Identify the business value of generative AI solutions — AB-731 Practice Quiz

Question 1

A retail company is deploying a customer-facing generative AI chatbot that can access an internal product database to answer questions. During a security review, the team discovers that malicious users could craft inputs designed to override the system's instructions and extract confidential pricing rules. As the AI Transformation Leader, which security consideration should you prioritize to address this specific threat?

Accepted Answer

Implementing application-layer input validation and prompt injection defenses to prevent malicious instructions from manipulating the model

Answer

Encrypting the product database at rest to ensure the underlying data cannot be read directly

Answer

Requiring multi-factor authentication for all internal employees who administer the system

Answer

Fine-tuning the model on a larger dataset so its responses become more accurate

Question 2

A marketing operations team spends most of each week manually drafting first-version product descriptions for thousands of new catalog items. The content follows predictable patterns, volume is high, and human editors always review the output before publishing. Leadership asks where generative AI would deliver the clearest business value. Which use of generative AI best fits this situation?

Accepted Answer

Automating the generation of high-volume first-draft product descriptions so human editors can focus on review and refinement

Answer

Replacing the human editors entirely so descriptions are published directly from model output without review

Answer

Using generative AI to calculate precise inventory reorder quantities for each catalog item

Answer

Deploying generative AI to make final legal compliance determinations on each published description

Question 3

A retail company plans to deploy a generative AI assistant that helps hiring managers draft candidate evaluation summaries based on interview notes. During a pilot, leaders notice the AI-generated summaries consistently use more favorable language for candidates from certain backgrounds and less favorable language for others, even when the interview notes are comparable. As the AI Transformation Leader, how should you characterize this issue when briefing executives on its business impact?

Accepted Answer

It is a bias challenge inherited from the model's training data that can create legal, ethical, and reputational risk and must be mitigated before broad deployment.

Answer

It is a token cost inefficiency that can be resolved by shortening the prompts sent to the model.

Answer

It is a grounding gap that will disappear automatically once retrieval-augmented generation is added to the solution.

Answer

It is an expected fabrication behavior that is acceptable because generative models always invent some details.

Question 4

A legal services firm receives thousands of lengthy contracts each month. Attorneys spend hours reading each document to extract key clauses and produce a concise plain-language summary for clients. Leadership wants a solution that reduces this manual effort while producing readable summaries in natural prose. Which capability of generative AI makes it the most appropriate choice for this specific business need?

Accepted Answer

Its ability to comprehend unstructured text and generate coherent, human-readable summaries in natural language

Answer

Its ability to classify structured numerical data into predefined risk categories

Answer

Its ability to forecast future contract volumes based on historical time-series trends

Answer

Its ability to detect statistical anomalies in transaction records for fraud alerts

Question 5

A retail company is planning a customer-facing chatbot that will handle a very high volume of simple, repetitive product-availability questions. During budget planning, the AI leader notices that using the largest, most capable model for every request would significantly increase monthly costs. Most queries are straightforward and do not require advanced reasoning. Which approach best controls cost while still meeting the business need?

Accepted Answer

Route the simple, high-volume queries to a smaller, less expensive model and reserve the largest model only for complex requests.

Answer

Use the largest available model for all queries to guarantee the highest possible response quality.

Answer

Fine-tune the largest model on the entire product catalog so it can answer any question type.

Answer

Increase the temperature setting so the model generates shorter, cheaper responses.

Question 6

A retail company runs a generative AI chatbot that produces detailed product descriptions. Monthly costs have risen sharply, and finance asks the AI Transformation Leader to identify the primary factor increasing spend. The team confirms request volume is stable, but each response has grown much longer over recent releases. Which change would most directly reduce the model's per-response cost?

Accepted Answer

Instruct the model to generate more concise responses, reducing the number of output tokens per request

Answer

Fine-tune the model on the company's product catalog to improve description accuracy

Answer

Add a retrieval-augmented generation layer to ground responses in the latest catalog data

Answer

Enable multi-factor authentication for all users accessing the chatbot

Question 7

A retail company wants to launch a generative AI chatbot to help customers with product questions across its full catalog. During a pilot, leadership notices the chatbot gives strong answers about the company's electronics products but weak, vague answers about its clothing and home-goods lines. An investigation reveals the knowledge base used to ground the chatbot was assembled mostly from electronics support documentation. As the AI Transformation Leader, what is the MOST accurate explanation of the business risk this reveals?

Accepted Answer

The grounding data is not representative of the full range of products the solution is expected to serve, so the business value will be uneven across customer segments.

Answer

The model is too small and must be replaced with a larger pretrained foundation model to handle all product categories.

Answer

The chatbot's temperature setting is too high, causing inconsistent answers for non-electronics products.

Answer

The solution requires fine-tuning on brand voice to improve answers for clothing and home-goods products.

Question 8

A retail company builds a generative AI assistant that summarizes product reviews for shoppers. During testing, leaders notice the summaries frequently contain inconsistent product details and contradictory sentiment. On investigation, the source review database is found to contain many duplicate entries, unlabeled spam reviews, and reviews mistakenly attached to the wrong products. Before investing further in prompt engineering or model changes, what should the transformation leader prioritize to improve output reliability?

Accepted Answer

Improving the quality of the underlying review data by removing duplicates, filtering spam, and correcting product associations

Answer

Increasing the maximum token limit so the model can process more reviews at once

Answer

Switching to a larger, more capable pretrained foundation model

Answer

Adding stricter authentication controls to the assistant's API endpoint

Question 9

A retail company plans to build a generative AI assistant that drafts personalized marketing emails. The project lead argues that simply collecting the largest possible volume of historical email data will guarantee the best-performing solution. As the AI Transformation Leader, what is the most important consideration you should raise about the data used for this initiative?

Accepted Answer

The data must be high quality and representative of the diverse customer segments and scenarios the assistant will serve, not merely large in volume.

Answer

The data volume alone is the decisive factor, so the team should prioritize acquiring as many historical emails as possible regardless of source.

Answer

The data should be limited exclusively to the highest-performing past campaigns so the model only learns from successful examples.

Answer

The data type is irrelevant because generative models can produce quality output from any unstructured text input.

Question 10

A pharmaceutical company wants to use a generative AI assistant to help researchers summarize internal drug-trial documents. These documents contain confidential intellectual property and personal patient data. Leadership asks the AI Transformation Leader which security consideration is MOST critical when selecting and configuring the solution.

Accepted Answer

Ensuring the model provider will not use the company's prompts and documents to train or improve its foundation models, and that data stays within the company's compliance boundary

Answer

Choosing the model with the largest context window so entire documents fit in a single prompt

Answer

Selecting a pretrained model rather than a fine-tuned model to reduce deployment time

Answer

Applying prompt engineering techniques to make summaries more concise and readable

Question 11

A retail company wants to build a generative AI assistant that helps merchandisers analyze both written product descriptions and photographs of store shelf displays, then generates recommendations for improving product placement. As the AI Transformation Leader, you are evaluating what kind of data the underlying model must be able to process to meet this requirement. Which characteristic of the data most directly determines whether a single generative AI model can support this scenario?

Accepted Answer

The solution requires a model that supports multimodal input, because it must process both text and image data types together.

Answer

The solution requires a model trained only on structured numerical sales data, because placement recommendations are quantitative.

Answer

The solution requires the largest available context window, because merchandisers will paste long descriptions.

Answer

The solution requires a fine-tuned model, because pretrained models cannot combine two inputs in one response.

Question 12

A legal services firm wants to deploy a generative AI assistant that drafts summaries of case law for attorneys. During a pilot, the assistant occasionally produces summaries citing court cases and rulings that do not actually exist, presented in a confident and authoritative tone. As the AI Transformation Leader, how should you characterize this behavior and its primary business implication?

Accepted Answer

This is a fabrication (hallucination), an inherent reliability risk of generative models that requires human review and grounding to mitigate before high-stakes use.

Answer

This is model bias caused by an unrepresentative training dataset, and it should be resolved by rebalancing the demographic distribution of the data.

Answer

This is a data security breach, because the model is exposing confidential case records that were included in the training data.

Answer

This is an authentication failure, indicating that unauthorized users are injecting false content into the model's responses.

Question 13

A healthcare startup deploys a generative AI assistant that answers patient questions about medications. During testing, the assistant occasionally invents dosage instructions and cites clinical studies that do not exist, even though its outputs sound authoritative and fluent. Leadership asks you to name the specific generative AI challenge this behavior represents so the right mitigation can be prioritized. Which challenge is being described?

Accepted Answer

Fabrication (hallucination), where the model generates plausible-sounding but factually incorrect or invented content

Answer

Bias, where the model produces outputs that unfairly favor or disadvantage certain groups

Answer

Token cost overrun, where longer prompts and responses increase operating expenses

Answer

Latency, where the model takes too long to return a response to the user

Question 14

A marketing agency wants a generative AI model that consistently writes product descriptions in their client's highly distinctive brand voice and tone. The desired style cannot be reliably achieved through prompt instructions alone, and the agency has a large curated dataset of thousands of approved past descriptions written in that exact style. Which approach best meets this requirement?

Accepted Answer

Fine-tune a pretrained model using the curated dataset of approved descriptions

Answer

Use retrieval-augmented generation (RAG) to pull past descriptions at query time

Answer

Rely solely on longer, more detailed prompts describing the brand voice

Answer

Train a brand-new foundation model from scratch on the agency's data

Question 15

A pharmaceutical company wants a generative AI assistant that reliably uses highly specialized medical and regulatory terminology in the exact style and structured format required by their internal clinical documentation standards. They have thousands of examples of correctly written past documents. Prompt engineering with examples has produced inconsistent formatting and occasional terminology errors. Which approach is most likely to deliver the required consistent, domain-specific output?

Accepted Answer

Fine-tune a pretrained model using the company's large set of correctly formatted example documents

Answer

Implement retrieval-augmented generation (RAG) to pull terminology from an external glossary at query time

Answer

Increase the maximum token limit on the pretrained model to allow longer responses

Answer

Add more few-shot examples to every prompt until the formatting stabilizes

Question 16

A retail company wants to deploy a customer-facing chatbot that answers general product questions using everyday conversational language. The AI Transformation Leader must decide between using a pretrained foundation model as-is versus investing in a fine-tuned model. The team has limited budget, no specialized dataset, and needs to launch within two weeks. General language understanding is sufficient for the use case. Which approach delivers the best business value given these constraints?

Accepted Answer

Use the pretrained foundation model as-is, since it already handles general conversational language and avoids the time and cost of fine-tuning.

Answer

Fine-tune the model on a large purchased third-party dataset to maximize response accuracy before launch.

Answer

Fine-tune the model using the small amount of internal chat logs available to guarantee brand-specific tone.

Answer

Delay the project until a specialized, labeled training dataset can be assembled for fine-tuning.

Question 17

A healthcare provider is deploying a generative AI assistant to help clinicians answer questions about internal treatment protocols. Compliance leadership states that every AI-generated answer must be traceable to a specific approved source document that a clinician can review. Which business requirement should the transformation leader define to satisfy this need?

Accepted Answer

The solution must ground responses in the approved protocol documents and return source citations with each answer

Answer

The solution must use a larger pretrained model to increase the accuracy of medical answers

Answer

The solution must apply few-shot prompting to demonstrate the desired answer format

Answer

The solution must fine-tune the model on historical clinician chat transcripts

Question 18

A financial services firm is deploying a generative AI assistant to answer employee questions about the company's internal HR benefits. During testing, the assistant frequently gives confident but incorrect answers drawn from general internet knowledge rather than the firm's actual benefits documentation. As the AI Transformation Leader, which approach should you recommend to ensure responses are anchored to the company's authoritative HR content?

Accepted Answer

Ground the model by connecting it to the firm's HR documents so responses are based on that authoritative data

Answer

Increase the temperature parameter so the model generates more creative and varied answers

Answer

Switch to a larger pretrained foundation model with more general world knowledge

Answer

Add more example questions to the prompt to make the assistant respond faster

Question 19

A retail company's data science team is planning a project to deploy a demand-forecasting model. Their AI Transformation Leader wants the team to follow the standard machine-learning lifecycle. After the team has finished collecting and preparing high-quality historical sales data, which stage should immediately follow before the model can be used to serve business predictions?

Accepted Answer

Train and evaluate the model

Answer

Deploy the model to production

Answer

Monitor the model for data drift

Answer

Define the business problem and success metrics

Question 20

A software company wants to boost developer productivity by introducing an AI tool that suggests code completions, generates unit tests from function signatures, and explains unfamiliar code blocks in plain language. Leadership asks the AI Transformation Leader to recommend the type of solution that best fits all three of these needs. Which recommendation is most appropriate?

Accepted Answer

A generative AI model (large language model) trained on natural language and code, integrated into the development environment

Answer

A traditional supervised classification model that labels each line of code as valid or invalid

Answer

A time-series forecasting model to predict future defect counts in the codebase

Answer

A rule-based static analysis linter that flags syntax violations against a fixed ruleset

More AB-731 practice