A manager asks whether predictive analytics can tell them which of two marketing designs to run next quarter, given that no such campaign has been run before. What is the appropriate advice?
Select an answer to reveal the explanation.
Short Explanation
A model can only tell you about things resembling what it has seen. Two designs nobody has ever run are outside that, so the honest answer is to generate the evidence by running a test while exploring whatever related history you do have.
Full Explanation
The correct advice is that without comparable history there is nothing to learn from, so you should run an experiment to generate the data while exploring what related history exists in the meantime. Predictive models generalise from historical examples, so a decision about options with no comparable precedent falls outside what one can support. The productive response is to design an experiment that produces the missing evidence, while exploratory analysis of adjacent history informs how the test is set up.
The advice to fit a model on all available marketing data and use its output as the decision is incorrect because extrapolating a model beyond the range of its training data yields output that looks authoritative and is unfounded. Some history is not better than none when it does not resemble the question.
The advice that predictive analytics cannot inform marketing decisions and that the choice should be made on judgement alone is incorrect because it overcorrects. Analytics informs marketing decisions routinely; the limitation here is this specific absence of precedent.
The advice to federate the marketing platform so the model can access current campaign data and predict the outcome is incorrect because federation supplies fresher access to the same absent history. A connector does not create observations that were never made.
For the exam, watch for scenarios where the required history does not exist - the intended answer is usually to generate it rather than to model around it.