and applicable privacy requirements Correct answer.
  4. Retaining the data indefinitely to avoid future collection costs

Explanation

Responsible use requires that new analytics purposes align with what was disclosed to individuals and with privacy law. Prior storage does not authorize unlimited reuse, vendor sharing requires agreements, and indefinite retention conflicts with minimization principles. Purpose alignment is the governing test.

Question 17Data and Technology (content domain; official weight breakdown not published in source)

Underwriters at a carrier receive automated alerts when a submitted risk's model score falls outside an expected range. What is the primary value of this control?

  1. It guarantees that no inaccurate data enter the system
  2. It removes the need for any documented underwriting guidelines
  3. It eliminates the possibility of model error
  4. It flags unusual cases for human review before a decision is finalized Correct answer.

Explanation

Exception alerts direct human attention to atypical cases, supporting oversight of automated scoring. They do not replace written guidelines, prevent bad data entry, or eliminate model error. The control adds a review layer where risk of error is highest.

Question 18Data and Technology (content domain; official weight breakdown not published in source)

An insurer uses text analytics on recorded customer service calls to identify frequent causes of complaints. Which data type is being analyzed?

  1. Transactional data
  2. Unstructured data Correct answer.
  3. Structured data
  4. Metadata

Explanation

Recorded and transcribed speech is unstructured data because it lacks a predefined tabular format. Structured and transactional data reside in defined fields and records. Metadata describes other data rather than consisting of the call content itself.

Question 19Data and Technology (content domain; official weight breakdown not published in source)

A carrier wants to compare the profitability of two agency segments after adjusting for differences in premium volume. Which analytical step is most important?

  1. Counting the total number of claims without regard to premium
  2. Excluding all segments with fewer than one hundred policies
  3. Normalizing the comparison using a common measure such as loss ratio Correct answer.
  4. Reviewing only the largest single claim in each segment

Explanation

A ratio such as loss ratio adjusts for size differences and allows a fair comparison between segments. Raw claim counts and single large claims distort comparisons across differing premium volumes. Arbitrarily excluding smaller segments removes information rather than normalizing it.

Question 20Data and Technology (content domain; official weight breakdown not published in source)

An insurer's systems cannot share data because each department maintains its own incompatible database. Which term describes this condition?

  1. Data silos Correct answer.
  2. Data masking
  3. Data redundancy
  4. Data latency

Explanation

Data silos exist when separate systems or departments hold isolated data that cannot be readily integrated. Redundancy is duplicated storage, latency is delay in availability, and masking hides sensitive values. The defining problem here is isolation across incompatible systems.

Question 21Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants to forecast claim frequency for a new line of business with no historical experience of its own. Which approach is most reasonable?

  1. Selecting a forecast at random to avoid bias
  2. Borrowing experience from similar lines and adjusting for known differences Correct answer.
  3. Using only the single largest competitor's published results
  4. Assuming claim frequency will be zero until data accumulate

Explanation

When direct experience is unavailable, actuaries use analogous data and adjust for differences in exposure and risk characteristics. Assuming zero frequency ignores known risk, relying on one competitor is unreliable, and random selection is indefensible. Credible analogy plus adjustment is the standard approach.

Question 22Data and Technology (content domain; official weight breakdown not published in source)

A company adopts a policy requiring that sensitive data be encrypted both while stored and while transmitted. Which risk does this policy most directly reduce?

  1. Delays in claims processing
  2. Inaccurate calculation of loss ratios
  3. Unauthorized disclosure of data if storage or transmission is compromised Correct answer.
  4. Errors in policyholder addresses

Explanation

Encryption at rest and in transit protects confidentiality if storage media or network traffic are intercepted. It does not correct calculation errors, processing delays, or address inaccuracies. The control targets unauthorized access to sensitive information.

Question 23Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants a repeatable, documented process for moving data from source systems into an analytics environment, including checks for completeness. What is this process called?

  1. A marketing funnel
  2. A loss reserve
  3. An ETL pipeline Correct answer.
  4. A reinsurance treaty

Explanation

Extract, transform, and load pipelines move data from sources into analytics environments and can include quality checks at each stage. Marketing funnels, reinsurance treaties, and loss reserves are business concepts unrelated to data movement. The pipeline is the technical process that ensures reliable data delivery.

Question 24Data and Technology (content domain; official weight breakdown not published in source)

A model used to triage claims for possible fraud produces many false positives, burdening investigators. Which action best addresses this problem?

  1. Raising the score threshold for referral while monitoring the effect on missed fraud Correct answer.
  2. Ignoring the false positives and referring every claim
  3. Disabling the model entirely without further analysis
  4. Removing all human review from the process

Explanation

Adjusting the referral threshold trades off investigative workload against detection, and the effect on missed fraud should be monitored. Eliminating review, abandoning the model outright, or referring every claim are disproportionate responses. Calibrating the threshold is the targeted remedy.

Question 25Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants to give agents real-time quotes by connecting its rating engine directly to agency management systems. Which technology enables this integration?

  1. A paper application form
  2. A telephone hotline
  3. An application programming interface Correct answer.
  4. A data archive

Explanation

Application programming interfaces let separate systems exchange data automatically, enabling real-time rating within agency systems. Paper forms, archives, and hotlines involve manual or stored processes and cannot provide automated integration. The interface is the standard mechanism for system-to-system data exchange.

Question 26Data and Technology (content domain; official weight breakdown not published in source)

An insurer's analytics unit needs a single trusted repository for structured policy and claims data that supports historical reporting and is updated on a scheduled batch basis. Which type of system is designed primarily for this purpose?

  1. A content management system
  2. A rules engine
  3. A customer portal
  4. A data warehouse Correct answer.

Explanation

A data warehouse consolidates structured data from multiple source systems into a single repository optimized for query and reporting, typically loaded on a scheduled batch cycle. It differs from operational systems, which are built to process transactions rather than to support broad historical analysis.

Question 27Data and Technology (content domain; official weight breakdown not published in source)

An underwriting data set contains a field for annual revenue in which some records are blank, some list values in thousands, and some include currency symbols. What is the most appropriate first step before using the field in analysis?

  1. Standardize and validate the values, then decide how to handle missing entries Correct answer.
  2. Delete every record that contains a blank field
  3. Move the field to a separate spreadsheet and ignore it
  4. Average the field and replace all entries with that average

Explanation

Data quality problems such as inconsistent units, stray characters, and missing values must be resolved through standardization, validation, and a documented treatment for missing data before analysis. Deleting all affected records or overwriting values with an average can distort results and hide the underlying issue.

Question 28Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants to store large volumes of semi-structured and unstructured data, such as claim photos and call transcripts, and expects the volume to grow rapidly. Which approach best fits this requirement?

  1. A single spreadsheet workbook
  2. A data lake Correct answer.
  3. A relational database with fixed columns
  4. A printed archive

Explanation

A data lake stores raw data in native formats, including unstructured and semi-structured content, and scales readily as volume grows. Relational databases and spreadsheets impose rigid schemas that fit poorly with transcripts and images at scale.

Question 29Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants a visual summary that shows the distribution of claim counts by cause of loss so that managers can quickly see which causes occur most often. Which tool or technique is most directly suited to this need?

  1. A backup schedule
  2. A data dictionary
  3. A regression equation
  4. A bar chart of claim counts by cause Correct answer.

Explanation

A bar chart displays counts or frequencies across categories, making it an effective way to compare causes of loss at a glance. Regression, data dictionaries, and backup schedules serve analytical, documentation, and operational purposes rather than categorical comparison.

Question 30Data and Technology (content domain; official weight breakdown not published in source)

An insurer's claims staff need a report that shows open claim counts by adjuster, refreshed each morning. Which characteristic of the report matters most for this operational use?

  1. It is timely and reflects current data Correct answer.
  2. It contains as many fields as possible
  3. It uses the most complex statistical model available
  4. It is stored only in printed form

Explanation

Operational reporting depends on timeliness and relevance, so the report should reflect current claim status when staff use it. Complexity and field volume do not improve usefulness and can obscure the information needed for daily decisions.

Question 31Data and Technology (content domain; official weight breakdown not published in source)

An insurer's data team wants to combine policy records from three systems that use different customer identifiers. What must be established to reliably link records that refer to the same customer?

  1. A shorter retention period
  2. A common key or matching rule Correct answer.
  3. A larger server
  4. A new marketing campaign

Explanation

Linking records across systems requires a shared identifier or a defined matching rule so that records referring to the same entity can be associated correctly. Infrastructure, retention, and marketing changes do not resolve identity conflicts between sources.

Question 32Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants to protect policyholder data by limiting each employee's access to only the records needed for that employee's job. Which control concept does this describe?

  1. Version control
  2. Load balancing
  3. Least privilege Correct answer.
  4. Data duplication

Explanation

Least privilege restricts access to the minimum necessary for a role, reducing exposure of sensitive records. Duplication, load balancing, and version control address availability, performance, and code management rather than access restriction.

Question 33Data and Technology (content domain; official weight breakdown not published in source)

An insurer's analytics team uses a model to rank applications by likelihood of inspection, and inspectors work through the list from the top. What does the ranking allow the insurer to do?

  1. Guarantee that ranked risks will produce losses
  2. Replace all underwriting judgment
  3. Prioritize limited inspection resources on higher-risk submissions Correct answer.
  4. Eliminate the need for any inspections

Explanation

A ranking model helps allocate scarce resources by directing attention to submissions with higher predicted risk. It does not eliminate inspections, guarantee outcomes, or replace professional judgment.

Question 34Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants a documented, searchable description of each data element in its claims system, including definition, format, and permitted values. What is this artifact called?

  1. An audit opinion
  2. A loss run
  3. A reinsurance treaty
  4. A data dictionary Correct answer.

Explanation

A data dictionary defines data elements, their formats, and allowable values so users interpret fields consistently. Loss runs, reinsurance treaties, and audit opinions are unrelated business documents.

Question 35Data and Technology (content domain; official weight breakdown not published in source)

An insurer's fraud analytics team builds a model that flags claims for review. Investigators can examine only a small share of claims, so the team tunes the model to rank claims by predicted fraud likelihood. Which evaluation approach is most appropriate for this use?

  1. Measuring how well the model ranks claims at the top of the list Correct answer.
  2. Reviewing the model's programming language
  3. Counting the total number of claims received
  4. Measuring the average claim payment

Explanation

When only a limited number of cases can be reviewed, the value of the model depends on how well it concentrates true positives near the top of the ranking. Overall payment averages, claim volume, and implementation language do not measure ranking performance.

Question 36Data and Technology (content domain; official weight breakdown not published in source)

An insurer builds a model that predicts small commercial policy lapses using data from 2018 through 2021, then tests it on 2022 policies. What is the purpose of the 2022 test?

  1. To estimate how well the model performs on data it has not seen Correct answer.
  2. To reduce the size of the training data
  3. To increase the number of predictors in the model
  4. To confirm that the model's code compiles

Explanation

Holding out later data provides an estimate of out-of-sample performance, which is more informative than performance on the data used to fit the model. It does not change the predictor set, shrink training data as a goal, or test code compilation.

Question 37Data and Technology (content domain; official weight breakdown not published in source)

An insurer's model shows strong accuracy in the aggregate but performs poorly for a small segment of rural policyholders. Which concern does this illustrate?

  1. Unequal performance across population segments Correct answer.
  2. Insufficient network bandwidth
  3. Poor spreadsheet formatting
  4. Excessive data storage

Explanation

A model that performs well overall but poorly for a subgroup raises concerns about disparate performance and fairness across segments. Storage, bandwidth, and formatting are technical matters that do not explain subgroup performance gaps.

Question 38Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants to use an external data source that contains information about applicants' purchasing habits. Before using it in underwriting, which action is most important?

  1. Blend the data with internal records without documentation
  2. Verify the source's accuracy, permissible purpose, and compliance with applicable rules Correct answer.
  3. Use the data immediately because it is externally produced
  4. Assume the vendor has resolved all legal issues

Explanation

External data must be evaluated for accuracy, permissible use, and regulatory compliance before it influences underwriting decisions. Assuming vendor compliance or skipping documentation leaves the insurer exposed to error and legal risk.

Question 39Data and Technology (content domain; official weight breakdown not published in source)

An insurer's analytics team needs to join policy data with geospatial hazard data. The policy file uses ZIP codes while the hazard file uses latitude and longitude. What must the team do before joining the files?

  1. Delete the location fields from both files
  2. Increase the number of records in each file
  3. Encrypt both files before joining
  4. Convert one location reference into a form compatible with the other Correct answer.

Explanation

Records can be joined only when location references share a common form, so one representation must be converted or mapped to the other. Deleting fields, adding records, or encrypting files does not resolve the mismatch.

Question 40Data and Technology (content domain; official weight breakdown not published in source)

An insurer's data governance council must decide who may approve changes to a critical claims data field used in regulatory reporting. Which role should hold this authority?

  1. The software vendor that built the system
  2. The mailroom supervisor
  3. The designated data owner for that field Correct answer.
  4. Any analyst who uses the field

Explanation

Data governance assigns accountability for a data element to a designated owner who approves changes affecting its quality and use. Users, vendors, and unrelated staff lack the assigned accountability for that field.

Question 41Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants to automate the extraction of coverage limits and effective dates from thousands of scanned policy documents. Which technology is best suited to this task?

  1. A dashboard filter
  2. A backup rotation
  3. Optical character recognition combined with document extraction rules Correct answer.
  4. A relational join

Explanation

Extracting structured fields from scanned documents requires converting images to text and then applying extraction logic to identify the relevant values. Joins, dashboard filters, and backup rotation do not process document images.

Question 42Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants to compare loss ratios across regions while adjusting for differences in the mix of business in each region. Which analytical technique is most appropriate?

  1. A list of regional office addresses
  2. A pie chart of premium by region
  3. A simple count of claims by region
  4. A multivariate model that controls for business mix Correct answer.

Explanation

Adjusting for differences in business mix requires a technique that controls for multiple variables simultaneously, such as a multivariate model. Simple counts, pie charts, and address lists do not isolate the effect of mix on loss ratios.

Question 43Data and Technology (content domain; official weight breakdown not published in source)

An insurer's analytics team wants to reduce the number of variables in a model while retaining most of the information they contain. Which technique is designed for this purpose?

  1. Principal component analysis Correct answer.
  2. Data archiving
  3. Random sampling
  4. Manual data entry

Explanation

Principal component analysis transforms correlated variables into a smaller set of components that capture most of the variation in the original data. Sampling, archiving, and manual entry do not reduce dimensionality while preserving information.

Question 44Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants to ensure that a model used for claims triage can be explained to regulators and claimants. Which property of the model is most relevant?

  1. The number of users logged in
  2. Interpretability of how inputs influence the output Correct answer.
  3. The physical size of the server hosting it
  4. The color scheme of its dashboard

Explanation

Explaining a decision requires understanding how model inputs influence the output, which is the property of interpretability. Server size, user counts, and dashboard design do not explain model reasoning.

Question 45Data and Technology (content domain; official weight breakdown not published in source)

An insurer's data team discovers that a vendor feed arrives with duplicate policy records each month. Which control most directly prevents duplicates from entering the analytics environment?

  1. A longer data retention period
  2. A weekly staff meeting
  3. A de-duplication rule applied during data loading Correct answer.
  4. A larger storage allocation

Explanation

Applying a de-duplication rule at the point of loading prevents duplicate records from entering downstream systems. Storage, meetings, and retention changes do not detect or remove duplicate records.

Question 46Data and Technology (content domain; official weight breakdown not published in source)

An insurer's model for predicting claim severity uses a variable that is only available after a claim is closed. Why is this a problem for the intended use?

  1. The variable has too many decimal places
  2. The variable is too expensive to store
  3. The variable would not be available when the prediction must be made Correct answer.
  4. The variable is stored in the wrong file format

Explanation

A predictor must be available at the time the prediction is needed; a post-close variable creates leakage and cannot support decisions made earlier in the claim lifecycle. Storage cost, precision, and file format are secondary technical concerns.

Question 47Data and Technology (content domain; official weight breakdown not published in source)

An insurer's analytics platform must support both a nightly batch scoring job and immediate scoring when a customer submits an online application. Which architecture best meets both needs?

  1. A printed report delivered weekly
  2. A single nightly batch job only
  3. A manual scoring process
  4. A design that combines batch processing with a real-time scoring service Correct answer.

Explanation

Meeting both scheduled and immediate scoring requirements calls for an architecture that combines batch processing with a real-time service. Batch-only, manual, or printed approaches cannot deliver immediate results for online applications.

Question 48Data and Technology (content domain; official weight breakdown not published in source)

An insurer's model performs well in testing but degrades sharply after deployment because customer behavior shifted following a change in the economy. What is the best description of this problem?

  1. Model drift caused by a change in the underlying population Correct answer.
  2. A software licensing issue
  3. A hardware failure
  4. A data entry typo

Explanation

Model drift occurs when the relationship between inputs and outcomes changes in the population, causing performance to decline after deployment. Licensing, hardware, and isolated typos do not explain a broad performance shift tied to economic change.

Question 49Data and Technology (content domain; official weight breakdown not published in source)

An insurer wants to combine structured claims data with free-text adjuster notes to improve severity predictions. Which approach best uses both data types?

  1. Discard the notes because they are unstructured
  2. Convert all structured variables to text
  3. Apply text analytics to the notes and combine the resulting features with structured variables in one model Correct answer.
  4. Use only the notes and ignore structured data

Explanation

Combining derived text features with structured variables lets the model use information from both sources. Discarding notes or structured data wastes available signal, and converting structured variables to text does not improve modeling.

Question 50Data and Technology (content domain; official weight breakdown not published in source)

An insurer's model uses a proxy variable that correlates with a protected characteristic, producing adverse outcomes for a group even though the characteristic was not an input. What is the most appropriate response?

  1. Increase the model's complexity
  2. Assess the variable's disparate impact and consider removing or adjusting it Correct answer.
  3. Keep the variable because the characteristic was not directly used
  4. Add the protected characteristic as an input

Explanation

Indirect discrimination can occur through proxy variables, so the insurer should assess disparate impact and consider removing or adjusting the variable. Relying on the absence of the characteristic as a defense ignores proxy effects, and adding the characteristic or complexity does not resolve the concern.