Free PMI CPMAI Practice Questions with Answers

74 free PMI CPMAI practice questions from the CertStash pack of 295, each with the correct answer and a full explanation. The rule is the same for every exam: ten free on the product page, 74 free here, all 295 in the pack, and each step includes the one before it.

Get all 295 questions (US$39) · Download the first 10 as a PDF

Question 1

Your team is working on an NLP model and has just operationalized the first model. Your team makes updates to the model, overwrites the original model, and puts this new model into operation. However, one of the teams using the model has seen a decrease in performance and is asking to use the original model.

What critical error did your team make?

They did not have data governance in place
They did not practice model versioning and keep all versions of the model
They did not have a model retraining pipeline that took into account models
They did not practice model iteration and properly iterate on the model

Show answer and explanation

Correct answer: B. They did not practice model versioning and keep all versions of the model

Model versioning is the practice of keeping every trained model as an immutable, retrievable artifact rather than overwriting it. CPMAI Phase VI treats a model as a released asset: when a new version underperforms for some consumers, the previous version must still be available so the team can roll back while they investigate. Overwriting the original destroyed the only copy of a working model, which is why the request cannot be satisfied.

Why the other options are wrong

A. Data governance concerns how data is classified, secured and accessed. No data was mishandled here - a model artifact was destroyed.
C. A retraining pipeline governs when a model is refreshed. The team did retrain; the failure was not retaining the earlier result.
D. The team did iterate - they produced a second model. Iterating is correct CPMAI practice, so this describes what they did right, not the error.

Question 2

Enhancing and cleaning data is an important action during which phase of CPMAI?

Phase VI
Phase I
Phase V
Phase III
Phase II
Phase IV

Show answer and explanation

Correct answer: D. Phase III

Phase III of CPMAI is Data Preparation. This is where raw data collected in Phase II is cleaned, de-duplicated, corrected, standardized and enriched, and where feature engineering and exploratory analysis take place. Enhancing and cleaning are the defining activities of this phase, and they must be complete before Phase IV Data Modeling can begin, because a model can only be as good as the data prepared for it.

Why the other options are wrong

A. Phase VI is Model Operationalization - deploying, monitoring and versioning a finished model.
B. Phase I is Business Understanding - defining the problem, the ROI and the AI Go/No Go decision.
C. Phase V is Model Evaluation - measuring whether the model meets the business goal set in Phase I.
E. Phase II is Data Understanding - identifying what data exists, where it lives and whether it is sufficient.
F. Phase IV is Data Modeling - selecting algorithms and training the model on data already prepared.

Question 3

Your team is ready to operationalize the model they have been working on. It’s a model that is meant to be used on an “edge device”, specifically a mobile phone and the user may sometimes be in remote locations without regular access to the internet.

What’s the most important thing to consider here?

Make sure that you can use Generative AI solutions on an edge device
Make sure the model lives in a hybrid environment
Make sure the model is available over a cloud-based API
Make sure the model lives on the edge device so it can be used regardless of internet connection

Show answer and explanation

Correct answer: D. Make sure the model lives on the edge device so it can be used regardless of internet connection

Operationalization has to match the environment the model actually runs in. The stated constraint is that the device is frequently offline, so any deployment that depends on a network round trip will fail exactly when the user needs it. Placing the model on the edge device itself means inference happens locally and the application keeps working regardless of connectivity, which is the standard CPMAI answer for intermittently connected edge deployments.

Why the other options are wrong

A. Whether the solution is generative is unrelated to the problem; the constraint is connectivity, not model type.
B. A hybrid environment still assumes some connectivity for the cloud half, so the offline case is left unsolved.
C. A cloud-based API is precisely the wrong choice here - it is unreachable whenever the device has no internet.

Question 4

For AI projects the code and systems don’t matter as much as the data. In fact, big data is what’s powering much of this latest wave of AI.

What’s most important for your company to consider around data?

Because of almost-infinite storage and compute power, collect as much data as possible and deal with organizing it later.
Collect enormous amounts of data - the more data the better.
Understanding which algorithms are best for your data needs.
Have team members that have experience, understanding of tools, and the ability to deal with massive volumes of data.

Show answer and explanation

Correct answer: D. Have team members that have experience, understanding of tools, and the ability to deal with massive volumes of data.

CPMAI is data-centric, but its point is that value comes from an organization's ability to manage data, not from the raw volume of it. Data only becomes useful when people with the right experience, tooling and understanding can collect, clean, govern and engineer it. Capability is the constraint that determines whether a data asset can be turned into a working model, which is why staffing for that capability is the most important consideration.

Why the other options are wrong

A. Collect-now-organize-later produces large, ungoverned data swamps that cost more to untangle than they return.
B. More data is not automatically better; unlabeled, irrelevant or poor-quality data adds cost without adding signal.
C. Algorithm selection is a Phase IV concern and is secondary to whether the data can be prepared at all.

Question 5

Using machine learning and other cognitive approaches to understand how to take past / existing behavior and predict future outcomes or help humans make decisions about future outcomes using insight learned from past behavior / interactions / data is a core part to which pattern(s) of AI?

Goal Driven Systems
Predictive Analytics & Decision Support and Patterns and Anomalies
Recognition Pattern
Predictive Analytics & Decision Support

Show answer and explanation

Correct answer: D. Predictive Analytics & Decision Support

The Predictive Analytics and Decision Support pattern covers exactly this description: learning from past behavior and existing data to forecast future outcomes or to help a human make a better decision. It is one of the seven patterns of AI, and it is distinguished by its forward-looking, human-supporting purpose rather than by the specific algorithm used.

Why the other options are wrong

A. Goal Driven Systems learn optimal actions by trial and reward, as in reinforcement learning and game playing, not by projecting past behavior forward.
B. Patterns and Anomalies is a separate pattern concerned with finding groupings and outliers in existing data; the description given is prediction and decision support alone.
C. The Recognition pattern identifies and classifies objects in images, audio or text - it labels what something is, not what will happen next.

Question 6

As an organization building an AI solution for your current customers based in NYC, but with possible plans for future expansion, how should you handle worldwide AI laws and regulations?

Make sure to follow relevant data, privacy, and other important laws both in the US and where you’re like to expand to in the coming year
Make sure to follow relevant data, privacy, and other important laws as it pertains to NYC
Make sure to follow relevant data, privacy, and other important laws as it pertains to the United States
You’re too small of an organization to be worried about laws at the moment

Show answer and explanation

Correct answer: A. Make sure to follow relevant data, privacy, and other important laws both in the US and where you’re like to expand to in the coming year

Regulatory exposure follows the data and the users, not the office address. A system built today will still be in operation when the expansion happens, and retrofitting privacy and AI compliance into a deployed model is far more expensive than designing for it. The defensible approach is to meet the laws that apply now in the US and to design against the requirements of the markets the organization plans to enter.

Why the other options are wrong

B. Scoping compliance to NYC alone ignores US federal and state law that already applies and guarantees rework on expansion.
C. US-only compliance covers today but not the planned expansion, which is the specific detail the question supplies.
D. Data protection and AI regulation apply regardless of company size; being small is not an exemption.

Question 7

Your team is starting a new facial recognition project and you want to ensure that the project is being done with Trustworthy AI in mind.

At what phase of CPMAI would Trustworthy AI be considered?

Phase I
Phase II
Phase III
Phase IV
Phase V
Phase VI
All phases
None of the phases

Show answer and explanation

Correct answer: G. All phases

Trustworthy AI is not a stage gate that a project passes once - it is a concern that runs through all six CPMAI phases. Bias can enter at data understanding, at data preparation, at modeling, at evaluation and at operationalization, and transparency, explainability, governance and responsibility have to be addressed at each. For a sensitive application such as facial recognition, every phase carries its own trust obligations.

Why the other options are wrong

A. Phase I raises the ethical question but cannot address bias in data or models that do not exist yet.
B. Phase II is where data sources are assessed for bias, but that is only one of several exposure points.
C. Phase III preparation can introduce or remove bias, yet trust concerns do not end there.
D. Phase IV modeling must be examined for fairness, but so must the phases before and after it.
E. Phase V evaluation tests for disparate performance, which is necessary but not sufficient on its own.
F. Phase VI monitoring catches drift and misuse in production, again only part of the picture.
H. Trustworthy AI is a core CPMAI concern, so excluding it from every phase is the opposite of the framework's guidance.

Question 8

You recently completed an image recognition project at your company that was focused on identifying different types of cars. You have now been assigned a new image recognition project that is focused on identifying different types of animals. You know you can shortcut model development by using a specific technique.

What is this technique called?

Reinforcement Learning
Generative AI
Transfer Learning
Pre-Trained Models

Show answer and explanation

Correct answer: C. Transfer Learning

Transfer learning reuses a model already trained on one task as the starting point for a related task, so the general visual features learned while classifying cars - edges, shapes, textures - carry over to classifying animals. Only the later layers need retraining on the new labeled set, which dramatically reduces the data and compute needed compared with training from scratch.

Why the other options are wrong

A. Reinforcement learning trains an agent through reward signals from an environment; it is not a way to reuse a trained image classifier.
B. Generative AI creates new content rather than shortening the training of a classifier.
D. A pre-trained model is the input to transfer learning, not the technique itself - the question asks what the technique is called.

Question 9

You’re working with an inexperienced team and this is all their first AI project. You’re trying to work on a supervised learning binary classification problem to determine if emails are spam or not.

What is the best approach for this project?

Pick a simple algorithm such a naive bayes
Pick a neural network algorithm since you know this works well for supervised learning approaches
Pick an ensemble method since you’re not sure which algorithm will perform best
Pick a simple algorithm such as Gaussian mixture

Show answer and explanation

Correct answer: A. Pick a simple algorithm such a naive bayes

With an inexperienced team on a first AI project, the right choice is the simplest algorithm that fits the problem. Naive Bayes is a supervised classifier, is a long-established baseline for spam detection, trains quickly, needs little tuning and produces a result the team can interpret and explain. CPMAI favours starting simple and iterating, because an understandable baseline tells you whether the problem is tractable at all.

Why the other options are wrong

B. A neural network adds tuning, compute and opacity that a first-time team cannot yet manage, for a problem a simple classifier already solves.
C. An ensemble compounds the complexity of several models and makes failures harder to diagnose.
D. Gaussian mixture models are an unsupervised clustering method and do not fit a supervised binary classification task.

Question 10

Your team is working on an AI system to provide a more personalized experience for customers on your website.

What should the team do in regard to determining the pattern of AI with regards to the ROI of the project?

First identify the AI pattern you want to use and then figure out the ROI
First determine the pattern of AI you want to use and then work with stakeholders to come up with ROI
First identify the objective you’re trying to solve or the ROI you desire and then use that to figure out that correct pattern
First talk to senior managers who set the ROI of the project

Show answer and explanation

Correct answer: C. First identify the objective you’re trying to solve or the ROI you desire and then use that to figure out that correct pattern

CPMAI Phase I is business-first: define the problem and the return you are trying to achieve, then let that determine which of the seven patterns of AI applies. Choosing a pattern or a technology first is technology-led thinking, and it produces projects that demonstrate a capability without delivering a business outcome. The objective drives the pattern, never the reverse.

Why the other options are wrong

A. Selecting the pattern first inverts CPMAI and risks fitting a business case to a chosen technology.
B. This still starts from the pattern; involving stakeholders afterwards does not correct the order.
D. ROI is validated with stakeholders, but a number handed down without a defined problem gives the team nothing to build against.

Question 11

A team is getting ready to begin working on a ML project. They need to build a data preparation pipeline and someone on the team suggests they reuse the same pipeline they created for their last project.

What’s wrong with this suggestion?

Pipelines are model operationalization need specific.
Pipelines are pattern and model need specific.
Pipelines are pattern needs specific so as long as it’s the same pattern then you can reuse the pipeline.
There is no issue. Pipelines can be reused as needed between projects

Show answer and explanation

Correct answer: B. Pipelines are pattern and model need specific.

A data preparation pipeline is shaped by both the pattern of AI and the specific needs of the model being built - the features required, the labels, the formats, the volumes and the transformations. Two projects rarely share all of those, so a pipeline lifted wholesale from a previous project will prepare the wrong data in the wrong shape.

Why the other options are wrong

A. Operationalization needs affect how a model is deployed, not how the training data is prepared.
C. Sharing a pattern is not sufficient; two projects on the same pattern can still need entirely different features and data sources.
D. Reuse without adaptation is precisely the problem - components may be reusable, but the pipeline as a whole is not.

Question 12

In the case that an algorithm you want to use isn’t algorithmically explainable, AI systems should try to do the following:

Provide a means to have contestability of the algorithm selected
Provide a means to interpret AI results so that cause and effect can be represented
Provide a means to have a different team on the project
Provide a means to reverse-engineer the algorithm to inspect its performance

Show answer and explanation

Correct answer: B. Provide a means to interpret AI results so that cause and effect can be represented

When an algorithm is not explainable in its own terms, the obligation shifts to interpretability: providing a means to interpret the results so that cause and effect can be represented to the people affected. Techniques that show which inputs drove an output give a defensible account of the decision even when the internal mechanism cannot be read directly.

Why the other options are wrong

A. Contestability is a valuable safeguard but it lets people challenge an outcome rather than understand it.
C. Changing the team does not make an opaque algorithm any more interpretable.
D. Reverse-engineering the algorithm inspects the mechanism and is impractical for the models that raise this problem in the first place.

Question 13

An energy technology company needs to convey data understanding to the leadership team for an AI project aimed at predicting equipment failures. The project involves analyzing sensor data from various equipment.

Which necessary initial task should the project manager take?

Use a generative adversarial network (GAN) to simulate equipment behavior.
Construct a graphical model using DALL-E to depict failure predictions.
Employ a support vector machine to classify sensor data.
Develop a preprocessing pipeline to clean and normalize sensor data.

Show answer and explanation

Correct answer: D. Develop a preprocessing pipeline to clean and normalize sensor data.

Before anything can be conveyed about the data, the sensor data has to be usable. Building a preprocessing pipeline to clean and normalize it - handling missing readings, aligning sampling rates, removing sensor noise and putting values on comparable scales - is the necessary first task, and it produces the reliable basis on which any analysis presented to leadership rests.

Why the other options are wrong

A. A GAN simulates data and is an advanced modeling step, not an initial data understanding task.
B. DALL-E generates images from text and has no role in depicting equipment failure analysis.
C. A support vector machine is a modeling technique applied after the data is prepared.

Question 14

Use cognitive technologies/AI when you can’t code the rules or you can’t scale easily with people or automation. As a good rule of thumb when deciding if AI is right for the project you should:

Decide if it’s a statistics pattern. If it’s statistical then go with the AI project.
Decide if it’s probabilistic or deterministic patterns. If it’s deterministic then go with the AI project.
See if simple rules work. If yes, then pick the right AI solution to solve the problem.
Decide if it’s probabilistic or deterministic patterns. If it’s probabilistic then go with the AI project.

Show answer and explanation

Correct answer: D. Decide if it’s probabilistic or deterministic patterns. If it’s probabilistic then go with the AI project.

AI earns its place when the problem is probabilistic - when the answer depends on patterns and likelihoods that cannot be written down as rules in advance. Deterministic problems have a known rule for every input and are better and more cheaply served by conventional code or automation. Testing whether the pattern is probabilistic is the practical rule of thumb for AI fit.

Why the other options are wrong

A. Statistics appear in almost every data problem, so this test does not separate AI-suitable work from ordinary analysis.
B. This has the test backwards - deterministic problems are the ones that should not use AI.
C. If simple rules work, the correct conclusion is to use those rules rather than to add an AI solution.

Question 15

Your team is working on a new facial recognition application. Since this technology has the potential to be mis-used you think it’s important to set guidelines for the proper use of this application and you want to make sure the AI system is built for some positive purpose.

What area of Trustworthy AI does this best fall under?

Transparent AI
Governed AI
Responsible AI
Explainable AI

Show answer and explanation

Correct answer: C. Responsible AI

Responsible AI is the dimension concerned with whether a system should be built and used at all - its purpose, its potential for misuse, and the guidelines that keep it aimed at a beneficial outcome. Setting rules for proper use of facial recognition and insisting the application serve a positive purpose is a responsibility question, not a technical disclosure one.

Why the other options are wrong

A. Transparent AI is about disclosing that AI is in use and how it operates, not about governing the purpose it serves.
B. Governed AI concerns the policies, controls and oversight structures applied to a system once it exists.
D. Explainable AI is about making individual model decisions interpretable, which is separate from whether the use itself is proper.

Question 16

You need to hire a data scientist to join your team.

What skill sets should you be looking for when hiring and interviewing this person? (Choose all that apply.)

Prompt engineering skills
Understanding of tools and technologies for manipulating, collecting, and preparing large data sets
Critical thinking skills
Understanding of algorithms
Automation skills, especially around creating RPA bots
Strong math skills, especially in calculus and statistics

Show answer and explanation

Correct answer: A, B, C, D, F. Prompt engineering skills; Understanding of tools and technologies for manipulating, collecting, and preparing large data sets; Critical thinking skills; Understanding of algorithms; Strong math skills, especially in calculus and statistics

A data scientist on a CPMAI project needs the breadth to take a business problem through data and into a model. That means the tooling skills to manipulate and prepare large data sets, an understanding of which algorithms suit which problems, the mathematical grounding in statistics and calculus that underpins them, critical thinking to frame the problem and question the results, and increasingly prompt engineering as generative models become part of the toolkit.

Why the other options are wrong

E. Building RPA bots is an automation engineering skill; RPA is rules-based process automation and sits outside the data scientist's role.

Question 17

Your team is running a forecasting project and wants to use previous user data to better predict future outcomes. However, your team doesn’t have access to all the data it needs.

What’s the best course of action?

Move ahead as planned and hope you get access to the data once you need it. Since you’re using an iterative approach you can always go back to steps as needed later on.
Cautiously move forward knowing you may need to pause mid-project which is ok.
Move ahead as planned so you stay on time with your project.
Do not move forward until you have access to all the data you need.

Show answer and explanation

Correct answer: B. Cautiously move forward knowing you may need to pause mid-project which is ok.

CPMAI is iterative and does not require every data source to be secured before work begins. The team can proceed on what it has, delivering a first iteration while access to the remaining data is pursued, provided everyone accepts that the project may have to pause at a phase boundary until that access arrives. Proceeding cautiously with that expectation set is the balanced course.

Why the other options are wrong

A. Moving ahead and hoping treats the data gap as somebody else's problem and sets no expectation that a pause may be needed.
C. Prioritizing the schedule over the data gap is how projects reach modeling with nothing to train on.
D. Waiting for complete data before starting is a waterfall stance and forfeits everything the team could learn from a first iteration.

Question 18

Creating machine learning models can be complicated. Your team wants to use tools called Automated Machine Learning (AutoML) to simplify the process. You know of another team that has used AutoML tools and it’s saved the team a lot of time.

However, what’s the one area you should not have the AutoML tool help with?

Automatic model assessment
Iterative modeling and evaluation
Automatic hyperparameter tuning
Automatic model selection
Automatic algorithm selection

Show answer and explanation

Correct answer: B. Iterative modeling and evaluation

AutoML is well suited to the mechanical parts of model building - trying candidate algorithms, tuning hyperparameters and scoring the results - because those are search problems a machine performs faster than a person. What it cannot take over is the iterative modeling and evaluation loop itself, which is where humans judge whether the model actually answers the business question defined in Phase I, whether its errors are acceptable and what to change next.

Why the other options are wrong

A. Automatic model assessment computes metrics, a mechanical task AutoML handles well.
C. Hyperparameter tuning is a systematic search and is one of AutoML's strongest use cases.
D. Automatic model selection is a standard, appropriate AutoML function.
E. Algorithm selection is exactly the kind of comparison AutoML is built to automate.

Question 19

Recently your company has been getting a large number of spam emails and some employees have been clicking on these suspicious emails causing a headache for IT. The head of IT wants to create a more robust spam filter and your team has been tasked with this project.

What type of algorithm would you select for this problem?

Clustering
Regression
Binary (or Binomial) Classification
Multiclass Classification

Show answer and explanation

Correct answer: C. Binary (or Binomial) Classification

A spam filter answers a two-way question: is this message spam or not? That makes it a binary, or binomial, classification problem, where a supervised model is trained on labeled examples of both classes and assigns each new email to one of exactly two categories.

Why the other options are wrong

A. Clustering is unsupervised and groups similar messages without knowing which group is spam, so it produces no decision.
B. Regression predicts a continuous numeric value, not a yes-or-no category.
D. Multiclass classification is for three or more categories; spam detection has only two outcomes.

Question 20

A project manager meets with a customer for initial discussions about an upcoming project. At the end of the meeting, the customer asks the project manager for a rough estimate of the project duration. Based on her experience with three similar projects, the project manager provides an estimate of 8-10 months.

What’s wrong with this timeframe?

It’s underestimating the project timeline by 3 months
It fits into a waterfall timeframe, but not an agile project timeframe
It’s not accounting for data preparation timelines
It’s not accounting for potential project delays

Show answer and explanation

Correct answer: C. It’s not accounting for data preparation timelines

An estimate drawn from three similar software projects will not include the data work, and data engineering and preparation is routinely 80 percent of an AI project's effort. Collecting, cleaning, labeling and governing data is what stretches AI timelines beyond what comparable software experience suggests, so an estimate that omits it is unsound regardless of the number attached.

Why the other options are wrong

A. Adding three months to a figure that was built on the wrong basis does not correct the basis.
B. The problem is not which delivery model is used - an agile AI project still has to prepare the data.
D. Contingency for delays is ordinary planning; the specific omission here is the data work itself.

Question 21

During the planning phase, a project manager identifies multiple patterns for their AI project.

Which two actions should the project manager do? (Choose two.)

Split the project into iterations for different patterns.
Focus on noncognitive parts before addressing cognitive patterns.
Combine all patterns into a single comprehensive iteration.
Select patterns based on stakeholder preference.
Proceed with the most complex pattern to challenge the team.

Show answer and explanation

Correct answer: A, B. Split the project into iterations for different patterns; Focus on noncognitive parts before addressing cognitive patterns.

When several patterns are in play, each brings its own data requirements, skills and evaluation criteria, so the project is split into iterations with each iteration addressing a pattern. Handling the non-cognitive parts first is the practical sequence: they are faster and more predictable to deliver, they establish the surrounding process and integration, and they give the cognitive work a working foundation to build on.

Why the other options are wrong

C. Combining every pattern into one iteration abandons the iterative structure and makes the scope unmanageable.
D. Stakeholder preference does not determine which patterns a problem actually requires.
E. Choosing the most complex pattern to challenge the team is the opposite of scoping an iteration for delivery.

Question 22

As the project manager, you are leading a brainstorming session with key stakeholders around a new Hyperpersonalization project.

What’s a key feature for this project that should happen to ensure success?

Develop a unique profile of each individual, and have that profile learn and adapt over time for a wide variety of purposes
Develop a unique profile of each individual, and manually update that profile over time for a wide variety of purposes
Develop a unique profile of each individual, and having that profile both learn and adapt over time as well as be programmed for a wide variety of purposes
Develop a unique profile of each type of individual, and have that profile stay the same over the lifetime of that user

Show answer and explanation

Correct answer: C. Develop a unique profile of each individual, and having that profile both learn and adapt over time as well as be programmed for a wide variety of purposes

Hyperpersonalization depends on a profile that is unique to each individual and that keeps changing. To be useful the profile has to learn and adapt from ongoing behaviour, and it also has to be programmable so the organization can direct it at different purposes - marketing, service, recommendations - rather than a single fixed use. Both properties together are what makes the pattern work.

Why the other options are wrong

A. Learning and adapting alone leaves no way to target the profile at specific business purposes.
B. Manual updates cannot keep pace with individual behaviour at scale, which defeats the point of the pattern.
D. A profile per type of individual is segmentation, not personalization, and freezing it means it never reflects how the user actually changes.

Question 23

During which phase of an AI project should you consider Trustworthy AI considerations?

Phase I: Business Understanding
Phase II: Data Understanding
Phase VI: Model Operationalization
Every Phase of the AI project

Show answer and explanation

Correct answer: D. Every Phase of the AI project

Trustworthy AI is a continuous obligation across every CPMAI phase, not a checkpoint. Bias, privacy exposure, transparency, explainability and accountability arise at different points - in how the problem is framed, which data is chosen, how it is prepared, how the model is built and evaluated, and how it behaves once live - so each phase carries its own trust considerations.

Why the other options are wrong

A. Phase I raises the question but cannot address risks in data and models that do not exist yet.
B. Phase II is where data bias is assessed, which is one exposure point among several.
C. Phase VI catches misuse and drift in production, far too late to be the only place trust is considered.

Question 24

An IT services company is operationalizing an AI-based customer service chatbot. The project manager needs to track the workflow of key resources such as natural language processing experts, cloud computing resources, and integration specialists.

What will enable the project manager to visualize and manage the cadence of resource activities?

Conducting a root cause analysis
Using a kanban board
Applying a critical path method
Implementing a balanced scorecard

Show answer and explanation

Correct answer: B. Using a kanban board

A kanban board visualizes work as it moves through stages and makes the cadence of activity across specialists immediately visible - what each resource is working on, what is queued and where work is piling up. That is exactly what is needed to manage the flow of NLP experts, cloud resources and integration specialists through an operationalization effort.

Why the other options are wrong

A. Root cause analysis investigates a specific failure after it has occurred.
C. Critical path method identifies the longest dependent sequence in a schedule but does not show ongoing work in progress.
D. A balanced scorecard reports strategic performance measures rather than day-to-day resource flow.

Question 25

In a clustering analysis for data use, the project team finds that the clusters are not meaningful and do not provide actionable insights.

Which activity should the project manager do with the project team?

Establish data governance protocols.
Conduct an algorithm analysis on the data sources.
Assess the trade-offs of the various algorithms.
Identify the data gaps and address deficiencies.

Show answer and explanation

Correct answer: D. Identify the data gaps and address deficiencies.

Clusters that carry no meaning usually reflect the data rather than the algorithm. If the features do not capture the dimensions that actually distinguish the records, or if the data is incomplete or unrepresentative, no clustering method will produce actionable groups. Identifying the data gaps and correcting the deficiencies addresses the cause instead of retuning around it.

Why the other options are wrong

A. Governance protocols control how data is managed and do not make the resulting clusters meaningful.
B. Analyzing algorithms against the data sources treats the method as the fault when the input is the more likely cause.
C. Weighing algorithm trade-offs is worth doing after the data is sound; done first it changes technique without changing the signal available.

25 questions in. Want the rest? The first half of the pack, free by email · The full pack

Question 26

Clean, well-labeled, datasets used for machine learning are partitioned into three subsets: Training sets, Validation sets, and Test sets. As your team is doing this, what’s the best way to split up this data?

Split by alphabetical order
Split by patterned subsampling
Split by random subsampling
Use the same data for all sets

Show answer and explanation

Correct answer: C. Split by random subsampling

Training, validation and test sets must each be representative of the same underlying distribution, and random subsampling is what achieves that. Any ordered or patterned split risks concentrating a class, a time period or a source in one subset, which makes evaluation results meaningless because the test set no longer resembles what the model will see in production.

Why the other options are wrong

A. Alphabetical order is a systematic bias - it can group related records and skew the class balance of each subset.
B. Patterned subsampling imposes structure on the split and can align with structure already in the data, reintroducing bias.
D. Reusing the same data for all three sets means the model is evaluated on what it memorized, so the scores are worthless.

Question 27

Data Engineering is 80%+ of most AI projects, so building a good Data Engineering Environment is key to AI Project Success. As the manager of this project, you need to make sure you have correct staffing needs.

What’s the most critical role to staff for in the Big Data / Data Engineering Environment?

Data Scientists
Data Engineering and Data Scientists
Senior management
Data Engineering
All roles are critical to staff in the Four different AI Tech environments

Show answer and explanation

Correct answer: D. Data Engineering

In the big data and data engineering environment, data engineering is the role that has to be staffed. Data engineering is where the bulk of an AI project's effort goes - building the ingestion, storage, transformation and pipeline infrastructure - and without it there is no prepared data for anyone else to work with.

Why the other options are wrong

A. Data scientists depend on infrastructure that data engineers build; staffing them first leaves them with nothing to work on.
B. The question asks for the most critical role within this specific environment, and that is data engineering.
C. Senior management sponsors the work but is not a staffing role in the data engineering environment.
E. This avoids the question, which asks specifically about the big data and data engineering environment.

Question 28

A financial services firm is building an AI model to detect fraudulent transactions. Identifying and validating data sources is critical to the model’s success.

What is an effective method that helps to ensure data accuracy?

Utilizing data lineage tools to track data origin and transformations
Employing a federated database system for decentralized data access
Setting up a batch processing system for data cleansing
Implementing a blockchain-based ledger for transaction data

Show answer and explanation

Correct answer: A. Utilizing data lineage tools to track data origin and transformations

Data lineage tools record where each data element originated and every transformation applied to it. That traceability is what lets the team verify that transaction data is accurate and unaltered, diagnose where an error was introduced, and satisfy the audit requirements that come with financial fraud detection.

Why the other options are wrong

B. A federated database changes how data is accessed across systems without recording its origin or transformations.
C. Batch cleansing corrects data but leaves no record of what the data was before or where it came from.
D. A blockchain ledger makes records tamper-evident once written but does not establish that they were accurate when entered.

Question 29

Your team has collected petabytes of data for your AI project. As the project lead, you understand this is too much data to use for this iteration of the project.

What is the best course of action to take with this data?

Data Deduping to reduce overall size and data complexity.
Data integration focused on reducing the number of data sources.
Data selection and attribute pruning to reduce overall size and data complexity.
Careful algorithm selection that reduces the need for data.

Show answer and explanation

Correct answer: C. Data selection and attribute pruning to reduce overall size and data complexity.

When the data volume exceeds what an iteration can handle, the right response is to reduce it deliberately: select the records that are relevant to this iteration's question and prune the attributes that carry no signal for it. That cuts both size and dimensionality while keeping the data representative, which resampling or deduplication alone will not achieve.

Why the other options are wrong

A. Deduplication removes only exact repeats and will not bring petabytes down to a workable scale.
B. Consolidating sources changes where data comes from without reducing how much of it there is.
D. Choosing an algorithm to fit the data volume inverts CPMAI - the problem determines the algorithm, not the storage bill.

Question 30

You’re working with a small inexperienced team on a new ML project. Choosing the best algorithm with the best settings given the training and test data is proving to be very hard for them. You lack the critical data science resources available on your team, and can’t wait weeks until a data science resource becomes available to join your team.

What’s your best course of action?

Outsource the project ASAP
Find a citizen data scientist to help
Use an AutoML solution
Put the project on hold until the resources needed become available

Show answer and explanation

Correct answer: C. Use an AutoML solution

AutoML exists for exactly this gap: it searches algorithms and hyperparameter settings automatically and returns a scored comparison, letting a team without data science depth produce a workable model now. It keeps the project moving without waiting on a hire or handing the work outside.

Why the other options are wrong

A. Outsourcing surrenders control and the team's own learning to solve a problem that tooling addresses.
B. A citizen data scientist works with low-code tools in their domain and would face the same algorithm and tuning difficulty.
D. Pausing the project is the outcome the question is trying to avoid, and it wastes the iteration.

Question 31

Which of the following best describes the technical definition of Machine Learning?

An approach to using increasing levels of intelligence to solve greater cognitive needs from unintelligent automation to autonomous business process.
A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E.
The application of pre-defined rules and algorithms to solve complex problems.
The use of computing technology to enable machines to gain cognitive intelligence.

Show answer and explanation

Correct answer: B. A computer program is said to learn from experience E with respect to some class of tasks T and performance measure P if its performance at tasks in T, as measured by P, improves with experience E.

The formal definition of machine learning, from Tom Mitchell, is that a program learns from experience E with respect to a class of tasks T and a performance measure P if its performance at T, measured by P, improves with E. It is the technical definition because it specifies learning operationally - measurable improvement with experience - rather than describing it loosely.

Why the other options are wrong

A. This describes a maturity progression from automation toward autonomy, not what machine learning is.
C. Applying pre-defined rules is the definition of rules-based systems, which is the opposite of learning from data.
D. This is a broad description of AI generally and does not define the learning mechanism.

Question 32

Your organization wants to keep an eye on AI systems for Governance purposes.

What are the most crucial things to consider? (Choose all that apply.)

Vendor procurement methods
AI System testing requirements
Continuous System monitoring
Data source identification
Algorithm selection
Human chain of accountability
Key Performance Indicators (KPIs)
ROI determination

Show answer and explanation

Correct answer: B, C, F, G. AI System testing requirements; Continuous System monitoring; Human chain of accountability; Key Performance Indicators (KPIs)

AI governance is about ongoing oversight of a system in operation. That means defined testing requirements before release, continuous monitoring afterwards to catch drift and misbehaviour, a documented human chain of accountability so a named person is answerable for the system's decisions, and KPIs that make performance and impact measurable over time.

Why the other options are wrong

A. Procurement method is a purchasing concern and does not govern how the AI system behaves once running.
D. Identifying data sources is a Phase II data understanding activity rather than an ongoing governance control.
E. Algorithm selection is a technical modeling decision made in Phase IV.
H. ROI determination belongs to Phase I business justification, not to operational oversight.

Question 33

You want to make sure that in your HR hiring system that applicants have the ability to contest the result. In what layer of the Trustworthy AI framework do we address this need?

Responsible AI
Ethical AI
Transparent AI
Explainable AI
Governed AI

Show answer and explanation

Correct answer: C. Transparent AI

Transparent AI covers the visibility a person has into an AI-driven decision that affects them - being told that AI was used, what it considered, and having a route to question or contest the outcome. Giving applicants the ability to challenge a hiring decision is an obligation of the transparency layer.

Why the other options are wrong

A. Responsible AI concerns whether the system should be built and used for its stated purpose.
B. Ethical AI addresses the moral principles behind the system rather than the individual's recourse against a decision.
D. Explainable AI makes the internal reasoning of a model interpretable, which supports contesting but is not the same as providing the right to contest.
E. Governed AI is the policy and oversight structure applied to the system as a whole.

Question 34

You just joined a new company and they want to start their first AI project. Senior management thinks the best approach is to just buy AI from a vendor. You know that AI is something you do, not something you buy.

What is your next best course of action to address this?

Say nothing and let the team figure it out for themselves
Share prior experiences with how your last team addressed this problem and how you solved it
Help senior management do research on AI vendors
Share prior experiences with how your last team addressed this problem and their data quality issues

Show answer and explanation

Correct answer: B. Share prior experiences with how your last team addressed this problem and how you solved it

AI is something an organization does, not something it purchases, because value depends on the organization's own data, problems and iteration. The effective way to shift that belief in senior management is evidence rather than argument: describing a comparable project you have run, how it was approached and what result it produced, gives them a concrete alternative to buying a product.

Why the other options are wrong

A. Staying silent leaves a known misconception to drive the company's first AI project.
C. Researching vendors reinforces the buy-it assumption instead of correcting it.
D. Leading with data quality problems raises an obstacle without showing the successful approach that answers management's question.

Question 35

The growth of Big Data has led to a desire to be able to do more to process and extract more value from Big Data. Simply storing data and providing analytics is no longer enough anymore to remain competitive.

To keep your organization competitive, you need to:

Make sure the technical team has deep understanding of big data and how best to extract value from big data to unleash it for competitive advantage.
Make sure senior management has deep understanding of big data and how best to extract value from big data to unleash it for competitive advantage.
Make sure all senior leadership is data literate, understands the Vs of big data, data’s connections to your specific team, and how to extract value from big data to unleash it for competitive advantage.
Make sure everyone on the team has an understanding of data, its connections to the organization, and how to extract value from big data to unleash it for competitive advantage.

Show answer and explanation

Correct answer: C. Make sure all senior leadership is data literate, understands the Vs of big data, data’s connections to your specific team, and how to extract value from big data to unleash it for competitive advantage.

Extracting competitive value from big data is a leadership problem before it is a technical one. If senior leadership is data literate - understanding the volume, velocity, variety and veracity characteristics of the organization's data, how those connect to their own function, and what value can be drawn out - then investment, priorities and governance line up behind the data strategy. Without that, technical capability alone has no mandate to act on.

Why the other options are wrong

A. Technical depth is necessary but cannot set organizational priorities or unlock investment on its own.
B. Understanding held only at the top does not translate into practice unless it reaches the functions that own the data.
D. Broad team awareness is valuable, but the question asks what keeps the organization competitive, and that requires literate senior leadership.

Question 36

You have been tasked at your organization to manage a large language model (LLM) project.

Identify what LLMs are useful for. (Choose all that apply.)

Process automation
Text summarization
Machine Translation
Classify and categorize content
Code generation
Improve search quality

Show answer and explanation

Correct answer: B, C, D, E, F. Text summarization; Machine Translation; Classify and categorize content; Code generation; Improve search quality

Large language models work on text and code. They summarize documents, translate between languages, classify and categorize content, generate and complete code, and improve search by matching on meaning rather than keywords. All of these are language tasks where the model produces or interprets text.

Why the other options are wrong

A. Process automation executes defined business steps across systems and is the domain of RPA and workflow tools; an LLM may sit inside such a process but is not itself the automation.

Question 37

The team is working to build a data preparation pipeline for the conversational chatbot project.

Which phase of CPMAI is this done?

Phase I
Phase II
Phase III
Phase IV
Phase V
Phase VI

Show answer and explanation

Correct answer: B. Phase II

The data preparation pipeline is designed in Phase II, Data Understanding, because it has to be built against the data requirements, sources and formats identified there. Phase II establishes what data exists, where it comes from and what condition it is in, and the pipeline is the mechanism defined to move it. Phase III is then where that pipeline is run and the data itself is cleaned and transformed.

Why the other options are wrong

A. Phase I is Business Understanding and settles the problem and ROI, before any data source is known.
C. Phase III executes the preparation - it is where data is cleaned and enriched, not where the pipeline requirements are established.
D. Phase IV trains models on data the pipeline has already delivered.
E. Phase V evaluates a trained model.
F. Phase VI operationalizes and monitors it.

Question 38

A financial services firm is assessing the success of a newly operationalized AI system for fraud detection. The project manager needs to evaluate the model against business key performance indicators. (KPIs).

What is an effective method to help ensure the accuracy of this evaluation?

Implementing a single comprehensive metric
Consulting with external experts and auditors
Utilizing a diverse set of validation techniques
Reviewing quarterly business financial reports

Show answer and explanation

Correct answer: C. Utilizing a diverse set of validation techniques

A single metric can be satisfied by a model that fails the business in other ways - high overall accuracy can hide poor performance on the fraud cases that matter. Using a diverse set of validation techniques, covering different metrics, data slices and time periods, gives a picture that is much harder to game and is what makes an evaluation against business KPIs trustworthy.

Why the other options are wrong

A. One comprehensive metric compresses away exactly the detail that reveals where a fraud model fails.
B. External experts and auditors add assurance but are a review process rather than a validation method.
D. Quarterly financial reports are far too slow and too coarse to attribute results to model performance.

Question 39

A team has started working on their first AI project and they are running this project like a traditional software development project. About two months into the project the team is hitting some major issues, and you’re tasked with coming in to help manage this project. Immediately you realize that AI projects need to be treated like data centric projects.

What’s the next best course of action?

Bring in data centric methodology best practices to get this project back on track
Get the existing team up to speed and make sure existing Agile approaches can support the AI effort
Hire an entirely new team making sure there is at least one data scientist on this centrteam
Hire an outside consulting firm to handle the technical aspects while you train the team yourself on data centric best practices

Show answer and explanation

Correct answer: A. Bring in data centric methodology best practices to get this project back on track

The diagnosis is that the project is being run as a software project when it is a data project, so the correction is to apply data-centric methodology practices - iterative phases driven by data understanding, preparation and evaluation rather than by feature delivery. That addresses the actual cause and can be done with the team already in place.

Why the other options are wrong

B. Agile practices are useful but were designed around software delivery; layering them on top does not supply the data-centric structure the project is missing.
C. Replacing the team discards two months of domain knowledge to solve a methodology problem that retraining fixes.
D. Outsourcing the technical work removes the team from the very practices they need to learn, and leaves the project dependent on a third party.

Question 40

In an aerospace project focused on predictive maintenance using AI, the project team is facing challenges in coordinating the AI models’ operationalization across various manufacturing sites. Strong governance and corporate guardrails are established, but each site has different computational capabilities and network latencies.

What is an effective method that helps to ensure consistent AI performance across these sites?

Using site-specific AI model tuning
Operationalizing a decentralized AI architecture
Implementing a centralized AI model repository
Utilizing cloud-based AI services uniformly

Show answer and explanation

Correct answer: A. Using site-specific AI model tuning

The sites differ in computational capability and network latency, so a single uniform deployment will run well on some and badly on others. Tuning the model for each site - adjusting model size, batching, quantization or inference frequency to match the hardware and network available - is what delivers consistent performance, and the established governance and guardrails keep those variations within approved bounds.

Why the other options are wrong

B. Decentralizing the architecture changes where processing happens without resolving the capability differences at each site.
C. A central repository standardizes distribution of models but not how they perform on unequal hardware.
D. Applying cloud services uniformly is undermined by the network latency differences that are part of the problem.

Question 41

A transportation company is preparing data for an AI model to optimize fleet management. The project team is working with large amounts of structured and unstructured data.

If the project manager avoids addressing the variety of data during preparation, what will be the result?

Decreased data processing speed
Increased data consistency
Reduced model performance
Improved model accuracy

Show answer and explanation

Correct answer: C. Reduced model performance

Structured and unstructured data have to be reconciled into a consistent representation before a model can learn from them. Skipping that means inconsistent formats, encodings and units reach training as noise, and the model learns a weaker signal. The direct result is reduced model performance.

Why the other options are wrong

A. Processing may slow, but that is an operational cost rather than the outcome for the model itself.
B. Ignoring format variety produces less consistency, not more.
D. Accuracy falls when data variety is unaddressed; it does not improve.

Question 42

When building your model, you need to make sure you’re not only checking for performance and making sure the model is giving the expected results. You also need to make sure the model is accomplishing the business objective.

At what phase of CPMAI is this most appropriate to do this?

Phase IV
Phase I
Phase II
Phase VI
Phase III
Phase V

Show answer and explanation

Correct answer: F. Phase V

Phase V, Model Evaluation, is where the model is judged against the business objective set in Phase I, not merely against technical accuracy metrics. A model can score well and still fail to deliver the intended outcome, so this phase deliberately checks both: does it perform, and does it accomplish what the business asked for?

Why the other options are wrong

A. Phase IV is Data Modeling - building and training the model, before it is evaluated against the business goal.
B. Phase I defines the business objective; it is where the target is set, not where it is measured.
C. Phase II is Data Understanding, which precedes any model existing.
D. Phase VI is Model Operationalization, which follows a successful evaluation.
E. Phase III is Data Preparation, again before a model exists to evaluate.

Question 43

A project manager needs to select an effective technique to address distinct outputs and objectives in AI/machine learning (ML) model training.

What should the project manager use?

A generic prompt template
A sequence of prompts tailored to each output
A single comprehensive prompt for all tasks
A fixed set of predefined prompts

Show answer and explanation

Correct answer: B. A sequence of prompts tailored to each output

When outputs and objectives differ, a sequence of prompts tailored to each one gives each task its own instructions, context and success criteria. This produces more reliable results than a single prompt trying to cover everything, and it lets each step be tested and refined on its own.

Why the other options are wrong

A. A generic template ignores the distinctions between the outputs that the question says are the problem.
C. One comprehensive prompt covering all tasks dilutes instructions and produces unfocused output for every one of them.
D. A fixed predefined set cannot adapt to distinct objectives as they change.

Question 44

You have an Anomaly Detection project you’re working on and you need a simple approach of clustering data into classified groups.

Which algorithm is the best choice given this situation?

K-Means Clustering
Neural Network
Decision Tree
Hidden Markov Model

Show answer and explanation

Correct answer: A. K-Means Clustering

K-means clustering is the simple, standard choice for grouping unlabeled data into classified groups, and it supports anomaly detection directly: points that sit far from every cluster centroid are the outliers. It is computationally cheap, easy to explain and needs no labeled training data.

Why the other options are wrong

B. A neural network is far heavier than a simple clustering requirement and needs substantially more data and tuning.
C. A decision tree is a supervised method requiring labeled examples, which a clustering task does not have.
D. Hidden Markov models describe sequences of hidden states over time rather than grouping data into clusters.

Question 45

You are establishing the data requirements for the project.

Which of the following tasks is the least likely to impact data requirements?

The quality of the data you collect
The makeup of your data team
The volume of the data you collect
The location/source of your data collection

Show answer and explanation

Correct answer: B. The makeup of your data team

Data requirements are determined by the problem and the data itself - how much data is needed, how good it has to be, and where it will come from. The composition of the data team affects how the requirements are met and how quickly, but it does not change what the data requirements actually are.

Why the other options are wrong

A. Quality directly sets the requirement - poor source data means more must be collected or cleaned to reach a usable standard.
C. Volume is a data requirement in its own right and drives storage, labeling and processing needs.
D. Source and location determine format, availability and which privacy regimes apply, all of which shape the requirements.

Question 46

A team is evaluating different AI models for their project. They are considering error rates and overall performance. If the team had selected a model based solely on the error rate, what would be the outcome?

A balanced performance across all metrics
An increase in stakeholder satisfaction based on performance
A better performance across the chosen domains
A potential to overlook other critical performance metrics

Show answer and explanation

Correct answer: D. A potential to overlook other critical performance metrics

Error rate is a single aggregate number and it hides how the errors are distributed. A model chosen on that basis alone may have poor precision or recall on the cases that matter most, be unfair across groups, or be too slow or too costly to run. Selecting on one metric risks overlooking other critical performance measures.

Why the other options are wrong

A. Balanced performance across metrics is what a multi-metric evaluation aims for, not what optimizing a single one produces.
B. Stakeholders judge outcomes, and a model that scores well on error rate can still fail the outcomes they care about.
C. A low error rate overall gives no assurance of performance within specific domains or segments.

Question 47

An IT services company is integrating an AI solution to automate its customer service functions. The integration team is facing resistance from the customer’s employees.

Which action should the project manager perform to manage this risk?

Conduct all-hands meetings on the benefits.
Mandate immediate transition from management.
Offer the option to join another team.
Implement a gradual phased rollout.

Show answer and explanation

Correct answer: D. Implement a gradual phased rollout.

A gradual phased rollout is the standard way to manage adoption resistance. It lets employees build confidence with a limited scope first, surfaces real problems while they are still small, and demonstrates that the system supports rather than displaces them - which addresses the cause of the resistance instead of overriding it.

Why the other options are wrong

A. All-hands presentations on benefits are one-way communication and rarely shift resistance rooted in how work will actually change.
B. Mandating the transition forces compliance while leaving the resistance intact, and usually deepens it.
C. Offering people a transfer sidesteps the adoption problem rather than resolving it.

Question 48

Your team is working on an image recognition system to help identify plants. They have collected a large amount of data but need to get this data labeled.

Which phase of CPMAI is this done?

Phase I
Phase II
Phase III
Phase IV
Phase V
Phase VI

Show answer and explanation

Correct answer: C. Phase III

Data labeling belongs to Phase III, Data Preparation, the phase that turns the data understood in Phase II into a form a model can learn from. Cleaning, transforming, augmenting and labeling the collected plant images all happen here, because a supervised image recognition model needs every training image tagged with the plant it shows before modeling can begin.

Why the other options are wrong

A. Phase I is Business Understanding and concerns the problem and its ROI, not the preparation of data.
B. Phase II, Data Understanding, establishes what data exists and whether it is sufficient and of good quality, but the labeling itself is prepared in the next phase.
D. Phase IV trains models on data that is already labeled and prepared.
E. Phase V evaluates a trained model.
F. Phase VI operationalizes and monitors the model in production.

Question 49

You’re looking to take an image and have a Generative AI solution generate additional content beyond the bounds of the current image size.

What Generative AI approach can you use?

Prompt engineering for new image generation
Use of inpainting to replace image components
Use of super-resolution to enhance the existing image
Use of Generative Outpainting

Show answer and explanation

Correct answer: D. Use of Generative Outpainting

Outpainting is the generative technique that extends an image beyond its original borders, generating new content that continues the scene outward while remaining consistent with what is already there. It is the direct answer to producing content past the current image size.

Why the other options are wrong

A. Prompt engineering for new image generation creates a fresh image rather than extending the existing one.
B. Inpainting fills or replaces regions inside the existing frame; it does not enlarge it.
C. Super-resolution increases pixel density within the same boundaries, adding detail rather than area.

Question 50

A team is retraining a model and creating a new version of that model.

What’s the most important thing for the team to have in place before doing this?

Model operations
Data operations
Model discovery
Model Governance

Show answer and explanation

Correct answer: D. Model Governance

Model governance is what must exist before a retrain produces a new version. It defines who may approve a new model, how versions are recorded and retained, what testing a version must pass before release, and who is accountable for the decisions it makes. Without it, retraining silently replaces a model with no approval trail and no way back to the previous version.

Why the other options are wrong

A. Model operations run the deployment mechanics but do not set the approval and accountability rules that govern a version change.
B. Data operations manage the data pipeline feeding the retrain rather than the control over the model version itself.
C. Model discovery concerns finding and cataloguing available models, which is not the control needed before releasing a new version.

50 questions in. Want the rest? The first half of the pack, free by email · The full pack

Question 51

Your model has been working fine for the last three months, however recently you notice the model’s performance has greatly declined.

What seems to have been overlooked in your workflow pipeline?

Model retraining
Model Operationalization
Model Drift
Model reevaluation

Show answer and explanation

Correct answer: A. Model retraining

A model that performed well for three months and then declined has been overtaken by drift - the live data no longer matches what it was trained on. The pipeline element that answers this is model retraining, refreshing the model on recent data on a schedule or on a monitoring trigger. Its absence is why performance was allowed to decay unchecked.

Why the other options are wrong

B. The model was operationalized successfully; it ran in production for three months.
C. Model drift is the symptom being observed, not the missing pipeline step that would have prevented it.
D. Re-evaluation would measure the decline but would not correct it - retraining is what restores performance.

Question 52

An inexperienced team is training a neural network model on a desktop computer which is taking a significant amount of time.

Which recommended approach will increase the speed of the model training time?

Utilize a contractor to provide the model training portion.
Divide the dataset into multiple smaller datasets and train the model on each dataset over a desktop computer.
Train the model on graphics processing units (GPUs).
Train the model over multiple desktop computers.

Show answer and explanation

Correct answer: C. Train the model on graphics processing units (GPUs).

GPUs execute the parallel matrix operations that dominate neural network training far faster than a desktop CPU, which makes moving training onto them the single largest speed-up available. It requires no change to the data or the model design, so an inexperienced team can adopt it without redesigning their approach.

Why the other options are wrong

A. A contractor runs the same training on the same hardware; the constraint is compute, not staffing.
B. Training separately on smaller subsets produces several weaker models rather than one model trained faster.
D. Distributing across multiple desktops adds coordination and network overhead for a fraction of the gain a GPU delivers.

Question 53

In order for Supervised Learning approaches to work, they must be fed clean, well-labeled data that the system can use to learn from examples.

But how do you get Labeled Data?

As a team leader at a small startup, what approach would not be beneficial when trying to gather labeled data?

Contract with Third Party Data Labeling Firms
Hire a Contractor Workforce
Get your Users to Do it
Find a source of already labeled data

Show answer and explanation

Correct answer: C. Get your Users to Do it

Getting users to label data works only where there is a large, engaged user base whose labeling is a natural by-product of using the product. A small startup has neither the volume of users nor the traffic to produce a usable labeled set that way, so it is the one option here that would not help.

Why the other options are wrong

A. Third-party labeling firms are a standard route and are accessible to small companies at modest volume.
B. A contractor workforce gives direct control over labeling quality and scales to the size the startup needs.
D. Finding data that is already labeled is the cheapest option of all when a suitable source exists.

Question 54

You have been tasked with creating a model that will recommend products based on what other customers have similarly purchased.

Which algorithm is the best choice given this situation?

K Nearest Neighbor
K-means
Neural Network
Hyperpersonalization

Show answer and explanation

Correct answer: A. K Nearest Neighbor

Recommending products based on what similar customers bought is a nearest-neighbor problem: find the customers closest to this one in behaviour, then surface what they purchased. K-nearest neighbor does exactly that, using labeled purchase history to score similarity and produce a recommendation.

Why the other options are wrong

B. K-means groups customers into clusters without producing a per-customer recommendation, and the similar name makes it a common trap.
C. A neural network could be trained for this but is heavier than the problem requires and is not the best choice given a direct similarity method.
D. Hyperpersonalization is a pattern of AI, not an algorithm.

Question 55

An organization is to undertake a multi-pattern AI project. They want to build a robot that is able to roam the halls as well as converse with employees and answer basic questions.

What is the best approach for handling this project?

Run each pattern as its own project, with their own CPMAI phase iterations, data requirements, and project needs
Run each pattern in isolation, with separate teams
Run it as a hybrid approach and some phases are run separately while other phases are combined together
Run it as one project, combining teams, data requirements, and project needs

Show answer and explanation

Correct answer: C. Run it as a hybrid approach and some phases are run separately while other phases are combined together

A robot that both navigates and converses spans more than one pattern of AI, and the patterns have genuinely different data requirements and skills. A hybrid approach works best: phases where the patterns diverge, such as data understanding and preparation, run separately, while phases where they must come together, such as business understanding and final operationalization, are combined so the product behaves as one system.

Why the other options are wrong

A. Fully independent projects lose the shared business case and risk two components that never integrate.
B. Isolated teams guarantee an integration failure at the end, since neither builds toward a single robot.
D. Combining everything ignores that each pattern needs its own data, its own preparation and its own evaluation.

Question 56

A healthcare project manager is evaluating whether to implement an AI-powered diagnostic tool. The initial cost is US$500,000 with an expected return on investment (ROI) of 15% within the first year. The project needs to satisfy multiple stakeholders including hospital administrators and medical staff.

Which method will maximize a positive ROI for the AI implementation?

Monitoring AI model performance against key performance indicators.
Ensuring all AI and non-AI components are integrated seamlessly.
Acquiring alternatives to the AI solution as a contingency plan.
Seeking verbal commitments from interested parties at each project phase.

Show answer and explanation

Correct answer: A. Monitoring AI model performance against key performance indicators.

An ROI target is only realized if performance is tracked against it. Monitoring the model against key performance indicators shows whether the expected return is materializing, catches degradation before it erodes the benefit, and gives administrators and clinical staff the evidence they need to keep backing the system.

Why the other options are wrong

B. Seamless integration is necessary for the system to work but does not by itself demonstrate or protect the return.
C. Buying contingency alternatives adds cost and directly reduces the ROI being protected.
D. Verbal commitments are not measurable and cannot substitute for tracking actual performance.

Question 57

A company’s leadership team has requested insights into the AI model’s ability to support decision-making processes without requiring them to understand complex technical details.

Which step should the project manager take?

Discuss how ensemble methods improve the model's robustness.
Explain the role of neural network architectures in prediction accuracy.
Demonstrate how the model's output can be integrated and used in end-user systems.
Describe the model’s backpropagation and gradient descent optimization.

Show answer and explanation

Correct answer: C. Demonstrate how the model's output can be integrated and used in end-user systems.

Leadership asked about decision support, not architecture. Demonstrating how the model's output is integrated into the systems people already use shows exactly what changes in practice and what value it adds, in terms that need no technical background. That answers the question actually asked.

Why the other options are wrong

A. Ensemble methods are an internal modeling detail of the kind leadership explicitly did not ask for.
B. Neural network architecture is technical depth that obscures rather than answers the decision-support question.
D. Backpropagation and gradient descent are training mechanics and are furthest of all from what leadership needs to know.

Question 58

Your team is testing the NLP model they just created to make sure it’s performing as expected. Some of your team members want to move this model to production and move to the next iteration.

What’s wrong with this workflow?

You need to make sure the AI Go/No Go questions have been addressed
Nothing is wrong with this workflow. You can move to the next iteration
Team members should not be able to move to new projects until senior management signs off
Model Evaluation requires continuous model evaluation, retraining, and operationalization

Show answer and explanation

Correct answer: D. Model Evaluation requires continuous model evaluation, retraining, and operationalization

A model that tests well is not finished. CPMAI Phase V requires continuous evaluation, retraining and operationalization, because performance decays as real-world data drifts away from the training distribution. Treating the test result as a finish line and moving on leaves the deployed model unmonitored, with no plan for the retraining it will need.

Why the other options are wrong

A. The AI Go/No Go questions belong to Phase I, before the project begins; the gap here is at the evaluation stage.
B. There is a real problem: the workflow ends evaluation instead of making it continuous.
C. Management sign-off is an organizational policy, not the methodological failure the question describes.

Question 59

During CPMAI Phase II it’s important to not only understand the sources of your data, but also what data is required for training as well as identifying the features that are required.

When looking to gather data, what approach is best when determining how much data you need?

The “goldilocks” approach
The “less is better” approach
The “more is better” approach
There is no correct approach

Show answer and explanation

Correct answer: A. The “goldilocks” approach

The goldilocks approach means gathering the amount of data that is right for the specific problem - enough to represent the pattern the model must learn, without collecting volumes the team cannot clean, label, store or govern. CPMAI ties data volume to the requirements identified in Phase II rather than to an absolute rule.

Why the other options are wrong

B. Deliberately minimizing data risks a training set too small or too narrow to represent the problem.
C. More is better ignores the cost of preparing and governing data, and adds noise as readily as signal.
D. There is a correct approach - matching data volume to the identified requirements.

Question 60

A project manager is reviewing the performance of an AI model used for predictive analytics in sales. The model’s accuracy is within acceptable limits; however, its precision is low.

What is the cause for the precision issue?

The feature selection process is flawed.
The model is overfitting the training data.
The training data is unbalanced.
The model is underfitting the validation data.

Show answer and explanation

Correct answer: C. The training data is unbalanced.

Precision falls while overall accuracy holds when the classes are unbalanced. If the positive cases are a small minority, a model can score well on accuracy simply by being right about the dominant class, while the positive predictions it does make are frequently wrong - which is exactly what low precision measures.

Why the other options are wrong

A. Poor feature selection normally degrades accuracy as well, rather than leaving it within acceptable limits.
B. Overfitting typically shows as a gap between training and validation performance across metrics, not as precision alone.
D. Underfitting produces weak performance on every metric, including the accuracy that is stated to be acceptable.

Question 61

A government agency plans to implement a new AI-driven solution for automating risk analysis. The project team needs to ensure that all stakeholders accept the solution and the project scope is well-defined. They must identify whether the AI approach is the best solution compared to traditional methods.

Which method meets this objective?

Utilizing a hybrid approach combining cognitive and noncognitive parts to satisfy all parties
Performing a comprehensive AI go/no-go assessment focusing on technology and data factors
Developing a prototype using generative adversarial networks (GANs)
Conducting a detailed analysis to evaluate other potential AI solutions

Show answer and explanation

Correct answer: B. Performing a comprehensive AI go/no-go assessment focusing on technology and data factors

The AI Go/No Go assessment is the structured method for deciding whether AI is the right answer at all. Working through the technology and data feasibility questions establishes whether the problem needs cognitive capability, whether the data exists to support it and whether a traditional approach would serve better - and running it with stakeholders produces both the scope definition and the acceptance the team needs.

Why the other options are wrong

A. A hybrid approach is a design decision that presumes the AI question has already been answered.
C. Building a GAN prototype commits engineering effort before establishing that AI is warranted.
D. Comparing AI solutions to each other assumes AI is the right category, which is the question being asked.

Question 62

A development team is tasked with creating an AI system to assist physicians with diagnosing medical conditions. They encountered cases where symptoms do not always lead to well-defined diagnoses.

Which approach should the project manager integrate to handle the inherent uncertainty?

Enhance the knowledge base with more detailed rules.
Implement a more complex retrained model.
Increase the number of input variables.
Keep a human in the loop with all decision-making.

Show answer and explanation

Correct answer: D. Keep a human in the loop with all decision-making.

Where symptoms do not map to well-defined diagnoses, the uncertainty is inherent in the domain and no amount of model work removes it. Keeping a human in the loop for all decision-making means the clinician holds the diagnosis while the AI supplies supporting evidence. In a high-stakes medical setting that is both the safe design and the accountable one.

Why the other options are wrong

A. More detailed rules cannot capture cases that genuinely have no well-defined answer.
B. A more complex model produces more confident output without making an ambiguous case any less ambiguous.
C. Adding input variables increases noise and dimensionality without resolving the underlying uncertainty.

Question 63

Your team is trying to determine which pattern best fits their AI problem. To do this the project team is running through the seven patterns of AI to figure out what pattern best applies to their problem.

Which of the following is the best approach?

When in doubt, go with the patterns and anomalies pattern as all AI projects are about pattern matching.
Determine what you’re trying to accomplish and see which pattern(s) of AI fit best.
Apply every pattern to the project.
When in doubt, don’t apply any pattern of AI

Show answer and explanation

Correct answer: B. Determine what you’re trying to accomplish and see which pattern(s) of AI fit best.

The pattern follows from the objective. Establishing what the team is trying to accomplish and then matching that to the pattern or patterns of AI that fit is the CPMAI approach, because the pattern is what then determines the data requirements, the skills needed and the algorithms that are appropriate.

Why the other options are wrong

A. Defaulting to one pattern when uncertain guarantees the wrong data and the wrong skills for most problems.
C. Applying all seven patterns spreads effort across approaches that do not fit and delivers none of them.
D. Declining to identify a pattern leaves the project with no basis for planning data or staffing.

Question 64

You’re testing your model and it is overly sensitive to the fluctuations of data and having trouble generalizing.

What type of problem is this?

You are underfitting the data
You are overfitting the data
You have selected the wrong algorithm
You have selected the wrong data

Show answer and explanation

Correct answer: B. You are overfitting the data

Overfitting is when a model learns the noise and idiosyncrasies of the training data rather than the underlying pattern. It shows exactly as described: high sensitivity to small fluctuations in the data and poor generalization to examples it has not seen. The usual remedies are more or more varied training data, fewer features, regularization, or a simpler model.

Why the other options are wrong

A. Underfitting is the opposite - the model is too simple to capture the pattern and performs poorly on training data as well.
C. A wrong algorithm can contribute, but the symptom described names a specific, well-defined condition.
D. Wrong data would produce irrelevant predictions rather than oversensitivity to fluctuation.

Question 65

A logistics company is planning to separate cognitive and noncognitive parts of an AI project. They need to manage teams working on these parts.

What is a necessary initial task?

Implementing a customer-driven feedback loop in planning
Developing a project charter outlining responsibilities for each team
Utilizing a project management software to monitor progress
Conducting a skills assessment to assign team members accordingly

Show answer and explanation

Correct answer: D. Conducting a skills assessment to assign team members accordingly

Splitting cognitive from non-cognitive work means the two streams need genuinely different skills - data science and modeling on one side, process automation and integration on the other. A skills assessment establishes who is capable of what, and that has to happen before people can be assigned to either team or before responsibilities can be documented.

Why the other options are wrong

A. A customer feedback loop is valuable for the product but does not resolve how to staff the two streams.
B. A project charter records responsibilities, but it cannot allocate people until their skills are known.
C. Project management software tracks work that has already been assigned.

Question 66

Your team is working on a new loan decision model that takes a number of factors and data points into consideration and then automatically approves or denies a loan. After a month in operation someone does a review and notices that the system is denying a large number of loans from a certain demographic when all other factors from people in other regions (such as age, salary, and credit score) are the same.

What is most likely happening here?

Data privacy issues leading to data sharing concerns
Biased data sets leading to algorithmic discrimination
Generative AI models hallucinating data results
Nothing is wrong, algorithmic decisions will never be 100%

Show answer and explanation

Correct answer: B. Biased data sets leading to algorithmic discrimination

When a model produces systematically different outcomes for one demographic while every other factor is equivalent, the cause is almost always bias in the training data, which the algorithm then reproduces and applies at scale. This is algorithmic discrimination: the model has learned a historical pattern of unequal treatment and is now enforcing it consistently.

Why the other options are wrong

A. Privacy and data sharing concern who may access data, not whether decisions are made unfairly.
C. Hallucination is a failure mode of generative models producing invented content; this is a classification model producing consistent, measurable bias.
D. A systematic pattern against one demographic is not ordinary model error, and treating it as acceptable is a legal and ethical exposure.

Question 67

You’re running an AI project and want to speed up training of the model so that you can complete your current CPMAI iteration within the two-week timeframe the team has set.

What’s one approach to speed up model training?

Use data from a previous project
Use cloud-based technology
Use Brute force method
Use or extend a pre-trained model

Show answer and explanation

Correct answer: D. Use or extend a pre-trained model

Using or extending a pre-trained model is the most effective way to shorten training. The model already carries learned representations from a large corpus, so only the final layers need to be trained on the project's own data. That cuts training time from what would be a full run to a fraction of it, which is what fitting inside a two-week iteration requires.

Why the other options are wrong

A. Data from a previous project may not be relevant and does nothing to reduce the cost of training on it.
B. Cloud compute can shorten wall-clock time by scaling hardware, but it does not reduce the amount of training the model needs.
C. Brute force increases the work done rather than reducing it.

Question 68

During the transition to an AI solution, the project manager discovers that certain tasks may not require cognitive Al capabilities and can be handled through traditional automation methods. As a result, the project team starts segregating tasks based on their cognitive requirements.

What should the team consider?

Utilizing traditional automation solutions
Assessing traditional task complexity
Applying AI capabilities for noncognitive tasks
Proceeding with intelligent functionalities

Show answer and explanation

Correct answer: A. Utilizing traditional automation solutions

Once tasks are separated by cognitive requirement, the non-cognitive ones should be handled with traditional automation. Those tasks follow known rules, need no learning and are cheaper, faster and more reliable to automate conventionally, which leaves AI effort concentrated where genuine cognitive capability is needed.

Why the other options are wrong

B. Assessing complexity is useful but the question asks what to do with the tasks once they are already segregated.
C. Applying AI to non-cognitive tasks is the overkill the segregation exercise exists to prevent.
D. Proceeding with intelligent functionality across the board ignores the separation the team has just made.

Question 69

You have just joined a team and they are working on a new project. The project lead isn’t sure what type of technology should be used on this project - AI or a traditional software development approach.

What is the best way to determine if you have the criteria for a good AI / ML Project?

Evaluate whether the solution can be done with automation.
Determine if the project fits within the scope, budget, and timeline set out.
Determine whether the project has a cognitive technology component and meets a short-term need.
Determine the long-term need for the organization and build the project to that long-term goal.

Show answer and explanation

Correct answer: C. Determine whether the project has a cognitive technology component and meets a short-term need.

A good AI project needs a genuine cognitive component - something requiring perception, prediction, language or judgement rather than fixed rules - and it needs to address a need that can be met in a short iteration. CPMAI insists on short-cycle value delivery, so a project that is both cognitively appropriate and near-term is the one that fits.

Why the other options are wrong

A. Establishing that automation can do it argues against AI rather than testing whether the project qualifies as AI.
B. Scope, budget and timeline are project management constraints that apply to any project and say nothing about AI suitability.
D. Building to a long-term goal is the opposite of CPMAI's short iterative cycles and defers all value.

Question 70

A healthcare provider had physicians review a potential diagnostic AI application. During their final review, the project team along with the physicians, discovered that the AI model exhibits a higher than acceptable false-positive rate.

Before making the go/no-go AI decision, which next step should be performed by the team?

Reevaluate the business objectives and outcomes.
Focus on the model’s ethical implications.
Adjust the hyperparameters for better generalization.
Increase the training data volume.

Show answer and explanation

Correct answer: A. Reevaluate the business objectives and outcomes.

An unacceptable false-positive rate in a diagnostic setting is a business and clinical question before it is a technical one. The team must return to the business objectives and outcomes to establish what rate is genuinely tolerable, what the cost of each false positive is to a patient and the provider, and whether the project still meets its intended purpose. That determination is what the go/no-go decision rests on.

Why the other options are wrong

B. Ethical implications matter but follow from the objectives; the immediate need is to establish what the model must achieve.
C. Hyperparameter adjustment is a technical remedy applied once it is known what target the model has to hit.
D. More training data may help, but adding it before revisiting the objectives is work with no defined target.

Question 71

You want to create a model to figure out if a customer would be likely to repurchase a certain item. The project owner doesn’t want you to create anything too complicated, and you have a limited data set to work with.

Which algorithm is the best choice given these constraints?

Ensemble models
Naive Bayes
Neural Networks
Generative AI

Show answer and explanation

Correct answer: B. Naive Bayes

Naive Bayes is the right fit against both constraints: it performs well on limited data because it makes strong independence assumptions rather than needing many examples to learn feature interactions, and it is simple to implement, fast to train and easy to explain to a project owner who wants nothing complicated.

Why the other options are wrong

A. Ensembles combine several models and add exactly the complexity the project owner ruled out.
C. Neural networks need substantially more data than is available and are the most complex option on the list.
D. Generative AI creates new content and does not address a binary repurchase prediction.

Question 72

Upper management is looking to roll out a new product and wants to see if there are any patterns and insights that can be discovered from customer data. Your team has been tasked to discover these potential patterns and structures within this data.

Which type of machine learning approach would be most appropriate to pick for this problem?

All would work equally well
Unsupervised Learning
Supervised Learning
Reinforcement Learning

Show answer and explanation

Correct answer: B. Unsupervised Learning

Discovering patterns and structures that nobody has specified in advance is the defining use of unsupervised learning. There are no labels and no known outcome to predict - the task is to let clustering and association methods surface groupings and relationships in the customer data that the business has not yet identified.

Why the other options are wrong

A. The approaches are not interchangeable; supervised methods cannot run without labels and reinforcement learning needs an environment and reward.
C. Supervised learning requires labeled examples of the answer, which do not exist when the patterns are still unknown.
D. Reinforcement learning optimizes actions through reward feedback and does not apply to pattern discovery in a static data set.

Question 73

A telecommunications company is preparing its data for a new AI-driven customer service chatbot. The project manager needs to ensure the data is precisely formatted and cleaned before moving on to the model-building phase.

Which principle does this represent?

Defining cognitive pattern recognition
Conducting AI readiness assessments
Implementing phase-specific data augmentation
Performing data preprocessing tasks

Show answer and explanation

Correct answer: D. Performing data preprocessing tasks

Formatting and cleaning data so it is fit for training is data preprocessing. It covers the standardization, cleansing, encoding and transformation work that has to be complete before model building begins, and it is the defining activity of the data preparation phase.

Why the other options are wrong

A. Cognitive pattern recognition concerns which pattern of AI the problem falls under, decided far earlier.
B. An AI readiness assessment judges whether the organization can undertake the project at all.
C. Data augmentation generates additional examples to expand a data set rather than cleaning the data already held.

Question 74

An energy technology firm is developing an AI model to predict energy consumption patterns. They need to gather the required data from smart meters, weather forecasts, and historical usage records.

To help ensure the AI model performs well, which action will address their needs?

Utilizing a data lake to store and process all data sources
Adopting a hybrid cloud approach for scalable data management
Leveraging federated learning to keep data decentralized
Integrating a data warehouse with real-time data ingestion

Show answer and explanation

Correct answer: A. Utilizing a data lake to store and process all data sources

Smart meter readings, weather forecasts and historical usage records arrive in different formats, structures and update frequencies. A data lake stores all of them in their native form and allows processing across them without forcing a schema up front, which is what makes it the right foundation for a model drawing on several dissimilar sources.

Why the other options are wrong

B. A hybrid cloud approach addresses where infrastructure runs, not how varied data sources are brought together.
C. Federated learning keeps data distributed for privacy reasons, which is not the constraint here and complicates combining the sources.
D. A data warehouse requires a defined schema and is a poor fit for the semi-structured and unstructured feeds involved.

That was 74 of 295.

The full PMI CPMAI pack has all 295 questions, each with the answer, the explanation and why the other options are wrong, plus a questions-only copy for timed runs. US$39, paid once, with free monthly updates and a pass-or-your-money-back guarantee.

Get the full pack

Not ready yet? Get the first 115 free as one PDF by email. Get the Half Pack →