Pass the CertNexus CAIP — the First Time

1. An insurance company processes about 40,000 auto claims per month. Experienced adjusters spend most of their day approving routine, low-value claims that follow highly predictable patterns. Leadership sponsors an ML project that will automatically approve routine claims and route only unusual or high-value claims to human adjusters. Which business value case does this project primarily represent?

   • A. Replacing the entire claims department with a fully autonomous system
   • B. Increasing the total number of claims the company receives each month
   • C. Automating repetitive decisions so staff focus on complex work
   • D. Eliminating the need for any human oversight of claim decisions

   Answer: C. The classic ML value case here is automation of high-volume, repetitive decisions, which frees expert adjusters to work on the complex cases where their judgment adds the most value. Options A and D overstate what the project does: unusual claims are explicitly routed to humans, so people remain in the loop. Option B confuses operational efficiency with demand generation; ML does not change how many claims arrive.

   Exam domain: Understanding the Artificial Intelligence Problem

2. A government transportation agency can fund only one machine learning project this year and asks a consultant to rank four candidate initiatives by their probability of success as ML projects. Which initiative should be ranked highest?

   • A. Calculating toll charges, which are fully determined by a published rate table
   • B. Prioritizing which of 12,000 bridges to inspect first, using 20 years of historical inspection reports and sensor readings
   • C. Replacing paper permit applications with a standard online web form
   • D. Enforcing a new statute that requires exact, rule-defined compliance checks with no tolerance for error

   Answer: B. Bridge inspection prioritization has the shape of a strong ML use case: a large decision space, abundant historical data, and patterns too complex to encode as fixed rules. Options A and D are deterministic problems where the logic is fully specified: simple rules are more accurate, cheaper, and auditable, so ML adds risk without value. Option C is process digitization, not a prediction problem at all.

   Exam domain: Understanding the Artificial Intelligence Problem

3. An electronics manufacturer wants to automatically detect surface defects such as scratches and solder faults on circuit boards as they move down the assembly line, using cameras mounted above the conveyor. Which type of learning system best fits this business case?

   • A. Natural language processing (NLP)
   • B. Speech recognition of spoken audio
   • C. A recommendation system for users
   • D. Image recognition (computer vision)

   Answer: D. Detecting visual defects in camera images is a textbook computer vision / image recognition use case, typically implemented with convolutional neural networks. NLP and speech recognition operate on text and audio, not images, and recommendation systems suggest items to users rather than inspect products. Matching the data modality (images) to the learning system is the key skill tested here.

   Exam domain: Understanding the Artificial Intelligence Problem

4. A telecom operator's call center wants to automatically route incoming customer phone calls. The planned system will first convert each caller's spoken words into text, then determine the caller's intent from that text in order to select the correct department. Which combination of learning systems does this solution require?

   • A. Image recognition followed by natural language processing
   • B. Speech recognition followed by natural language processing
   • C. Speech recognition followed by a recommendation system
   • D. Robotics followed by speech recognition

   Answer: B. The pipeline has two distinct stages: speech recognition transcribes audio into text, and natural language processing (specifically natural language understanding) extracts the caller's intent from that text. Option C fails because recommenders suggest items based on user preferences, not classify intent from language. Options A and D involve modalities (images, physical actuation) that do not appear anywhere in this workflow.

   Exam domain: Understanding the Artificial Intelligence Problem

5. A video streaming service has three years of subscriber records, each labeled with whether the subscriber canceled within 90 days of that record. The company wants to predict which current subscribers are likely to cancel in the next 90 days so retention offers can be targeted. How should this problem be framed?

   • A. Reinforcement learning, with cancellations as negative rewards
   • B. Supervised learning, as a regression problem predicting a continuous value
   • C. Unsupervised learning, as a clustering problem
   • D. Supervised learning, as a binary classification problem

   Answer: D. Historical labels (canceled or not) plus a categorical yes/no prediction target make this supervised binary classification: the standard framing for churn. Clustering is unsupervised and would ignore the valuable labels the company already has. Regression predicts continuous quantities, and reinforcement learning applies to sequential decision-making with an agent acting in an environment, not to one-shot prediction from historical records.

   Exam domain: Understanding the Artificial Intelligence Problem

Want more? Finish the quiz above to unlock 15 more free questions, or open the full AIP-210 practice app.