AI Business Value and Boundaries: 10 practice questions

Question 1 of 10

A model scores each case and an investigator then decides what to do with it. What is AI/ML contributing?

1. AA replacement for the investigator's accountability
2. BA written transcript of the investigator's own notes
3. CAssistance for a human decision, supplied as evidence
4. DAn exact result already prescribed by the applicable rules

Question 2 of 10

A team can label a few hundred emails by hand, but millions now arrive every day. Which documented reason to use ML does that match?

1. AThe correct label is already fixed by a published rule
2. BThe data is spread across too many separate systems
3. CYou cannot scale the manual judgement to that volume
4. DThe result has to be identical on every single run

Question 3 of 10

A payments platform sends any transaction that a model flags as suspicious straight to a review queue. What is AI/ML contributing?

1. AA ranked list of products tailored to each payer
2. BAutomation of a recognition step inside the workflow
3. CA deterministic calculation from a published rate table
4. DManual screening of every transaction by an analyst

Question 4 of 10

How should the list of ways AI/ML provides value be read when sizing up an application?

1. AAs a ranking that puts automation above the other entries
2. BAs a rule that bars one workload from showing two entries
3. CAs examples of where value appears, not as exclusive boxes
4. DAs a closed set in which only one entry can ever apply

Question 5 of 10

Which workload characteristic points toward deterministic application logic rather than ML?

1. AInput data changes and model accuracy needs monitoring
2. BHistorical data contains patterns useful for prediction
3. CStable steps completely specify the required result
4. DA forecast helps a planner choose an inventory level

Question 6 of 10

What belongs in a cost-benefit assessment for a proposed ML workload?

1. ATraining-record count compared with the number of deployment Regions
2. BPrediction volume compared with the number of deterministic policy rules
3. CExpected value compared with build, operation, monitoring, and improvement costs
4. DModel feature count compared with the number of available class labels

Question 7 of 10

Why can a technically feasible prediction project still be inappropriate?

1. AIts expected benefit can be lower than its lifecycle cost and risk
2. BA feasible model removes the need for data quality and monitoring
3. CPrediction feasibility means deterministic rules are unavailable in every case
4. DTechnical feasibility proves that its predictions are prescribed outcomes

Question 8 of 10

A payroll system must compute a legally prescribed deduction from a complete rate table. Where, if anywhere, does ML still belong around this system?

1. APredicting each employee's deduction from historical payslips
2. BFlagging anomalous timesheets that feed the calculation
3. CNowhere, because payroll work rules ML out completely
4. DRounding the deduction when the rate table is ambiguous

Question 9 of 10

A proposed model would support a decision that arises twice a year and saves little each time, and it would need labeled data, monitoring, and periodic retraining. What does the cost-benefit step call for next?

1. AApprove it, because producing the prediction is technically feasible
2. BReject it, because rules already prescribe the correct decision
3. CCompare those costs with the value gained over a simpler baseline
4. DDrop the baseline comparison, since no rules-based version exists

Question 10 of 10

A fraud platform scores millions of payments, opens review tickets automatically for the riskiest, and sends borderline scores to investigators. Which reading of it holds up?

1. AHuman review of borderline cases cancels the automated step
2. BIt shows all three: scale, assistance and an automated step
3. CIts score is a prescribed outcome because tickets open on it
4. DApplying it to millions of payments removes the assistance