AIF-C01 - Fundamentals of AI and ML (20% of the exam) - Section 1.2
Identify practical use cases for AI.
Recognise where AI adds value (forecasting, anomaly detection, personalisation, content generation) and where simpler rules or analytics are the better fit. Match a business problem to an appropriate AI capability.
Use case selectionWhen not to use ML
Practice question for this objective
lock_openFree sampleFundamentals of AI and MLeasy
An online grocery retailer is reviewing a list of candidate projects and wants to commit only to those where machine learning would add genuine value over hand-written rules or basic reporting. Which TWO of the following tasks are well suited to a machine learning approach? Select TWO.
ACalculating each order's delivery fee from a published table that maps postcode bands to fixed charges set by management.
BSuggesting the next products an individual shopper is most likely to buy, drawn from years of varied browsing and purchase history.check_circle Correct
CRounding every basket total to two decimal places before the payment gateway processes the charge to the customer.
DSending an automatic email when a delivery is marked complete, using a fixed template the marketing team approved.
EForecasting demand for perishable stock next week from several years of sales, seasonality, and weather signals.check_circle Correct
Recognise that machine learning fits tasks with complex, data-rich patterns such as personalisation and forecasting, not tasks solved by exact rules. Machine learning earns its place when outcomes depend on intricate patterns learned from large, varied historical data that no fixed rule can express. Recommending products per shopper and forecasting perishable demand both rely on such patterns, while lookup tables, rounding, and template emails have single correct outputs that deterministic logic handles exactly.
Why A is wrong: A fixed lookup table gives one exact charge per band, so a deterministic rule is correct and cheaper; ML would add risk without benefit here.
Why B is correct: Personalisation learns each shopper's patterns from rich behavioural history, a relationship too complex and shifting for hand-written rules to capture.
Why C is wrong: Rounding is a precise arithmetic operation with one defined answer, so a simple formula is exact and ML offers nothing useful.
Why D is wrong: A status change triggering a fixed template is straightforward event-driven logic; there is no pattern to learn, so ML is unnecessary.
Why E is correct: Demand forecasting models nonlinear patterns across many noisy historical signals, which is exactly where ML outperforms fixed thresholds or averages.
Answers that look right on this material and are not. Each one is a distractor from a different question in the AIF-C01 bank for this domain.
Unsupervised anomaly detection, because fraud is rare and therefore by definition an anomaly.
errorWhy it is wrong: Unsupervised anomaly detection is used when labels are absent. The firm has labelled fraud cases from investigations, so supervised approaches can be applied and will typically outperform unsupervised methods when labels exist.
Model deployment, where the team specifies serving requirements and latency targets for the production system.
errorWhy it is wrong: Tempting because deployment planning does involve defining requirements, but deployment occurs after a model is trained and evaluated. The described activities - clarifying the business objective and questioning whether ML is appropriate - are characteristic of the earliest lifecycle phase, not a late one.
Generative AI, because the model must generate new demand values that did not exist in historical data.
errorWhy it is wrong: Generative AI creates new data samples such as images or text. Demand forecasting predicts a numeric outcome from structured inputs; it is a regression problem, not a generative task.